Update ggml-cuda/mmvq.cu

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
cuda : fix bounds check for src0 rows in MMVQ kernel
2025-06-24 09:10:57 +00:00 · 2024-06-11 17:37:32 +03:00 · 2024-06-11 11:30:12 +03:00
989 changed files with 144569 additions and 360487 deletions
--- a/.devops/cublas.Dockerfile
+++ b/.devops/cublas.Dockerfile
@ -12,7 +12,7 @@ FROM ${BASE_CUDA_DEV_CONTAINER} as build
 ARG CUDA_DOCKER_ARCH=all

 RUN apt-get update && \
-    apt-get install -y build-essential git cmake libsdl2-dev wget git
+    apt-get install -y build-essential git cmake

 WORKDIR /app

@ -21,8 +21,8 @@ COPY . .
 # Set nvcc architecture
 ENV CUDA_DOCKER_ARCH=${CUDA_DOCKER_ARCH}
 # Enable cuBLAS
-ENV GGML_CUDA=1
+ENV WHISPER_CUBLAS=1

-RUN make base.en
+RUN make

 ENTRYPOINT ["/app/main"]
--- a/.devops/main-cuda.Dockerfile
+++ b/.devops/main-cuda.Dockerfile
@ -13,9 +13,11 @@ WORKDIR /app
 ARG CUDA_DOCKER_ARCH=all
 # Set nvcc architecture
 ENV CUDA_DOCKER_ARCH=${CUDA_DOCKER_ARCH}
+# Enable cuBLAS
+ENV WHISPER_CUBLAS=1

 RUN apt-get update && \
-    apt-get install -y build-essential libsdl2-dev wget cmake git \
+    apt-get install -y build-essential \
    && rm -rf /var/lib/apt/lists/* /var/cache/apt/archives/*

 # Ref: https://stackoverflow.com/a/53464012
@ -23,8 +25,7 @@ ENV CUDA_MAIN_VERSION=12.3
 ENV LD_LIBRARY_PATH /usr/local/cuda-${CUDA_MAIN_VERSION}/compat:$LD_LIBRARY_PATH

 COPY .. .
-# Enable cuBLAS
-RUN make base.en CMAKE_ARGS="-DGGML_CUDA=1"
+RUN make

 FROM ${BASE_CUDA_RUN_CONTAINER} AS runtime
 ENV CUDA_MAIN_VERSION=12.3
@ -32,9 +33,8 @@ ENV LD_LIBRARY_PATH /usr/local/cuda-${CUDA_MAIN_VERSION}/compat:$LD_LIBRARY_PATH
 WORKDIR /app

 RUN apt-get update && \
-  apt-get install -y curl ffmpeg wget cmake git \
+  apt-get install -y curl ffmpeg \
  && rm -rf /var/lib/apt/lists/* /var/cache/apt/archives/*

 COPY --from=build /app /app
-ENV PATH=/app/build/bin:$PATH
 ENTRYPOINT [ "bash", "-c" ]
--- a/.devops/main-intel.Dockerfile
+++ b/.devops/main-intel.Dockerfile
@ -1,28 +0,0 @@
-ARG ONEAPI_VERSION=2025.1.1-0-devel-ubuntu24.04
-
-FROM intel/oneapi-basekit:$ONEAPI_VERSION AS build
-WORKDIR /app
-
-RUN apt-get update && \
-    apt-get install -y build-essential libsdl2-dev wget cmake git \
-    && rm -rf /var/lib/apt/lists/* /var/cache/apt/archives/*
-
-COPY .. .
-# Enable SYCL
-ARG GGML_SYCL_F16=OFF
-RUN if [ "${GGML_SYCL_F16}" = "ON" ]; then \
-        echo "GGML_SYCL_F16 is set" \
-        && export OPT_SYCL_F16="-DGGML_SYCL_F16=ON"; \
-    fi && \
-    make base.en CMAKE_ARGS="-DGGML_SYCL=1 -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx ${OPT_SYCL_F16}"
-
-FROM intel/oneapi-basekit:$ONEAPI_VERSION AS runtime
-WORKDIR /app
-
-RUN apt-get update && \
-  apt-get install -y curl ffmpeg libsdl2-dev wget cmake git \
-  && rm -rf /var/lib/apt/lists/* /var/cache/apt/archives/*
-
-COPY --from=build /app /app
-ENV PATH=/app/build/bin:$PATH
-ENTRYPOINT [ "bash", "-c" ]
--- a/.devops/main-musa.Dockerfile
+++ b/.devops/main-musa.Dockerfile
@ -1,39 +0,0 @@
-ARG UBUNTU_VERSION=22.04
-# This needs to generally match the container host's environment.
-ARG MUSA_VERSION=rc4.0.1
-# Target the MUSA build image
-ARG BASE_MUSA_DEV_CONTAINER=mthreads/musa:${MUSA_VERSION}-mudnn-devel-ubuntu${UBUNTU_VERSION}
-# Target the MUSA runtime image
-ARG BASE_MUSA_RUN_CONTAINER=mthreads/musa:${MUSA_VERSION}-mudnn-runtime-ubuntu${UBUNTU_VERSION}
-
-FROM ${BASE_MUSA_DEV_CONTAINER} AS build
-WORKDIR /app
-
-RUN apt-get update && \
-    apt-get install -y build-essential libsdl2-dev wget cmake git && \
-    apt-get clean && \
-    rm -rf /var/lib/apt/lists/* /var/cache/apt/archives/* /tmp/* /var/tmp/*
-
-COPY .. .
-# Enable muBLAS
-RUN make base.en CMAKE_ARGS="-DGGML_MUSA=1"
-
-RUN find /app/build -name "*.o" -delete && \
-    find /app/build -name "*.a" -delete && \
-    rm -rf /app/build/CMakeFiles && \
-    rm -rf /app/build/cmake_install.cmake && \
-    rm -rf /app/build/_deps
-
-FROM ${BASE_MUSA_RUN_CONTAINER} AS runtime
-WORKDIR /app
-
-RUN apt-get update && \
-    apt-get install -y curl ffmpeg wget cmake git && \
-    apt-get clean && \
-    rm -rf /var/lib/apt/lists/* /var/cache/apt/archives/* /tmp/* /var/tmp/*
-
-COPY --from=build /app /app
-RUN du -sh /app/*
-RUN find /app -type f -size +100M
-ENV PATH=/app/build/bin:$PATH
-ENTRYPOINT [ "bash", "-c" ]
--- a/.devops/main.Dockerfile
+++ b/.devops/main.Dockerfile
@ -2,19 +2,18 @@ FROM ubuntu:22.04 AS build
 WORKDIR /app

 RUN apt-get update && \
-  apt-get install -y build-essential wget cmake git \
+  apt-get install -y build-essential \
  && rm -rf /var/lib/apt/lists/* /var/cache/apt/archives/*

 COPY .. .
-RUN make base.en
+RUN make

 FROM ubuntu:22.04 AS runtime
 WORKDIR /app

 RUN apt-get update && \
-  apt-get install -y curl ffmpeg libsdl2-dev wget cmake git \
+  apt-get install -y curl ffmpeg \
  && rm -rf /var/lib/apt/lists/* /var/cache/apt/archives/*

 COPY --from=build /app /app
-ENV PATH=/app/build/bin:$PATH
 ENTRYPOINT [ "bash", "-c" ]
--- a/.dockerignore
+++ b/.dockerignore
@ -1,3 +0,0 @@
-build*/
-.github/
-.devops/
--- a/.github/workflows/bindings-go.yml
+++ b/.github/workflows/bindings-go.yml
@ -10,13 +10,13 @@ on:
      - whisper.h

 jobs:
-  ubuntu-22:
-    runs-on: ubuntu-22.04
+  ubuntu-latest:
+    runs-on: ubuntu-latest
    steps:
-      - uses: actions/setup-go@v5
+      - uses: actions/setup-go@v3
        with:
-          go-version: '^1.23'
-      - uses: actions/checkout@v4
+          go-version: '^1.19'
+      - uses: actions/checkout@v1
      - run: |
          cd bindings/go
          make test
--- a/.github/workflows/bindings-ruby.yml
+++ b/.github/workflows/bindings-ruby.yml
@ -1,21 +1,22 @@
 name: Bindings Tests (Ruby)
-
 on:
  push:
-    branches:
-      - master
+    paths:
+      - bindings/ruby/**
+      - whisper.h
  pull_request:
-    types: [opened, synchronize, reopened]
+    paths:
+      - bindings/ruby/**
+      - whisper.h

 jobs:
-  ubuntu-22:
-    runs-on: ubuntu-22.04
-    defaults:
-      run:
-        working-directory: bindings/ruby
+  ubuntu-latest:
+    runs-on: ubuntu-latest
    steps:
      - uses: ruby/setup-ruby@v1
        with:
-          ruby-version: '3.2'
-      - uses: actions/checkout@v4
-      - run: rake test
+          ruby-version: '3.0'
+      - uses: actions/checkout@v1
+      - run: |
+          cd bindings/ruby/ext
+          ruby extconf.rb && make
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
--- a/.github/workflows/docker.yml
+++ b/.github/workflows/docker.yml
@ -11,18 +11,14 @@ jobs:
    name: Push Docker image to Docker Hub
    if: github.event.pull_request.draft == false

-    runs-on: ubuntu-22.04
+    runs-on: ubuntu-latest
    env:
      COMMIT_SHA: ${{ github.sha }}
    strategy:
      matrix:
        config:
-          - { tag: "main", dockerfile: ".devops/main.Dockerfile", platform: "linux/amd64" }
-          - { tag: "main-musa", dockerfile: ".devops/main-musa.Dockerfile", platform: "linux/amd64" }
-          - { tag: "main-intel", dockerfile: ".devops/main-intel.Dockerfile", platform: "linux/amd64" }
-          #TODO: the cuda image keeps failing - disable for now
-          #      https://github.com/ggerganov/whisper.cpp/actions/runs/11019444428/job/30602020339
-          #- { tag: "main-cuda", dockerfile: ".devops/main-cuda.Dockerfile", platform: "linux/amd64" }
+          - { tag: "main", dockerfile: ".devops/main.Dockerfile", platform: "linux/amd64,linux/arm64" }
+          - { tag: "main-cuda", dockerfile: ".devops/main-cuda.Dockerfile", platform: "linux/amd64" }

    steps:
      - name: Check out the repo
@ -30,8 +26,6 @@ jobs:

      - name: Set up QEMU
        uses: docker/setup-qemu-action@v3
-        with:
-          image: tonistiigi/binfmt:qemu-v7.0.0-28

      - name: Set up Docker Buildx
        uses: docker/setup-buildx-action@v3
@ -49,7 +43,7 @@ jobs:
        with:
          context: .
          push: true
-          platforms: ${{ matrix.config.platform }}
+          platforms: ${{ matrix.config.platforms }}
          tags: "ghcr.io/${{ github.repository }}:${{ matrix.config.tag }}-${{ env.COMMIT_SHA }}"
          file: ${{ matrix.config.dockerfile }}

@ -58,6 +52,6 @@ jobs:
        with:
          context: .
          push: ${{ github.event_name == 'push' }}
-          platforms: ${{ matrix.config.platform }}
+          platforms: ${{ matrix.config.platforms }}
          tags: "ghcr.io/${{ github.repository }}:${{ matrix.config.tag }}"
          file: ${{ matrix.config.dockerfile }}
--- a/.github/workflows/examples-wasm.yml
+++ b/.github/workflows/examples-wasm.yml
@ -1,91 +0,0 @@
-name: Examples WASM
-on:
-  push:
-    branches: ["master"]
-
-  workflow_dispatch:
-
-permissions:
-  contents: read
-  pages: write
-  id-token: write
-
-concurrency:
-  group: "pages"
-  cancel-in-progress: false
-
-jobs:
-  deploy-wasm-github-pages:
-    environment:
-      name: github-pages
-      url: ${{ steps.deployment.outputs.page_url }}
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-
-      - name: Setup Pages
-        uses: actions/configure-pages@v4
-
-      - name: Setup emsdk
-        uses: mymindstorm/setup-emsdk@v14
-
-      - name: Build WASM Examples
-        # Enable for real build later in whisper.cpp
-        run: |
-          mkdir -p build-em && cd build-em
-          emcmake cmake .. -DCMAKE_BUILD_TYPE=Release
-          make -j
-
-      - name: Create staging directory
-        run: mkdir -p staging
-
-      - name: Create .nojekyll file in staging directory
-        run: touch staging/.nojekyll
-
-      - name: Copy application files
-        run: |
-          build_dir=build-em/bin
-
-          ls ${build_dir}
-
-          # command.wasm
-          target_dir=staging/command.wasm
-          mkdir -p ${target_dir}
-          cp ${build_dir}/command.wasm/{index.html,command.js,helpers.js} ${target_dir}
-          cp ${build_dir}/libcommand.js ${target_dir}
-
-          # bench.wasm
-          target_dir=staging/bench.wasm
-          mkdir -p ${target_dir}
-          cp ${build_dir}/bench.wasm/{index.html,bench.js,helpers.js} ${target_dir}
-          cp ${build_dir}/libbench.js ${target_dir}
-
-          # stream.wasm
-          target_dir=staging/stream.wasm
-          mkdir -p ${target_dir}
-          cp ${build_dir}/stream.wasm/{index.html,stream.js,helpers.js} ${target_dir}
-          cp ${build_dir}/libstream.js ${target_dir}
-
-          # whisper.wasm (this will be the main example page)
-          target_dir=staging
-          mkdir -p ${target_dir}
-          cp ${build_dir}/whisper.wasm/{index.html,main.js,helpers.js} ${target_dir}
-          cp ${build_dir}/libmain.js ${target_dir}
-
-          # Copy Cross-Origin Isolation service worker
-          cp -v examples/coi-serviceworker.js staging/
-
-      - name: List files in staging directory (for debugging)
-        run: |
-          echo "Files in staging directory:"
-          find staging -type f | sort
-
-      - name: Upload artifact
-        uses: actions/upload-pages-artifact@v3
-        with:
-          path: ./staging
-
-      - name: Deploy to GitHub Pages
-        id: deployment
-        uses: actions/deploy-pages@v4
--- a/.github/workflows/examples.yml
+++ b/.github/workflows/examples.yml
@ -10,8 +10,8 @@ on:
      - whisper.h

 jobs:
-  addon_node-ubuntu-22:
-    runs-on: ubuntu-22.04
+  addon_node-ubuntu-latest:
+    runs-on: ubuntu-latest
    strategy:
      matrix:
        node-version: [ 16.x, 18.x ]
@ -22,7 +22,7 @@ jobs:
      - name: Dependencies
        run: |
          sudo apt-get update
-          sudo apt-get install build-essential git
+          sudo apt-get install build-essential
          sudo apt-get install cmake
          sudo apt-get install libsdl2-dev

--- a/.gitignore
+++ b/.gitignore
@ -1,29 +1,31 @@
 *.o
 *.a
-*.d
 .cache/
 .coreml/
 .test/
-.venv/
 .vs/
 .vscode/
 .DS_Store
 .vimspector.json
 /CMakeSettings.json
-/talk-llama.dSYM/

 build/
-build-*/
-build_*/
+build-coreml/
+build-em/
+build-debug/
+build-release/
+build-rwdi/
+build-static/
+build-cublas/
+build-no-accel/
+build-sanitize-addr/
+build-sanitize-thread/

 # SPM
 .build/
 .swiftpm
 *.metallib

-ggml-metal-embed.metal
-ggml-metal-embed.metal.tmp
-
 /main
 /stream
 /command
@ -50,8 +52,6 @@ extra/bench-gg.txt
 models/*.mlmodel
 models/*.mlmodelc
 models/*.mlpackage
-models/*-encoder-openvino.xml
-models/*-encoder-openvino-cache/
 bindings/java/.gradle/
 bindings/java/.idea/
 .idea/
@ -61,5 +61,3 @@ cmake-build-debug/
 .cxx/
 .gradle/
 local.properties
-.log
-.exe
--- a/.gitmodules
+++ b/.gitmodules
@ -0,0 +1,3 @@
+[submodule "bindings/ios"]
+	path = bindings/ios
+	url = https://github.com/ggerganov/whisper.spm
--- a/211
+++ b/211
@ -1,51 +1,34 @@
-# date: Tue Feb  4 13:03:35 EET 2025
+# date: Tue Apr  9 20:27:03 EEST 2024
 # this file is auto-generated by scripts/gen-authors.sh

 0/0 <zero@imaskeleton.me>
 0cc4m <picard12@live.de>
 0xsourcecode <134374803+0xsourcecode@users.noreply.github.com>
-65a <10104049+65a@users.noreply.github.com>
-AIWintermuteAI <32562299+AIWintermuteAI@users.noreply.github.com>
 AT <manyoso@users.noreply.github.com>
 Aarni Koskela <akx@iki.fi>
 Aaron Pham <29749331+aarnphm@users.noreply.github.com>
 Aaron Taylor <aaron@exphat.com>
 Abhilash Majumder <30946547+abhilash1910@users.noreply.github.com>
 Abitofevrything <54505189+abitofevrything@users.noreply.github.com>
-Adam Jones <domdomegg+git@gmail.com>
-Adrien Gallouët <adrien@gallouet.fr>
-Adrien Gallouët <angt@huggingface.co>
 AfryMask <AfryMask@163.com>
 Ahmad Bilal <ahmad.bilal@empglabs.com>
-Ahmad Tameem <113388789+Tameem-10xE@users.noreply.github.com>
 AidanBeltonS <87009434+AidanBeltonS@users.noreply.github.com>
-AidanBeltonS <aidan.belton@codeplay.com>
-Akarshan Biswas <akarshan.biswas@gmail.com>
-Akarshan Biswas <akarshanbiswas@fedoraproject.org>
 Akash Mahajan <akash7190@gmail.com>
 Akash Mahajan <akashmjn@stanford.edu>
 Al Hoang <3811822-hoanga@users.noreply.gitlab.com>
 Alan <unknown>
-Albert Jin <albert.jin@gmail.com>
-Alberto Cabrera Pérez <alberto.cabrera@codeplay.com>
-Alberto Cabrera Pérez <alberto.cabrera@intel.com>
 Aleksander Andrzejewski <18704749+aleksanderandrzejewski@users.noreply.github.com>
 Alex Azarov <alex@azarov.by>
 Alex Bacart <13940752+alex-bacart@users.noreply.github.com>
 Alex Evgrashin <aevgrashin@yandex.ru>
-Alex O'Connell <35843486+acon96@users.noreply.github.com>
 Alexandr Graschenkov <alexandr.graschenkov91@gmail.com>
 Alexandru Mariuti <alex@mariuti.com>
 Alexey Kharlamov <alexey@kharlamov.biz>
 Alfredo Montesinos <alfredo.montesinos@g.austincc.edu>
 Ali Alameh <ali.alameh@isae.edu.lb>
-Alter <0x7c48@gmail.com>
 Ananta Bastola <anantarajbastola@gmail.com>
-Andreas Kieslinger <47689530+aendk@users.noreply.github.com>
-Andreas Lubbe <git@lubbe.org>
 Andreu Huguet <andreuhuguet@gmail.com>
 Andrew Huynh <a5thuynh@gmail.com>
-Andrew Minh Nguyen <40281306+amqdn@users.noreply.github.com>
 Andrew S <andrews54757@gmail.com>
 Andy Maloney <asmaloney@gmail.com>
 Anton Kostin <masguit42@users.noreply.github.com>
@ -57,11 +40,8 @@ AustinMroz <austinmroz@utexas.edu>
 Avik Sengupta <avik@sengupta.net>
 Bader-eddine Ouaich <49657842+baderouaich@users.noreply.github.com>
 Baffin Lee <baffinlee@gmail.com>
-Ben Ashbaugh <ben.ashbaugh@intel.com>
 Ben Nortier <bjnortier@gmail.com>
 Benjamin Heiniger <benjamin.heiniger@bluewin.ch>
-Bernhard M. Wiedemann <githubbmwprimary@lsmod.de>
-Binozo <70137898+Binozo@users.noreply.github.com>
 Bo-Yi Wu <appleboy.tw@gmail.com>
 Boris Bliznioukov <blib@mail.com>
 Borislav Stanimirov <b.stanimirov@abv.bg>
@ -69,86 +49,47 @@ Brad Murray <59848399+bradmurray-dt@users.noreply.github.com>
 Brian Murray <brian@bmurray.ca>
 CRD716 <crd716@gmail.com>
 Canis Lupus <Canis-UK@users.noreply.github.com>
-Carlos Zoido <mrgalleta@gmail.com>
 Carolinabanana <140120812+Carolinabanana@users.noreply.github.com>
-CarterLi999 <664681047@qq.com>
 ChangSeok Oh <shivamidow@users.noreply.github.com>
-Changyeon Kim <cyzero.kim@samsung.com>
 Chaoqun <27287694+OpenWaygate@users.noreply.github.com>
-Charles Xu <63788048+chaxu01@users.noreply.github.com>
-Charles Xu <charles.xu@arm.com>
-Chen Xi <xi2.chen@intel.com>
-Chen Xi <xixichen08@foxmail.com>
-Chenguang Li <87689256+noemotiovon@users.noreply.github.com>
 Chia-Hsiang Cheng <88014292+garychia@users.noreply.github.com>
 Chidi Williams <williamschidi1@gmail.com>
-Chris Elrod <elrodc@gmail.com>
 Christian <12550267+iceychris@users.noreply.github.com>
-Christian Kastner <ckk@kvr.at>
 Clifford Heath <clifford.heath@gmail.com>
-Clint Herron <hanclinto@gmail.com>
 Colin <github@whoisc.cc>
-Conrad Kramer <conrad@conradkramer.com>
-Corey Earwood <iamcgn+github@gmail.com>
-CrispStrobe <154636388+CrispStrobe@users.noreply.github.com>
-DAN™ <dranger003@gmail.com>
 DGdev91 <DGdev91@users.noreply.github.com>
 Damian Czaja <trojan295@protonmail.com>
-Dan Johansson <164997844+eddnjjn@users.noreply.github.com>
-Dan Johansson <dan.johansson@arm.com>
 Daniel Bevenius <daniel.bevenius@gmail.com>
-Daniel Valdivia <18384552+dvaldivia@users.noreply.github.com>
-Daniel Ziegenberg <daniel@ziegenberg.at>
-Daniele <57776841+daniandtheweb@users.noreply.github.com>
-Dave <dave-fl@users.noreply.github.com>
-Dave Airlie <airlied@gmail.com>
-Dave Airlie <airlied@redhat.com>
-Daven Sanassy <daven@vochlea.co.uk>
 David <dnhkng@gmail.com>
 David Thorpe <djt@mutablelogic.com>
-DavidKorczynski <david@adalogics.com>
 Davidson Francis <davidsondfgl@gmail.com>
 Dener Stassun <denerstassun@gmail.com>
-Dibakar Gope <dibakar.gope@arm.com>
 Didzis Gosko <didzis@users.noreply.github.com>
-Diego Devesa <slarengh@gmail.com>
 Digipom <admin@digipom.com>
 Dimo <dimo@ieee.org>
-Djip007 <3705339+Djip007@users.noreply.github.com>
-Djip007 <djip.perois@free.fr>
 Dody Suria Wijaya <dodysw@gmail.com>
-Dou Xinpeng <15529241576@163.com>
-Dou Xinpeng <81913537+Dou-Git@users.noreply.github.com>
 Dr. Tom Murphy VII Ph.D <499244+tom7@users.noreply.github.com>
 Duncan McConnell <ddmcconnell4@gmail.com>
 Egor Egorov <me@egorfine.com>
 Elkana Bardugo <ttv200@gmail.com>
 Emmanuel Schmidbauer <eschmidbauer@gmail.com>
 Engininja2 <139037756+Engininja2@users.noreply.github.com>
-Eric Curtin <ericcurtin17@gmail.com>
 Eric Swanson <eswanson@alloscomp.com>
 Eric Tendian <erictendian@gmail.com>
-Eric Zhang <34133756+EZForever@users.noreply.github.com>
 Erik Scholz <Green-Sky@users.noreply.github.com>
 Evan Jones <evan.q.jones@gmail.com>
 Evan Martin <evan.martin@gmail.com>
 Eve <139727413+netrunnereve@users.noreply.github.com>
 Evgeny Kuznetsov <evgeny@kuznetsov.md>
 F1L1P <78918286+F1L1Pv2@users.noreply.github.com>
-Faisal Zaghloul <quic_fzaghlou@quicinc.com>
 Fangjun Kuang <csukuangfj@gmail.com>
 Felix <stenbackfelix@gmail.com>
 Finn Voorhees <finnvoorhees@gmail.com>
-FirstTimeEZ <179362031+FirstTimeEZ@users.noreply.github.com>
 FlippFuzz <41221030+FlippFuzz@users.noreply.github.com>
-Frankie Robertson <frankier@users.noreply.github.com>
 Gang Chen <goncha@gmail.com>
 Gavin Cai <gavin1818@hotmail.com>
 George Hindle <george@georgehindle.com>
 Georgi Gerganov <ggerganov@gmail.com>
-Gilad S <7817232+giladgd@users.noreply.github.com>
-Gilad S <giladgd@users.noreply.github.com>
-Gilad S. <7817232+giladgd@users.noreply.github.com>
 GitAritron <103900385+GitAritron@users.noreply.github.com>
 GiviMAD <GiviMAD@users.noreply.github.com>
 Gleicon Moraes <gleicon@gmail.com>
@ -157,66 +98,41 @@ Guillaume Wenzek <gwenzek@users.noreply.github.com>
 HY. Kelvin Lee <34256578+hykelvinlee42@users.noreply.github.com>
 Halalaluyafail3 <55773281+Halalaluyafail3@users.noreply.github.com>
 Hang <bebound@gmail.com>
-Haus1 <haus.xda@gmail.com>
 Herman Semenov <GermanAizek@yandex.ru>
-HimariO <dsfhe49854@gmail.com>
-Hong Bo PENG <penghb@cn.ibm.com>
 Hrishikesh Barman <geekodour@users.noreply.github.com>
-Hugo <hugo@whynothugo.nl>
 Ian Bicking <ian@ianbicking.org>
 Ian Bull <irbull@eclipsesource.com>
-Ihar Hrachyshka <ihrachys@redhat.com>
 Ikko Ashimine <eltociear@gmail.com>
-Ikko Eltociear Ashimine <eltociear@gmail.com>
 InconsolableCellist <23345188+InconsolableCellist@users.noreply.github.com>
 Ismatulla Mansurov <47342870+sapoepsilon@users.noreply.github.com>
-Ivan <nekotekina@gmail.com>
-Ivan Filipov <159561759+vanaka11@users.noreply.github.com>
 Ivan Gorin <ivangorin21@gmail.com>
-Ivo von Putzer Reibegg <ivo.putzer@gmail.com>
 JJ <103335846+computerscienceiscool@users.noreply.github.com>
 Jack Mousseau <jmousseau@users.noreply.github.com>
 JacobLinCool <jacoblincool@gmail.com>
 Jakub Ráček <blizzcz@gmail.com>
 Jared Van Bortel <jared@nomic.ai>
 Jay Binks <jaybinks@gmail.com>
-Jayant <jayantyadav202@gmail.com>
-Jeff Bolz <jbolz@nvidia.com>
-Jeroen Mostert <jeroen.mostert@cm.com>
 Jhen-Jie Hong <developer@jhen.me>
 Jhen-Jie Hong <iainst0409@gmail.com>
 JidongZhang-THU <1119708529@qq.com>
 Jo Liss <joliss42@gmail.com>
-Joe Todd <joe.todd@codeplay.com>
 Johan <jr.raffin@gmail.com>
 Johannes Gäßler <johannesg@5d6.de>
 John Balis <phobossystems@gmail.com>
-JohnnyB <jboero@users.noreply.github.com>
 Jonathan Soo <jcsoo@agora.com>
 Jonno <1160532+razodactyl@users.noreply.github.com>
 Joonas Pihlajamaa <joonas.pihlajamaa@iki.fi>
 Jose <34888496+Jerry-Master@users.noreply.github.com>
 Josh Bleecher Snyder <josharian@gmail.com>
-Josscii <jossciiweiyi@gmail.com>
 Judd <foldl@users.noreply.github.com>
 Jumper775 <78500318+jumpers775@users.noreply.github.com>
-Jun Hee Yoo <contact.jhyoo@gmail.com>
-Junil Kim <logyourself@gmail.com>
-Justina Cho <justcho5@gmail.com>
 Justine Tunney <jtunney@gmail.com>
-Justine Tunney <jtunney@mozilla.com>
-KITAITI Makoto <KitaitiMakoto@gmail.com>
 KP Kaiser <kirk@zothcorp.com>
 Kamilake <exjang0@gmail.com>
-Karol Kontny <82021046+kkontny@users.noreply.github.com>
-Karthick <j.karthic2004@gmail.com>
 Kartik Saranathan <278928+Kartiku@users.noreply.github.com>
 Kasumi <90275229+kasumi-1@users.noreply.github.com>
 Kawrakow <48489457+ikawrakow@users.noreply.github.com>
-Kendrick Taylor <kendrick@circuitsix.com>
 Kevin Brothaler <admin@digipom.com>
-Kevin Gibbons <bakkot@gmail.com>
-Konosuke Sakai <konosuke@konosuke.work>
 Konstantin Zhuravlyov <konstantin.zhuravlyov@amd.com>
 Kreijstal <rainb@tfwno.gf>
 Kylin <56434533+KyL0N@users.noreply.github.com>
@ -231,110 +147,56 @@ Luis Herrera <herrera-luis@users.noreply.github.com>
 Lukas Rist <glaslos@gmail.com>
 M. A. Ali <73258591+MightyStud@users.noreply.github.com>
 M. Eren Akbiyik <erenakbiyik@gmail.com>
-Ma Mingfei <mingfei.ma@intel.com>
 Maciek <maciek.mab122@gmail.com>
-Mahesh Madhav <67384846+heshpdx@users.noreply.github.com>
 Marcin Mielniczuk <marmistrz.dev@zoho.eu>
-Mark Karpelès <MagicalTux@users.noreply.github.com>
-Mark Zhuang <zhuangqiubin@gmail.com>
-Markus Tavenrath <mtavenrath@users.noreply.github.com>
-Martin Delille <martin@delille.org>
 Martin Warnaar <martinwarnaar@gmail.com>
-Masaya, Kato <62578291+msy-kato@users.noreply.github.com>
 Matheus de Sousa <23645013+keyehzy@users.noreply.github.com>
-Mathieu Baudier <mbaudier@argeo.org>
 Mathijs de Bruin <mathijs@mathijsfietst.nl>
 Matija Pevec <mightymatth@users.noreply.github.com>
-Matt Stephenson <mstephenson6@users.noreply.github.com>
-Max Krasnyansky <max.krasnyansky@gmail.com>
-Max Krasnyansky <quic_maxk@quicinc.com>
 Maximiliano Levi <8160966+maxilevi@users.noreply.github.com>
 Meng, Hengyu <hengyu.meng@intel.com>
-Mengqing Cao <cmq0113@163.com>
 Michael Podvitskiy <podvitskiymichael@gmail.com>
 Michael Rienstra <mrienstra@gmail.com>
 Mikhail Grigorev <sleuthhound@gmail.com>
 Mohammadreza Hendiani <hendiani.mohammadreza@gmail.com>
 Mohit Agarwal <mohit@sdf.org>
-Molly Sophia <mollysophia379@gmail.com>
 Murilo Santana <mvrilo@gmail.com>
-NETZkultur GmbH <mulholland@netzkultur.de>
-Natsu <chino@hotococoa.moe>
 Neil Chudleigh <nchudleigh@users.noreply.github.com>
-Neo Zhang <14088817+arthw@users.noreply.github.com>
 Neo Zhang Jianyu <jianyu.zhang@intel.com>
 Neuman Vong <neuman.vong@gmail.com>
-Nicholai Tukanov <nicholaitukanov@gmail.com>
 Nicholas Albion <nalbion@yahoo.com>
-Nico Bosshard <nico@bosshome.ch>
-Nicolò Scipione <nicolo.scipione@codeplay.com>
 Niels Mayer <Niels.Mayer@gmail.com>
-Nikita Sarychev <42014488+sARY77@users.noreply.github.com>
-Nikolaj Olsson <nikse.dk@gmail.com>
 Okabintaro <103938900+Okabintaro@users.noreply.github.com>
 Oleg Sidorov <me@whitebox.io>
 Oleg Sidorov <oleg@sidorov.nl>
-Olivier Chafik <ochafik@users.noreply.github.com>
 Ondrej Kokes <ondrej.kokes@gmail.com>
 Ouadie EL FAROUKI <ouadie.elfarouki@codeplay.com>
-PAB <pierreantoine.bannier@gmail.com>
 Paul Tsochantaris <ptsochantaris@icloud.com>
-Pedro Probst <pprobst@insiberia.net>
-Peng <hzp1024@qq.com>
-Peter <peter277@users.noreply.github.com>
 Philipp Zabel <philipp.zabel@gmail.com>
 Philippe Normand <phil@base-art.net>
-Philippe Normand <philn@igalia.com>
-Plamen Minev <pacominev@gmail.com>
-Prashant Vithule <119530321+Vithulep@users.noreply.github.com>
 Przemysław Pawełczyk <przemoc@gmail.com>
 Qianhe Chen <54462604+chenqianhe@users.noreply.github.com>
-R0CKSTAR <xiaodong.ye@mthreads.com>
-R0CKSTAR <yeahdongcn@gmail.com>
-Radoslav Gerganov <rgerganov@gmail.com>
 Radosław Gryta <radek.gryta@gmail.com>
-Rahul Vadhyar <107788610+RahulVadhyar@users.noreply.github.com>
-Raiya Araki <83504221+rai62@users.noreply.github.com>
 Reinforce-II <fate@eastal.com>
 Reinis Muiznieks <muiznieks.reinis@gmail.com>
 RelatedTitle <r3latedtitle@gmail.com>
-Rémy Oudompheng <oudomphe@phare.normalesup.org>
 RhinoDevel <RhinoDevel@users.noreply.github.com>
 Rich Jones <miserlou@gmail.com>
-Robert Ormandi <52251610+ormandi@users.noreply.github.com>
 Robin <robin.xw@hotmail.com>
 Roddur Dasgupta <roddurd@gmail.com>
 Roland Rabien <figbug@gmail.com>
-Romain Biessy <romain.biessy@codeplay.com>
-Ronsor <ronsor@ronsor.pw>
 Rotem Dan <rotemdan@gmail.com>
 Ryan Hitchman <hitchmanr@gmail.com>
 Ryan Metcalfe <107415876+RyanMetcalfeInt8@users.noreply.github.com>
 RyanChang <ftes90015@gmail.com>
-SRHMorris <69468379+SRHMorris@users.noreply.github.com>
-SXX <sxx1136965276@gmail.com>
-Sacha Arbonel <sacha.arbonel@hotmail.fr>
-Salman Faroz <stsfaroz@gmail.com>
-Salvatore Mesoraca <s.mesoraca16@gmail.com>
 Sam <49637763+Onlyartist9@users.noreply.github.com>
 Sam Pullara <spullara@gmail.com>
-Samuel Durante <44513615+samueldurantes@users.noreply.github.com>
 Sanchit Gandhi <93869735+sanchit-gandhi@users.noreply.github.com>
-Sandro Hanea <40202887+sandrohanea@users.noreply.github.com>
-Sergio López <slp@redhat.com>
 Sergio López <slp@sinrega.org>
-Shanshan Shen <467638484@qq.com>
-Shijie <821898965@qq.com>
-Shupei Fan <dymarkfan@outlook.com>
 Siddharth Ramakrishnan <srr2141@columbia.edu>
-Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
 Simon Moisselin <simon.moisstoll@gmail.com>
 Sindre Sorhus <sindresorhus@gmail.com>
 Slava Primenko <primenko.s@gmail.com>
-Srihari-mcw <96763064+Srihari-mcw@users.noreply.github.com>
-Stavros Panakakis <53979866+Stavrospanakakis@users.noreply.github.com>
-Stefan Sydow <s.sydow@heinlein-video.de>
-Stefan Sydow <stefan@sydow.email>
 Syahmi Azhar <prsyahmi@gmail.com>
 Syed Jafri <syedjafri97@gmail.com>
 Sơn Phan Trung <phantrungson17@gmail.com>
@ -343,63 +205,37 @@ Takeshi Inoue <inoue.takeshi@gmail.com>
 Tamotsu Takahashi <ttakah+github@gmail.com>
 Taras Glek <taras@thegp.com>
 Tauseef Mohiuddin <35351464+tauseefmohammed2@users.noreply.github.com>
-Thamster <Thamster@users.noreply.github.com>
 Thijs Raymakers <thijs@raymakers.nl>
 Thomas Fitzsimmons <fitzsim@fitzsim.org>
 Tiago Fassoni <tiagofassoni@users.noreply.github.com>
 Tienshiao Ma <tienshiao@tienshiao.org>
-Tim Miller <drasticactions@users.noreply.github.com>
 Timothy Cronin <40186632+4imothy@users.noreply.github.com>
 Tobrun <tobrun.van.nuland@gmail.com>
 Todd <taf2@users.noreply.github.com>
-Toliver <teejae@gmail.com>
 Tong Li <31761981+litongjava@users.noreply.github.com>
-Tony Wasserka <4840017+neobrain@users.noreply.github.com>
 Topping1 <78745143+Topping1@users.noreply.github.com>
 Travis Cline <travis.cline@gmail.com>
 UEXTM.com <84163508+uextm@users.noreply.github.com>
-UsernamesLame <156965854+UsernamesLame@users.noreply.github.com>
 Vadim Peretokin <vperetokin@hey.com>
 Valentin Gosu <1454649+valenting@users.noreply.github.com>
-Vin Misra <vinith@alum.mit.edu>
 Vulcan <93451215+trholding@users.noreply.github.com>
 WhiteOlivierus <36532695+WhiteOlivierus@users.noreply.github.com>
-William Tambellini <william.tambellini@gmail.com>
-William Tambellini <wtambellini@sdl.com>
-Wilson Silva <wilson.dsigns@gmail.com>
 Xiang (Kevin) Li <kevinli020508@gmail.com>
 Xiao-Yong Jin <jinxiaoyong@gmail.com>
 XiaotaoChen <chenxiaotao1234@gmail.com>
-Xingchen Song(宋星辰) <xingchensong1996@163.com>
-Xinpeng Dou <81913537+Dou-Git@users.noreply.github.com>
-Xuan Son Nguyen <thichthat@gmail.com>
 Yajing Tang <phillis@google.com>
 Yang Shen <aplshenyang@gmail.com>
 Yunès <jean.baptiste.yunes@free.fr>
-Yuri Khrustalev <ykhrustalev@users.noreply.github.com>
-Yusuf Redžić <48274562+redzic@users.noreply.github.com>
 ZaBlazzingZephyrus <119159668+blazingzephyr@users.noreply.github.com>
-Zhenwei Jin <109658203+kylo5aby@users.noreply.github.com>
-Zhiyuan Li <lizhiyuan@uniartisan.com>
-Zhiyuan Li <uniartisan2017@gmail.com>
 Zigfrid Zvezdin <ziggerZZ@gmail.com>
 Zollner <24618122+Zolliner@users.noreply.github.com>
-a3sh <38979186+A3shTnT@users.noreply.github.com>
-ag2s20150909 <19373730+ag2s20150909@users.noreply.github.com>
-agray3 <agray3@users.noreply.github.com>
 ai-at-home <149282006+ai-at-home@users.noreply.github.com>
-aldorof <aldorof@users.noreply.github.com>
 alonfaraj <alonfaraj@gmail.com>
-amd-dwang <dong.wang@amd.com>
-amritahs-ibm <amritahs@linux.vnet.ibm.com>
 andypayne <apayne@gmail.com>
 ardfork <134447697+ardfork@users.noreply.github.com>
-arizhih <40765267+arizhih@users.noreply.github.com>
 automaticcat <daogiatuank54@gmail.com>
-bandoti <141645996+bandoti@users.noreply.github.com>
 be-next <jerome.ramette@gmail.com>
 bert hubert <bert@hubertnet.nl>
-billyct <billy_allen@126.com>
 bmwl <brian.marshall@tolko.com>
 bobqianic <129547291+bobqianic@users.noreply.github.com>
 bocytko <bocytko+github@gmail.com>
@ -412,9 +248,7 @@ byte-6174 <88070277+byte-6174@users.noreply.github.com>
 cdosoftei <ciprian.dosoftei@gmail.com>
 clach04 <Chris.Clark@actian.com>
 compilade <113953597+compilade@users.noreply.github.com>
-compilade <git@compilade.net>
 conradg <conradjgodfrey@gmail.com>
-crummyh <elijah@crums.us>
 ddpasa <112642920+ddpasa@users.noreply.github.com>
 denersc <denerstassun@gmail.com>
 dscripka <dscripka@users.noreply.github.com>
@ -422,55 +256,28 @@ duthils <duthils@duthils.net>
 ecneladis <ecneladis@users.noreply.github.com>
 faker <nspyia2002@gmail.com>
 fitzsim <fitzsim@fitzsim.org>
-fj-y-saito <85871716+fj-y-saito@users.noreply.github.com>
 fraxy-v <65565042+fraxy-v@users.noreply.github.com>
 genevera (she/her) <genevera@users.noreply.github.com>
 geniusnut <geniusnut@gmail.com>
-gilbertgong <gilbert.gong@gmail.com>
-gn64 <yukikaze.jp@gmail.com>
-goldwaving <77494627+goldwaving@users.noreply.github.com>
 greeshmay <greeshmay@gmail.com>
-haopeng <657407891@qq.com>
-hipudding <huafengchun@gmail.com>
-hsinhoyeh <yhh92u@gmail.com>
 hydai <z54981220@gmail.com>
 iamthad <thadeus.j.fleming@gmail.com>
-issixx <46835150+issixx@users.noreply.github.com>
 james wolf <contractorwolf@hotmail.com>
-jdomke <28772296+jdomke@users.noreply.github.com>
-jettoblack <jettoblack@gmail.com>
-jiez <373447296@qq.com>
 joecryptotoo <80373433+joecryptotoo@users.noreply.github.com>
 jorismertz <35079666+jorismertz@users.noreply.github.com>
-junchao-loongson <68935141+junchao-loongson@users.noreply.github.com>
 junkfood <69683722+JunkFood02@users.noreply.github.com>
 jwijffels <jwijffels@bnosac.be>
-k.h.lai <adrian.k.h.lai@outlook.com>
 kamranjon <kamranjon@gmail.com>
 katsu560 <katsu560oo-@docomo.ne.jp>
 kennethge <57784063+kenneth-ge@users.noreply.github.com>
 keyehzy <msamuel@aluno.puc-rio.br>
-kunnis <kunnis@users.noreply.github.com>
-l3utterfly <gc.pthzfoldr@gmail.com>
 leejet <leejet714@gmail.com>
-leo-pony <nengjunma@outlook.com>
-lhez <quic_lih@quicinc.com>
 litong <31761981+litongjava@users.noreply.github.com>
-liuwei-git <14815172+liuwei-git@users.noreply.github.com>
 lnyan <lkwq007@gmail.com>
-luoyu-intel <yu.luo@intel.com>
 m.bell <m.bell@techsmith.com>
-mahorozte <41834471+mahorozte@users.noreply.github.com>
-mashizora <30516315+mashizora@users.noreply.github.com>
-matt23654 <matthew.webber@protonmail.com>
-matteo <matteogeniaccio@yahoo.it>
-mgrachten <maarten@grachten.eu>
 mkiol <mkiol@users.noreply.github.com>
-mky_coder <47767389+mkycoder@users.noreply.github.com>
 novag <7754358+novag@users.noreply.github.com>
 pajowu <pajowu@pajowu.de>
-pengxin99 <pengxin.yuan@intel.com>
-petterreinholdtsen <pere-github@hungry.com>
 polarmoon <90010972+polarmoon@users.noreply.github.com>
 rlapray <lapray.romain@gmail.com>
 sandrohanea <40202887+sandrohanea@users.noreply.github.com>
@ -480,31 +287,15 @@ shikokuchuo <53399081+shikokuchuo@users.noreply.github.com>
 slaren <slarengh@gmail.com>
 slashlib <slashlib@users.noreply.github.com>
 snadampal <87143774+snadampal@users.noreply.github.com>
-someone13574 <81528246+someone13574@users.noreply.github.com>
 st-gr <38470677+st-gr@users.noreply.github.com>
-stduhpf <stephduh@live.fr>
-stormofice <58337328+stormofice@users.noreply.github.com>
 texmex76 <40733439+texmex76@users.noreply.github.com>
 thefinaldegree <thefinaldegree@gmail.com>
-thewh1teagle <61390950+thewh1teagle@users.noreply.github.com>
-toboil-features <160222185+toboil-features@users.noreply.github.com>
 trixirt <trix@redhat.com>
 ulatekh <ulatekh@yahoo.com>
 undef <undefdev@gmail.com>
-uvos <devnull@uvos.xyz>
-uvos <philipp@uvos.xyz>
-valVk <valVk@users.noreply.github.com>
 venkr <venkateshrameshkumar+1@gmail.com>
 vicalloy <zbirder@gmail.com>
-wangshuai09 <391746016@qq.com>
-woachk <24752637+woachk@users.noreply.github.com>
-xctan <axunlei@gmail.com>
 xdrudis <xavierdrudis@yahoo.es>
-yuri@FreeBSD <yuri@FreeBSD>
-zhangjixiong <code.zjx@gmail.com>
-zhentaoyu <zhentao.yu@intel.com>
 zhouwg <6889919+zhouwg@users.noreply.github.com>
-zhouwg <zhouwg2000@gmail.com>
-谢乃闻 <sienaiwun@users.noreply.github.com>
 布客飞龙 <562826179@qq.com>
 Артём Земляк <azemlyak@smart-consulting.ru>
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@ -1,31 +1,25 @@
-cmake_minimum_required(VERSION 3.5) # for add_link_options and implicit target directories.
-project("whisper.cpp" C CXX)
-project("whisper.cpp" VERSION 1.7.5)
-include(CheckIncludeFileCXX)
+cmake_minimum_required (VERSION 3.5)

+# Allow for the creation of solution folders.
+set_property(GLOBAL PROPERTY USE_FOLDERS ON)
+
+project(whisper.cpp VERSION 1.6.2)
 set(SOVERSION 1)

-#set(CMAKE_WARN_DEPRECATED YES)
-set(CMAKE_WARN_UNUSED_CLI YES)
-
-set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
-
-if (NOT XCODE AND NOT MSVC AND NOT CMAKE_BUILD_TYPE)
-    set(CMAKE_BUILD_TYPE Release CACHE STRING "Build type" FORCE)
-    set_property(CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS "Debug" "Release" "MinSizeRel" "RelWithDebInfo")
-endif()
-
 # Add path to modules
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/")

 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)

-if (CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR)
+if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR)
    set(WHISPER_STANDALONE ON)
-
-    include(git-vars)
+    include(GitVars)
+    include(BuildTypes)

    # configure project version
+    if (EXISTS "${CMAKE_SOURCE_DIR}/bindings/ios/Makefile-tmpl")
+        configure_file(${CMAKE_SOURCE_DIR}/bindings/ios/Makefile-tmpl ${CMAKE_SOURCE_DIR}/bindings/ios/Makefile @ONLY)
+    endif()
    configure_file(${CMAKE_SOURCE_DIR}/bindings/javascript/package-tmpl.json ${CMAKE_SOURCE_DIR}/bindings/javascript/package.json @ONLY)
 else()
    set(WHISPER_STANDALONE OFF)
@ -35,16 +29,6 @@ if (EMSCRIPTEN)
    set(BUILD_SHARED_LIBS_DEFAULT OFF)

    option(WHISPER_WASM_SINGLE_FILE "whisper: embed WASM inside the generated whisper.js" ON)
-
-    # TODO: without these, we get the following error:
-    #       wasm-ld: error: --shared-memory is disallowed by whisper.cpp.o because it was not compiled with 'atomics' or 'bulk-memory' features.
-    set(CMAKE_C_FLAGS   "${CMAKE_C_FLAGS}   -pthread")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -pthread")
-
-    set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -s TOTAL_STACK=5242880")
-    set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -s TOTAL_STACK=5242880")
-
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-deprecated")
 else()
    if (MINGW)
        set(BUILD_SHARED_LIBS_DEFAULT OFF)
@ -53,199 +37,758 @@ else()
    endif()
 endif()

-option(BUILD_SHARED_LIBS "build shared libraries" ${BUILD_SHARED_LIBS_DEFAULT})
+# options

-#
-# option list
-#
+if (APPLE)
+    set(WHISPER_METAL_DEFAULT ON)
+else()
+    set(WHISPER_METAL_DEFAULT OFF)
+endif()
+
+option(BUILD_SHARED_LIBS              "whisper: build shared libs" ${BUILD_SHARED_LIBS_DEFAULT})

-# debug
 option(WHISPER_ALL_WARNINGS           "whisper: enable all compiler warnings"                   ON)
 option(WHISPER_ALL_WARNINGS_3RD_PARTY "whisper: enable all compiler warnings in 3rd party libs" OFF)

-# build
-option(WHISPER_FATAL_WARNINGS  "whisper: enable -Werror flag"               OFF)
-option(WHISPER_USE_SYSTEM_GGML "whisper: use system-installed GGML library" OFF)
+option(WHISPER_SANITIZE_THREAD        "whisper: enable thread sanitizer"    OFF)
+option(WHISPER_SANITIZE_ADDRESS       "whisper: enable address sanitizer"   OFF)
+option(WHISPER_SANITIZE_UNDEFINED     "whisper: enable undefined sanitizer" OFF)

-# sanitizers
-option(WHISPER_SANITIZE_THREAD    "whisper: enable thread sanitizer"    OFF)
-option(WHISPER_SANITIZE_ADDRESS   "whisper: enable address sanitizer"   OFF)
-option(WHISPER_SANITIZE_UNDEFINED "whisper: enable undefined sanitizer" OFF)
+option(WHISPER_BUILD_TESTS            "whisper: build tests"    ${WHISPER_STANDALONE})
+option(WHISPER_BUILD_EXAMPLES         "whisper: build examples" ${WHISPER_STANDALONE})

-# extra artifacts
-option(WHISPER_BUILD_TESTS    "whisper: build tests"          ${WHISPER_STANDALONE})
-option(WHISPER_BUILD_EXAMPLES "whisper: build examples"       ${WHISPER_STANDALONE})
-option(WHISPER_BUILD_SERVER   "whisper: build server example" ${WHISPER_STANDALONE})
-
-# 3rd party libs
-option(WHISPER_CURL "whisper: use libcurl to download model from an URL" OFF)
-option(WHISPER_SDL2 "whisper: support for libSDL2" OFF)
+option(WHISPER_SDL2                   "whisper: support for libSDL2" OFF)

 if (CMAKE_SYSTEM_NAME MATCHES "Linux")
-    option(WHISPER_FFMPEG "whisper: support building and linking with ffmpeg libs (avcodec, swresample, ...)" OFF)
+    option(WHISPER_FFMPEG                 "whisper: support building and linking with ffmpeg libs (avcodec, swresample, ...)" OFF)
 endif()

-option(WHISPER_COREML                "whisper: enable Core ML framework"  OFF)
-option(WHISPER_COREML_ALLOW_FALLBACK "whisper: allow non-CoreML fallback" OFF)
-option(WHISPER_OPENVINO              "whisper: support for OpenVINO"      OFF)
+option(WHISPER_NO_AVX                 "whisper: disable AVX"         OFF)
+option(WHISPER_NO_AVX2                "whisper: disable AVX2"        OFF)
+option(WHISPER_NO_AVX512              "whisper: disable AVX512"      ON)
+option(WHISPER_NO_AVX512_VBMI         "whisper: disable AVX512-VBMI" ON)
+option(WHISPER_NO_AVX512_VNNI         "whisper: disable AVX512-VNNI" ON)
+option(WHISPER_NO_FMA                 "whisper: disable FMA"         OFF)
+option(WHISPER_NO_F16C                "whisper: disable F16c"        OFF)

-# Required for relocatable CMake package
-include(${CMAKE_CURRENT_SOURCE_DIR}/cmake/build-info.cmake)
+option(WHISPER_OPENVINO               "whisper: support for OpenVINO" OFF)

-# override ggml options
-set(GGML_SANITIZE_THREAD    ${WHISPER_SANITIZE_THREAD})
-set(GGML_SANITIZE_ADDRESS   ${WHISPER_SANITIZE_ADDRESS})
-set(GGML_SANITIZE_UNDEFINED ${WHISPER_SANITIZE_UNDEFINED})
-set(GGML_ALL_WARNINGS       ${WHISPER_ALL_WARNINGS})
-set(GGML_FATAL_WARNINGS     ${WHISPER_FATAL_WARNINGS})
+if (APPLE)
+    option(WHISPER_NO_ACCELERATE         "whisper: disable Accelerate framework" OFF)
+    option(WHISPER_METAL                 "whisper: use Metal"                    ${WHISPER_METAL_DEFAULT})
+    option(WHISPER_METAL_NDEBUG          "whisper: disable Metal debugging"      OFF)
+    option(WHISPER_COREML                "whisper: enable Core ML framework"     OFF)
+    option(WHISPER_COREML_ALLOW_FALLBACK "whisper: allow non-CoreML fallback"    OFF)
+    option(WHISPER_METAL_EMBED_LIBRARY   "whisper: embed Metal library"          OFF)
+else()
+    option(WHISPER_BLAS                  "whisper: use BLAS libraries"                        OFF)
+    option(WHISPER_BLAS_VENDOR           "whisper: BLAS library vendor"                       Generic)
+    option(WHISPER_OPENBLAS              "whisper: prefer OpenBLAS"                           OFF)
+    option(WHISPER_OPENBLAS_INTERFACE64  "whisper: use OpenBLAS w/ 64-bit interface"          OFF)
+    option(WHISPER_CUDA                  "whisper: support for CUDA"                          OFF)
+    option(WHISPER_CUBLAS                "whisper: support for CUDA (deprecated)"             OFF)
+    option(WHISPER_HIPBLAS               "whisper: support for hipBLAS"                       OFF)
+    option(WHISPER_CLBLAST               "whisper: use CLBlast"                               OFF)
+    option(WHISPER_MKL                   "whisper: use Intel Math Kernel Library (MKL)"       OFF)
+    option(WHISPER_SYCL                  "whisper: use SYCL"                                  OFF)
+    option(WHISPER_SYCL_F16              "whisper: use 16 bit floats for sycl calculations"   OFF)
+endif()

-# transition helpers
-function (whisper_option_depr TYPE OLD NEW)
-    if (${OLD})
-        message(${TYPE} "${OLD} is deprecated and will be removed in the future.\nUse ${NEW} instead\n")
-        set(${NEW} ON)
+option(WHISPER_PERF "whisper: enable perf timings" OFF)
+
+# sanitizers
+
+if (NOT MSVC)
+    if (WHISPER_SANITIZE_THREAD)
+        set(CMAKE_C_FLAGS   "${CMAKE_C_FLAGS}   -fsanitize=thread")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsanitize=thread")
    endif()
-endfunction()

-whisper_option_depr(FATAL_ERROR WHISPER_CUBLAS              GGML_CUDA)
-whisper_option_depr(WARNING     WHISPER_CUDA                GGML_CUDA)
-whisper_option_depr(WARNING     WHISPER_KOMPUTE             GGML_KOMPUTE)
-whisper_option_depr(WARNING     WHISPER_METAL               GGML_METAL)
-whisper_option_depr(WARNING     WHISPER_METAL_EMBED_LIBRARY GGML_METAL_EMBED_LIBRARY)
-whisper_option_depr(WARNING     WHISPER_NATIVE              GGML_NATIVE)
-whisper_option_depr(WARNING     WHISPER_OPENMP              GGML_OPENMP)
-whisper_option_depr(WARNING     WHISPER_RPC                 GGML_RPC)
-whisper_option_depr(WARNING     WHISPER_SYCL                GGML_SYCL)
-whisper_option_depr(WARNING     WHISPER_SYCL_F16            GGML_SYCL_F16)
-whisper_option_depr(WARNING     WHISPER_CCACHE              GGML_CCACHE)
+    if (WHISPER_SANITIZE_ADDRESS)
+        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}     -fsanitize=address -fno-omit-frame-pointer")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsanitize=address -fno-omit-frame-pointer")
+    endif()

-if (GGML_CUDA AND NOT MSVC)
-    #GGML_CUDA enabled, add the necessary compile options -Wno-deprecated-gpu-targets
-    add_compile_options(-Wno-deprecated-gpu-targets)
+    if (WHISPER_SANITIZE_UNDEFINED)
+        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}     -fsanitize=undefined")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsanitize=undefined")
+    endif()
 endif()

-#
-# build the library
-#
+#set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -ffast-math")
+#set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -march=native")

-if (NOT TARGET ggml)
-    if (WHISPER_USE_SYSTEM_GGML)
-        find_package(ggml REQUIRED)
-        if (NOT ggml_FOUND)
-            message(FATAL_ERROR "System-installed GGML library not found.")
+# dependencies
+
+find_package(Threads REQUIRED)
+
+#compile flag sycl
+if (WHISPER_SYCL)
+    set(CMAKE_CXX_STANDARD 17)
+else()
+    set(CMAKE_CXX_STANDARD 11)
+endif()
+
+if (WHISPER_FFMPEG)
+    # As of cmake 3.27, there is no official cmake support for FindFFmpeg.
+    # Consequnelty we added a FindFFmpeg.cmake script the cmake subfolder:
+    # whisper.cpp does not need the full ffmpeg libs, just AVFORMAT AVCODEC AVUTIL SWRESAMPLE
+    # libswresample  performs highly optimized audio resampling, rematrixing and sample format conversion operations
+    # libavcodec provides a generic encoding/decoding framework and contains multiple decoders and encoders for audio, video and subtitle streams, and several bitstream filters.
+    # libavformat provides a generic framework for multiplexing and demultiplexing (muxing and demuxing) audio, video and subtitle streams.
+    find_package(FFmpeg REQUIRED)
+    if (NOT ${FFMPEG_FOUND})
+        message(FATAL_ERROR "Cannot find ffmpeg libs/headers")
+    endif()
+    message(STATUS "Found ffmpeg libs: ${FFMPEG_LIBRARIES}")
+    message(STATUS "Found ffmpeg headers in: ${FFMPEG_INCLUDE_DIRS}")
+    message(STATUS "ffmpeg definitions: ${FFMPEG_DEFINITIONS}")
+    message(STATUS "Found avformat ${AVFORMAT_VERSION}")
+    include_directories(${FFMPEG_INCLUDE_DIRS})
+    add_compile_definitions(WHISPER_FFMPEG)
+    set(WHISPER_EXTRA_LIBS  ${WHISPER_EXTRA_LIBS} ${FFMPEG_LIBRARIES})
+endif()
+
+# on APPLE
+if (APPLE)
+    # include Accelerate framework
+    if (NOT WHISPER_NO_ACCELERATE)
+        find_library(ACCELERATE_FRAMEWORK Accelerate)
+
+        if (ACCELERATE_FRAMEWORK)
+            message(STATUS "Accelerate framework found")
+
+            set(WHISPER_EXTRA_LIBS  ${WHISPER_EXTRA_LIBS}  ${ACCELERATE_FRAMEWORK})
+            set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -DGGML_USE_ACCELERATE -DACCELERATE_NEW_LAPACK -DACCELERATE_LAPACK_ILP64)
+        else()
+            message(FATAL_ERROR "Accelerate framework not found")
        endif()
-        add_library(ggml ALIAS ggml::ggml)
+    endif()
+
+    if (WHISPER_METAL)
+        find_library(FOUNDATION_LIBRARY         Foundation              REQUIRED)
+        find_library(METAL_FRAMEWORK            Metal                   REQUIRED)
+        find_library(METALKIT_FRAMEWORK         MetalKit                REQUIRED)
+
+        if (METAL_FRAMEWORK)
+            message(STATUS "Metal framework found")
+
+            set(WHISPER_EXTRA_LIBS ${WHISPER_EXTRA_LIBS}
+                ${FOUNDATION_LIBRARY}
+                ${METAL_FRAMEWORK}
+                ${METALKIT_FRAMEWORK}
+                )
+            set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -DGGML_USE_METAL)
+
+            if (WHISPER_METAL_NDEBUG)
+                set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -DGGML_METAL_NDEBUG)
+            endif()
+        else()
+            message(FATAL_ERROR "Metal framework not found")
+        endif()
+
+        set(GGML_SOURCES_METAL ggml-metal.m ggml-metal.h)
+
+        # copy ggml-common.h and ggml-metal.metal to bin directory
+        configure_file(ggml-common.h    bin/ggml-common.h    COPYONLY)
+        configure_file(ggml-metal.metal bin/ggml-metal.metal COPYONLY)
+
+        if (WHISPER_METAL_EMBED_LIBRARY)
+            enable_language(ASM)
+            set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -DGGML_METAL_EMBED_LIBRARY)
+
+            set(METALLIB_SOURCE "${CMAKE_CURRENT_SOURCE_DIR}/ggml-metal.metal")
+            set(COMMON_HEADER   "${CMAKE_CURRENT_SOURCE_DIR}/ggml-common.h")
+
+            file(MAKE_DIRECTORY "${CMAKE_BINARY_DIR}/autogenerated")
+            set(EMBED_METALLIB_ASSEMBLY "${CMAKE_BINARY_DIR}/autogenerated/ggml-embed-metallib.s")
+            set(EMBED_METALLIB_SOURCE "${CMAKE_BINARY_DIR}/autogenerated/ggml-metal-combined.metal")
+
+            add_custom_command(
+                OUTPUT ${EMBED_METALLIB_SOURCE}
+                COMMAND sed -e "/^#include \\\"ggml-common.h\\\"/r ${COMMON_HEADER}" -e "/^#include \\\"ggml-common.h\\\"/d" ${METALLIB_SOURCE} > ${EMBED_METALLIB_SOURCE}
+                DEPENDS ${METALLIB_SOURCE} ${COMMON_HEADER}
+                COMMENT "Generating combined Metal library for embedding"
+            )
+
+            add_custom_command(
+                OUTPUT ${EMBED_METALLIB_ASSEMBLY}
+                COMMAND echo ".section __DATA,__ggml_metallib" > ${EMBED_METALLIB_ASSEMBLY}
+                COMMAND echo ".globl _ggml_metallib_start" >> ${EMBED_METALLIB_ASSEMBLY}
+                COMMAND echo "_ggml_metallib_start:" >> ${EMBED_METALLIB_ASSEMBLY}
+                COMMAND echo ".incbin \\\"${EMBED_METALLIB_SOURCE}\\\"" >> ${EMBED_METALLIB_ASSEMBLY}
+                COMMAND echo ".globl _ggml_metallib_end" >> ${EMBED_METALLIB_ASSEMBLY}
+                COMMAND echo "_ggml_metallib_end:" >> ${EMBED_METALLIB_ASSEMBLY}
+                DEPENDS ${EMBED_METALLIB_SOURCE}
+                COMMENT "Generate assembly for embedded Metal library"
+            )
+
+            set(GGML_SOURCES_METAL ${GGML_SOURCES_METAL} ${EMBED_METALLIB_ASSEMBLY})
+        endif()
+    endif()
+
+    if (WHISPER_COREML)
+        find_library(FOUNDATION_FRAMEWORK Foundation)
+        find_library(COREML_FRAMEWORK CoreML)
+
+        if (COREML_FRAMEWORK)
+            message(STATUS "CoreML framework found")
+
+            set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -DWHISPER_USE_COREML)
+        else()
+            message(FATAL_ERROR "CoreML framework not found")
+        endif()
+
+        if (WHISPER_COREML_ALLOW_FALLBACK)
+            set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -DWHISPER_COREML_ALLOW_FALLBACK)
+        endif()
+    endif()
+endif()
+
+if (WHISPER_OPENBLAS)
+    set(WHISPER_BLAS_VENDOR "OpenBLAS")
+    set(WHISPER_BLAS ON)
+    # BLA_PKGCONFIG_BLAS is supported since CMake 3.25.
+    # FindBLAS.cmake pkg-config logic seems incomplete, because when
+    # BLA_SIZEOF_INTEGER is 8, then it should search for blas64 instead of blas.
+    # blas.pc/blas64.pc are not always provided, so let's be more specific
+    # and go with openblas.pc/openblas64.pc if WHISPER_OPENBLAS is on.
+    if (WHISPER_OPENBLAS_INTERFACE64)
+        set(WHISPER_BLAS_LIB "openblas64")
+    else ()
+        set(WHISPER_BLAS_LIB "openblas")
+    endif ()
+    set(BLA_PKGCONFIG_BLAS ${WHISPER_BLAS_LIB})
+    # OpenBLAS prebuilt libraries for Windows do not have "64" suffix in filename.
+    # (But .pc file has "64" suffix in filename for USE_64BITINT=1 Windows build.)
+    if (MSVC)
+        set(WHISPER_BLAS_LIB "openblas")
+    endif ()
+endif()
+
+if (WHISPER_BLAS)
+    if (NOT "$ENV{OPENBLAS_PATH}" STREQUAL "")
+        if (WHISPER_STATIC)
+            set(WHISPER_BLAS_LIB_PREFIX ${CMAKE_STATIC_LIBRARY_PREFIX})
+            set(WHISPER_BLAS_LIB_SUFFIX ${CMAKE_STATIC_LIBRARY_SUFFIX})
+        else ()
+            if (CMAKE_IMPORT_LIBRARY_SUFFIX)
+                set(WHISPER_BLAS_LIB_PREFIX ${CMAKE_IMPORT_LIBRARY_PREFIX})
+                set(WHISPER_BLAS_LIB_SUFFIX ${CMAKE_IMPORT_LIBRARY_SUFFIX})
+            else ()
+                set(WHISPER_BLAS_LIB_PREFIX ${CMAKE_SHARED_LIBRARY_PREFIX})
+                set(WHISPER_BLAS_LIB_SUFFIX ${CMAKE_SHARED_LIBRARY_SUFFIX})
+            endif ()
+        endif ()
+        # OpenBLAS prebuilt libraries hardcode "lib" prefix in filename even on Windows
+        if (WHISPER_OPENBLAS)
+            set(WHISPER_BLAS_LIB_PREFIX "lib")
+        endif ()
+        message(STATUS "BLAS compatible library path provided")
+        set(BLAS_LIBRARIES "$ENV{OPENBLAS_PATH}/lib/${WHISPER_BLAS_LIB_PREFIX}${WHISPER_BLAS_LIB}${WHISPER_BLAS_LIB_SUFFIX}")
+        message(STATUS "Libraries ${BLAS_LIBRARIES}")
+        set(BLAS_INCLUDE_DIRS "$ENV{OPENBLAS_PATH}/include")
+        message(STATUS "Include dirs ${BLAS_INCLUDE_DIRS}")
+        if (NOT EXISTS "${BLAS_LIBRARIES}")
+            message(FATAL_ERROR "BLAS library was not found. Environment variable OPENBLAS_PATH misdefined.")
+        endif ()
+        set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -DGGML_USE_OPENBLAS)
+        include_directories(${BLAS_INCLUDE_DIRS})
+        set(WHISPER_EXTRA_LIBS ${WHISPER_EXTRA_LIBS} ${BLAS_LIBRARIES})
+    else ()
+        if (WHISPER_STATIC)
+            # FindBLAS.cmake pkg-config logic seems incomplete, because when
+            # BLA_STATIC is on, then it should use pkg_check_modules_static
+            # instead of pkg_check_modules.
+            # Some manual variable overriding may be necessary if you don't
+            # achieve desired results.
+            set(BLA_STATIC 1)
+        endif ()
+        set(BLA_VENDOR ${WHISPER_BLAS_VENDOR})
+        if (WHISPER_OPENBLAS_INTERFACE64)
+            set(BLA_SIZEOF_INTEGER 8)
+        else ()
+            set(BLA_SIZEOF_INTEGER 4)
+        endif()
+        set(BLA_PREFER_PKGCONFIG 1)
+        find_package(BLAS)
+
+        if(BLAS_FOUND)
+            message(STATUS "BLAS compatible library found")
+            message(STATUS "Libraries ${BLAS_LIBRARIES}")
+            if (NOT DEFINED BLAS_INCLUDE_DIRS)
+                if (PKGC_BLAS_FOUND)
+                    set(BLAS_INCLUDE_DIRS "${PKGC_BLAS_INCLUDE_DIRS}")
+                else ()
+                    find_path(BLAS_INCLUDE_DIRS cblas.h /usr/include/openblas)
+                endif()
+            endif()
+            message(STATUS "Include dirs ${BLAS_INCLUDE_DIRS}")
+            set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -DGGML_USE_OPENBLAS)
+            include_directories(${BLAS_INCLUDE_DIRS})
+            set(WHISPER_EXTRA_LIBS ${WHISPER_EXTRA_LIBS} ${BLAS_LIBRARIES})
+        else()
+            message(FATAL_ERROR "BLAS library was not found")
+        endif()
+    endif ()
+endif ()
+
+if (WHISPER_MKL)
+    find_package(MKL CONFIG REQUIRED PATHS $ENV{MKLROOT})
+    message(STATUS "Imported oneMKL targets: ${MKL_IMPORTED_TARGETS}")
+    set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -DGGML_USE_OPENBLAS)
+    set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -DGGML_BLAS_USE_MKL)
+endif()
+
+if (WHISPER_CUBLAS)
+    message(WARNING "WHISPER_CUBLAS is deprecated and will be removed in the future.\nUse WHISPER_CUDA instead")
+    set(WHISPER_CUDA ON)
+endif()
+
+if (WHISPER_CUDA)
+    cmake_minimum_required(VERSION 3.17)
+
+    find_package(CUDAToolkit)
+
+    if (CUDAToolkit_FOUND)
+        message(STATUS "cuBLAS found")
+
+        enable_language(CUDA)
+
+        file(GLOB   GGML_SOURCES_CUDA "ggml-cuda/*.cu")
+        list(APPEND GGML_SOURCES_CUDA  ggml-cuda.h)
+        list(APPEND GGML_SOURCES_CUDA  ggml-cuda.cu)
+
+        add_compile_definitions(GGML_USE_CUDA)
+
+        if (WHISPER_STATIC)
+            if (WIN32)
+                # As of 12.3.1 CUDA Tookit for Windows does not offer a static cublas library
+                set(WHISPER_EXTRA_LIBS ${WHISPER_EXTRA_LIBS} CUDA::cudart_static CUDA::cublas CUDA::cublasLt CUDA::cufft)
+            else ()
+                set(WHISPER_EXTRA_LIBS ${WHISPER_EXTRA_LIBS} CUDA::cudart_static CUDA::cublas_static CUDA::cublasLt_static CUDA::cufft_static)
+            endif()
+        else()
+            set(WHISPER_EXTRA_LIBS ${WHISPER_EXTRA_LIBS} CUDA::cudart CUDA::cublas CUDA::cublasLt CUDA::cufft)
+        endif()
+
+        set(WHISPER_EXTRA_LIBS ${WHISPER_EXTRA_LIBS} CUDA::cuda_driver)
    else()
-        add_subdirectory(ggml)
-        if(WIN32)
-            # The following adds a _DISABLE_CONSTEXPR_MUTEX_CONSTRUCTOR macro and is a workaround for
-            # the Windows C++ standard library which does not support constexpr mutexes.
-            # From the release notes://github.com/microsoft/STL/wiki/Changelog
-            #  Disable constexpr mutex constructor on Windows
-            #  Fixed mutex's constructor to be constexpr. #3824 #4000 #4339
-            #  Note: Programs that aren't following the documented restrictions on binary compatibility may encounter
-            #  null dereferences in mutex machinery. You must follow this rule:
-            #  When you mix binaries built by different supported versions of the toolset, the Redistributable version
-            #  must be at least as new as the latest toolset used by any app component.
-            #  You can define _DISABLE_CONSTEXPR_MUTEX_CONSTRUCTOR as an escape hatch.
-            #
-            # Specifically to whisper.cpp this would cause a crash when using the Java bindings.
-            # resulting in a Invalid memory access error.
-            target_compile_definitions(ggml-base PRIVATE _DISABLE_CONSTEXPR_MUTEX_CONSTRUCTOR)
+        message(FATAL_ERROR "cuBLAS not found")
+    endif()
+endif()
+
+
+if (WHISPER_HIPBLAS)
+    list(APPEND CMAKE_PREFIX_PATH /opt/rocm)
+    if (NOT ${CMAKE_C_COMPILER_ID} MATCHES "Clang")
+        message(WARNING "Only LLVM is supported for HIP, hint: CC=/opt/rocm/llvm/bin/clang")
+    endif()
+    if (NOT ${CMAKE_CXX_COMPILER_ID} MATCHES "Clang")
+        message(WARNING "Only LLVM is supported for HIP, hint: CXX=/opt/rocm/llvm/bin/clang++")
+    endif()
+
+    find_package(hip)
+    find_package(hipblas)
+    find_package(rocblas)
+
+    if (${hipblas_FOUND} AND ${hip_FOUND})
+        message(STATUS "HIP and hipBLAS found")
+        set(GGML_HEADERS_ROCM "ggml-cuda.h")
+
+        file(GLOB GGML_SOURCES_ROCM "ggml-cuda/*.cu")
+        list(APPEND GGML_SOURCES_ROCM "ggml-cuda.cu")
+
+        add_compile_definitions(GGML_USE_HIPBLAS GGML_USE_CUDA)
+
+        set_source_files_properties(${GGML_SOURCES_ROCM} PROPERTIES LANGUAGE CXX)
+        if (WHISPER_STATIC)
+            message(FATAL_ERROR "Static linking not supported for HIP/ROCm")
+        endif()
+        set(WHISPER_EXTRA_LIBS ${WHISPER_EXTRA_LIBS} hip::device PUBLIC hip::host roc::rocblas roc::hipblas)
+    else()
+        message(FATAL_ERROR "hipBLAS or HIP not found. Try setting CMAKE_PREFIX_PATH=/opt/rocm")
+    endif()
+endif()
+
+if (WHISPER_CLBLAST)
+    find_package(CLBlast)
+    if (CLBlast_FOUND)
+        message(STATUS "CLBlast found")
+
+        set(GGML_SOURCES_OPENCL ggml-opencl.cpp ggml-opencl.h)
+
+        add_compile_definitions(GGML_USE_CLBLAST)
+
+        set(WHISPER_EXTRA_LIBS ${WHISPER_EXTRA_LIBS} clblast)
+    else()
+        message(FATAL_ERROR "CLBlast not found")
+    endif()
+endif()
+
+if( WHISPER_OPENVINO )
+    find_package(OpenVINO REQUIRED COMPONENTS Runtime)
+endif()
+
+if (WHISPER_SYCL)
+    if ( NOT DEFINED ENV{ONEAPI_ROOT})
+        message(FATAL_ERROR "Not detect ENV {ONEAPI_ROOT}, please install oneAPI & source it, like: source /opt/intel/oneapi/setvars.sh")
+    endif()
+    #todo: AOT
+
+    find_package(IntelSYCL REQUIRED)
+    if (WHISPER_SYCL_F16)
+        add_compile_definitions(GGML_SYCL_F16)
+    endif()
+    add_compile_definitions(GGML_USE_SYCL)
+
+    add_compile_options(-I./) #include DPCT
+    add_compile_options(-I/${SYCL_INCLUDE_DIR})
+
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-narrowing")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O3")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fsycl -L${MKLROOT}/lib")
+
+    set(GGML_HEADERS_SYCL ggml-sycl.h)
+    set(GGML_SOURCES_SYCL ggml-sycl.cpp)
+
+    set(WHISPER_EXTRA_LIBS ${WHISPER_EXTRA_LIBS} sycl OpenCL mkl_core pthread m dl mkl_sycl_blas mkl_intel_ilp64 mkl_tbb_thread)
+endif()
+# compiler flags
+
+if (NOT CMAKE_BUILD_TYPE AND NOT CMAKE_CONFIGURATION_TYPES)
+    set(CMAKE_BUILD_TYPE Release CACHE STRING "Build type" FORCE)
+    set_property(CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS "Debug" "Release" "RelWithDebInfo")
+endif ()
+
+if (WHISPER_ALL_WARNINGS)
+    if (NOT MSVC)
+        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} \
+            -Wall                           \
+            -Wextra                         \
+            -Wpedantic                      \
+            -Wshadow                        \
+            -Wcast-qual                     \
+            -Wstrict-prototypes             \
+            -Wpointer-arith                 \
+            -Wno-unused-function            \
+        ")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} \
+            -Wall                           \
+            -Wextra                         \
+            -Wpedantic                      \
+            -Wcast-qual                     \
+        ")
+    else()
+        # todo : msvc
+    endif()
+endif()
+
+if (NOT MSVC)
+    # TODO: temporary disabled until we figure out ggml-metal.m
+    #set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Werror=vla")
+    #set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fno-math-errno -ffinite-math-only -funsafe-math-optimizations")
+endif()
+
+message(STATUS "CMAKE_SYSTEM_PROCESSOR: ${CMAKE_SYSTEM_PROCESSOR}")
+
+if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm" OR ${CMAKE_SYSTEM_PROCESSOR} MATCHES "aarch64")
+    message(STATUS "ARM detected")
+elseif(${CMAKE_SYSTEM_PROCESSOR} MATCHES "ppc64le")
+    message(STATUS "PowerPC detected")
+else()
+    message(STATUS "x86 detected")
+    if (MSVC)
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /utf-8")
+        set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /utf-8")
+        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /utf-8")
+        if(NOT WHISPER_NO_AVX512)
+            set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:AVX512")
+            set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /arch:AVX512")
+            set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /arch:AVX512")
+            # MSVC has no compile-time flags enabling specific
+            # AVX512 extensions, neither it defines the
+            # macros corresponding to the extensions.
+            # Do it manually.
+            if (NOT WHISPER_NO_AVX512_VBMI)
+                add_compile_definitions($<$<COMPILE_LANGUAGE:C>:__AVX512VBMI__>)
+                add_compile_definitions($<$<COMPILE_LANGUAGE:CXX>:__AVX512VBMI__>)
+            endif()
+            if (NOT WHISPER_NO_AVX512_VNNI)
+                add_compile_definitions($<$<COMPILE_LANGUAGE:C>:__AVX512VNNI__>)
+                add_compile_definitions($<$<COMPILE_LANGUAGE:CXX>:__AVX512VNNI__>)
+            endif()
+        elseif(NOT WHISPER_NO_AVX2)
+            set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:AVX2")
+            set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /arch:AVX2")
+            set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /arch:AVX2")
+        elseif(NOT WHISPER_NO_AVX)
+            set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:AVX")
+            set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /arch:AVX")
+            set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /arch:AVX")
+        endif()
+    else()
+        if (EMSCRIPTEN)
+            set(CMAKE_C_FLAGS   "${CMAKE_C_FLAGS}   -pthread -s TOTAL_STACK=5242880")
+            set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -pthread -s TOTAL_STACK=5242880")
+        else()
+            if(NOT WHISPER_NO_AVX)
+                set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mavx")
+            endif()
+            if(NOT WHISPER_NO_AVX2)
+                set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mavx2")
+            endif()
+            if(NOT WHISPER_NO_AVX512)
+                set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mavx512f -mavx512cd -mavx512vl -mavx512dq -mavx512bw")
+                if(NOT WHISPER_NO_AVX512_VBMI)
+                    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mavx512vbmi")
+                endif()
+                if(NOT WHISPER_NO_AVX512_VNNI)
+                    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mavx512vnni")
+                endif()
+            endif()
+            if(NOT WHISPER_NO_FMA)
+                set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mfma")
+            endif()
+            if(NOT WHISPER_NO_F16C)
+                set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mf16c")
+            endif()
        endif()
    endif()
-    # ... otherwise assume ggml is added by a parent CMakeLists.txt
 endif()
-add_subdirectory(src)

 #
-# install
+# POSIX conformance
 #

+# clock_gettime came in POSIX.1b (1993)
+# CLOCK_MONOTONIC came in POSIX.1-2001 / SUSv3 as optional
+# posix_memalign came in POSIX.1-2001 / SUSv3
+# M_PI is an XSI extension since POSIX.1-2001 / SUSv3, came in XPG1 (1985)
+add_compile_definitions(_XOPEN_SOURCE=600)
+
+# Somehow in OpenBSD whenever POSIX conformance is specified
+# some string functions rely on locale_t availability,
+# which was introduced in POSIX.1-2008, forcing us to go higher
+if (CMAKE_SYSTEM_NAME MATCHES "OpenBSD")
+    remove_definitions(-D_XOPEN_SOURCE=600)
+    add_compile_definitions(_XOPEN_SOURCE=700)
+endif()
+
+# Data types, macros and functions related to controlling CPU affinity
+# are available on Linux through GNU extensions in libc
+if (CMAKE_SYSTEM_NAME MATCHES "Linux")
+    add_compile_definitions(_GNU_SOURCE)
+endif()
+
+# RLIMIT_MEMLOCK came in BSD, is not specified in POSIX.1,
+# and on macOS its availability depends on enabling Darwin extensions
+# similarly on DragonFly, enabling BSD extensions is necessary
+if (CMAKE_SYSTEM_NAME MATCHES "Darwin")
+    add_compile_definitions(_DARWIN_C_SOURCE)
+endif()
+if (CMAKE_SYSTEM_NAME MATCHES "DragonFly")
+    add_compile_definitions(_DARWIN_C_SOURCE)
+endif()
+
+# alloca is a non-standard interface that is not visible on BSDs when
+# POSIX conformance is specified, but not all of them provide a clean way
+# to enable it in such cases
+if (CMAKE_SYSTEM_NAME MATCHES "FreeBSD")
+    add_compile_definitions(__BSD_VISIBLE)
+endif()
+if (CMAKE_SYSTEM_NAME MATCHES "NetBSD")
+    add_compile_definitions(_NETBSD_SOURCE)
+endif()
+if (CMAKE_SYSTEM_NAME MATCHES "OpenBSD")
+    add_compile_definitions(_BSD_SOURCE)
+endif()
+
+if (WHISPER_PERF)
+    set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -DGGML_PERF)
+endif()
+
+#
+# whisper.coreml - Core ML support
+#
+
+if (WHISPER_COREML)
+    set(TARGET whisper.coreml)
+
+    add_library(${TARGET}
+        coreml/whisper-encoder.h
+        coreml/whisper-encoder.mm
+        coreml/whisper-encoder-impl.h
+        coreml/whisper-encoder-impl.m
+        )
+
+    include(DefaultTargetOptions)
+
+    target_include_directories(${TARGET} PUBLIC
+        .
+        )
+
+    target_link_libraries(${TARGET} PRIVATE ${FOUNDATION_FRAMEWORK} ${COREML_FRAMEWORK})
+
+    set_target_properties(${TARGET} PROPERTIES
+        COMPILE_FLAGS "-fobjc-arc"
+        )
+    set_target_properties(${TARGET} PROPERTIES FOLDER "libs")
+endif()
+
+if (WHISPER_OPENVINO)
+    set(TARGET whisper.openvino)
+
+    add_library(${TARGET} OBJECT
+        openvino/whisper-openvino-encoder.h
+        openvino/whisper-openvino-encoder.cpp
+        )
+
+    target_include_directories(${TARGET} PUBLIC
+        .
+        )
+
+    set_property(TARGET ${TARGET} PROPERTY POSITION_INDEPENDENT_CODE ON)
+    set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -DWHISPER_USE_OPENVINO)
+
+    target_link_libraries(${TARGET} PRIVATE openvino::runtime)
+    set_target_properties(${TARGET} PROPERTIES FOLDER "libs")
+endif()
+
+#
+# whisper - this is the main library of the project
+#
+
+set(TARGET whisper)
+
+add_library(${TARGET}
+    ggml.h
+    ggml.c
+    ggml-alloc.h
+    ggml-alloc.c
+    ggml-backend.h
+    ggml-backend.c
+    ggml-quants.h
+    ggml-quants.c
+    ${GGML_SOURCES_METAL}
+    ${GGML_SOURCES_CUDA}
+    ${GGML_SOURCES_OPENCL}
+    ${GGML_SOURCES_SYCL}        ${GGML_HEADERS_SYCL}
+    ${GGML_SOURCES_ROCM}        ${GGML_HEADERS_ROCM}
+    whisper.h
+    whisper.cpp
+    )
+
+if (WHISPER_CUDA)
+    target_sources(${TARGET} PRIVATE whisper-mel-cuda.cu)
+endif()
+
+include_directories (
+    .
+)
+# Set the version numbers
+set_target_properties(whisper PROPERTIES
+    VERSION ${PROJECT_VERSION}
+    SOVERSION ${SOVERSION}
+)
+
+include(DefaultTargetOptions)
+
+target_include_directories(${TARGET} PUBLIC
+    .
+    )
+
+if (WHISPER_COREML)
+    target_link_libraries(${TARGET} PRIVATE whisper.coreml)
+endif()
+
+if (WHISPER_OPENVINO)
+    target_link_libraries(${TARGET} PRIVATE whisper.openvino)
+endif()
+
+if (WHISPER_MKL)
+    target_link_libraries(${TARGET} PUBLIC MKL::MKL)
+endif()
+
+if (MSVC)
+    target_link_libraries(${TARGET} PRIVATE ${WHISPER_EXTRA_LIBS} ${CMAKE_THREAD_LIBS_INIT})
+
+    set(WHISPER_EXTRA_FLAGS ${WHISPER_EXTRA_FLAGS} -D_CRT_SECURE_NO_WARNINGS)
+else()
+    target_link_libraries(${TARGET} PRIVATE m ${WHISPER_EXTRA_LIBS} ${CMAKE_THREAD_LIBS_INIT})
+endif()
+
+if (BUILD_SHARED_LIBS)
+    set_target_properties(${TARGET} PROPERTIES POSITION_INDEPENDENT_CODE ON)
+    target_link_libraries(${TARGET} PUBLIC
+        ${CMAKE_DL_LIBS}
+        )
+
+    target_compile_definitions(${TARGET} PUBLIC
+        WHISPER_SHARED
+        GGML_SHARED
+        )
+
+    target_compile_definitions(${TARGET} PRIVATE
+        WHISPER_BUILD
+        GGML_BUILD
+        )
+
+    if (WHISPER_METAL)
+        # TODO: I think this should make ggml-metal.m "see" the ggml-metal.metal file from the "bin" directory
+        #       but for some reason it does not work here like it does in llama.cpp
+        set_target_properties(${TARGET} PROPERTIES RESOURCE "${CMAKE_CURRENT_SOURCE_DIR}/ggml-metal.metal")
+    endif()
+endif()
+
+if (GGML_SOURCES_CUDA)
+    message(STATUS "GGML CUDA sources found, configuring CUDA architecture")
+    # Only configure gmml CUDA architectures is not globally set
+    if (NOT DEFINED GGML_CUDA_ARCHITECTURES)
+        # Not overriden by user, so set defaults
+        set(GGML_CUDA_ARCHITECTURES 52 61 70)
+    endif()
+    message(STATUS "GGML Configuring CUDA architectures ${GGML_CUDA_ARCHITECTURES}")
+    set_property(TARGET whisper PROPERTY CUDA_ARCHITECTURES ${GGML_CUDA_ARCHITECTURES})
+    set_property(TARGET whisper PROPERTY CUDA_SELECT_NVCC_ARCH_FLAGS "Auto")
+endif()
+
+if (EMSCRIPTEN)
+    set_target_properties(${TARGET} PROPERTIES COMPILE_FLAGS "-msimd128")
+endif()
+
+target_compile_definitions(${TARGET} PUBLIC
+    ${WHISPER_EXTRA_FLAGS}
+    )
+
+set_target_properties(${TARGET} PROPERTIES PUBLIC_HEADER "ggml.h;whisper.h")
+set_target_properties(${TARGET} PROPERTIES FOLDER "libs")
+
 include(GNUInstallDirs)
-include(CMakePackageConfigHelpers)

-set(WHISPER_BUILD_NUMBER        ${BUILD_NUMBER})
-set(WHISPER_BUILD_COMMIT        ${BUILD_COMMIT})
-set(WHISPER_INSTALL_VERSION     ${CMAKE_PROJECT_VERSION})
+install(TARGETS ${TARGET}
+    LIBRARY  DESTINATION lib
+    ARCHIVE  DESTINATION lib/static
+    RUNTIME  DESTINATION bin
+    RESOURCE DESTINATION bin
+    PUBLIC_HEADER DESTINATION include
+    )

-set(WHISPER_INCLUDE_INSTALL_DIR ${CMAKE_INSTALL_INCLUDEDIR} CACHE PATH "Location of header  files")
-set(WHISPER_LIB_INSTALL_DIR     ${CMAKE_INSTALL_LIBDIR}     CACHE PATH "Location of library files")
-set(WHISPER_BIN_INSTALL_DIR     ${CMAKE_INSTALL_BINDIR}     CACHE PATH "Location of binary  files")
+#
+# bindings
+#

-get_directory_property(WHISPER_TRANSIENT_DEFINES COMPILE_DEFINITIONS)
-
-set_target_properties(whisper PROPERTIES PUBLIC_HEADER ${CMAKE_CURRENT_SOURCE_DIR}/include/whisper.h)
-install(TARGETS whisper LIBRARY PUBLIC_HEADER)
-
-configure_package_config_file(
-        ${CMAKE_CURRENT_SOURCE_DIR}/cmake/whisper-config.cmake.in
-        ${CMAKE_CURRENT_BINARY_DIR}/whisper-config.cmake
-    INSTALL_DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/whisper
-    PATH_VARS
-    WHISPER_INCLUDE_INSTALL_DIR
-    WHISPER_LIB_INSTALL_DIR
-    WHISPER_BIN_INSTALL_DIR )
-
-write_basic_package_version_file(
-    ${CMAKE_CURRENT_BINARY_DIR}/whisper-version.cmake
-    VERSION ${WHISPER_INSTALL_VERSION}
-    COMPATIBILITY SameMajorVersion)
-
-install(FILES ${CMAKE_CURRENT_BINARY_DIR}/whisper-config.cmake
-              ${CMAKE_CURRENT_BINARY_DIR}/whisper-version.cmake
-        DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/whisper)
-
-configure_file(cmake/whisper.pc.in
-        "${CMAKE_CURRENT_BINARY_DIR}/whisper.pc"
-        @ONLY)
-
-install(FILES "${CMAKE_CURRENT_BINARY_DIR}/whisper.pc"
-        DESTINATION lib/pkgconfig)
+add_subdirectory(bindings)

 #
 # programs, examples and tests
 #

 if (WHISPER_BUILD_TESTS AND NOT CMAKE_JS_VERSION)
-    include(CTest)
+    enable_testing()
    add_subdirectory(tests)
 endif ()

 if (WHISPER_BUILD_EXAMPLES)
    add_subdirectory(examples)
 endif()
-
-if (MSVC)
-    set(MSVC_WARNING_FLAGS
-        /wd4101  # Unreferenced local variable
-        /wd4005  # Macro redefinition
-        /wd4065  # switch statement contains 'default' but no 'case' labels
-        /wd4267  # Conversion from 'size_t' to a smaller type, possible loss of data
-        /wd4244  # Conversion from one type to another type, possible loss of ata
-        /wd4805  # Unsafe mix of type
-        /wd4305  # Truncation from 'type1' to 'type2' (often double to float)
-        /wd4996  # Function or variable may be unsafe/deprecated
-    )
-    function(disable_msvc_warnings target_name)
-        if(TARGET ${target_name})
-            target_compile_options(${target_name} PRIVATE ${MSVC_WARNING_FLAGS})
-        endif()
-    endfunction()
-
-    if (WHISPER_BUILD_EXAMPLES)
-        disable_msvc_warnings(whisper)
-        disable_msvc_warnings(common)
-        disable_msvc_warnings(common-sdl)
-        disable_msvc_warnings(lsp)
-        disable_msvc_warnings(wchess-core)
-        disable_msvc_warnings(whisper-command)
-        disable_msvc_warnings(whisper-cli)
-        disable_msvc_warnings(whisper-server)
-        disable_msvc_warnings(whisper-stream)
-        disable_msvc_warnings(whisper-talk-llama)
-        disable_msvc_warnings(whisper-bench)
-        disable_msvc_warnings(quantize)
-        disable_msvc_warnings(vad-speech-segments)
-    endif()
-endif()
--- a/532
+++ b/532
@ -1,12 +1,504 @@
+default: main bench quantize server
+
+ifndef UNAME_S
+UNAME_S := $(shell uname -s)
+endif
+
+ifndef UNAME_P
+UNAME_P := $(shell uname -p)
+endif
+
+ifndef UNAME_M
+UNAME_M := $(shell uname -m)
+endif
+
+ifndef NVCC_VERSION
+	ifeq ($(call,$(shell which nvcc))$(.SHELLSTATUS),0)
+		NVCC_VERSION := $(shell nvcc --version | egrep -o "V[0-9]+.[0-9]+.[0-9]+" | cut -c2-)
+	endif
+endif
+
+# In GNU make default CXX is g++ instead of c++.  Let's fix that so that users
+# of non-gcc compilers don't have to provide g++ alias or wrapper.
+DEFCC  := cc
+DEFCXX := c++
+ifeq ($(origin CC),default)
+CC  := $(DEFCC)
+endif
+ifeq ($(origin CXX),default)
+CXX := $(DEFCXX)
+endif
+
+CCV  := $(shell $(CC) --version | head -n 1)
+CXXV := $(shell $(CXX) --version | head -n 1)
+
+# Mac OS + Arm can report x86_64
+# ref: https://github.com/ggerganov/whisper.cpp/issues/66#issuecomment-1282546789
+ifeq ($(UNAME_S),Darwin)
+	ifneq ($(UNAME_P),arm)
+		SYSCTL_M := $(shell sysctl -n hw.optional.arm64)
+		ifeq ($(SYSCTL_M),1)
+			# UNAME_P := arm
+			# UNAME_M := arm64
+			warn := $(warning Your arch is announced as x86_64, but it seems to actually be ARM64. Not fixing that can lead to bad performance. For more info see: https://github.com/ggerganov/whisper.cpp/issues/66\#issuecomment-1282546789)
+		endif
+	endif
+endif
+
+#
+# Compile flags
+#
+
+CFLAGS   = -I.              -O3 -DNDEBUG -std=c11   -fPIC
+CXXFLAGS = -I. -I./examples -O3 -DNDEBUG -std=c++11 -fPIC
+LDFLAGS  =
+
+ifdef MACOSX_DEPLOYMENT_TARGET
+	CFLAGS   += -mmacosx-version-min=$(MACOSX_DEPLOYMENT_TARGET)
+	CXXFLAGS += -mmacosx-version-min=$(MACOSX_DEPLOYMENT_TARGET)
+	LDFLAGS  += -mmacosx-version-min=$(MACOSX_DEPLOYMENT_TARGET)
+endif
+
+# clock_gettime came in POSIX.1b (1993)
+# CLOCK_MONOTONIC came in POSIX.1-2001 / SUSv3 as optional
+# posix_memalign came in POSIX.1-2001 / SUSv3
+# M_PI is an XSI extension since POSIX.1-2001 / SUSv3, came in XPG1 (1985)
+CFLAGS   += -D_XOPEN_SOURCE=600
+CXXFLAGS += -D_XOPEN_SOURCE=600
+
+# Somehow in OpenBSD whenever POSIX conformance is specified
+# some string functions rely on locale_t availability,
+# which was introduced in POSIX.1-2008, forcing us to go higher
+ifeq ($(UNAME_S),OpenBSD)
+	CFLAGS   += -U_XOPEN_SOURCE -D_XOPEN_SOURCE=700
+	CXXFLAGS += -U_XOPEN_SOURCE -D_XOPEN_SOURCE=700
+endif
+
+# Data types, macros and functions related to controlling CPU affinity
+# are available on Linux through GNU extensions in libc
+ifeq ($(UNAME_S),Linux)
+	CFLAGS   += -D_GNU_SOURCE
+	CXXFLAGS += -D_GNU_SOURCE
+endif
+
+# RLIMIT_MEMLOCK came in BSD, is not specified in POSIX.1,
+# and on macOS its availability depends on enabling Darwin extensions
+# similarly on DragonFly, enabling BSD extensions is necessary
+ifeq ($(UNAME_S),Darwin)
+	CFLAGS   += -D_DARWIN_C_SOURCE
+	CXXFLAGS += -D_DARWIN_C_SOURCE
+endif
+ifeq ($(UNAME_S),DragonFly)
+	CFLAGS   += -D__BSD_VISIBLE
+	CXXFLAGS += -D__BSD_VISIBLE
+endif
+
+# alloca is a non-standard interface that is not visible on BSDs when
+# POSIX conformance is specified, but not all of them provide a clean way
+# to enable it in such cases
+ifeq ($(UNAME_S),FreeBSD)
+	CFLAGS   += -D__BSD_VISIBLE
+	CXXFLAGS += -D__BSD_VISIBLE
+endif
+ifeq ($(UNAME_S),NetBSD)
+	CFLAGS   += -D_NETBSD_SOURCE
+	CXXFLAGS += -D_NETBSD_SOURCE
+endif
+ifeq ($(UNAME_S),OpenBSD)
+	CFLAGS   += -D_BSD_SOURCE
+	CXXFLAGS += -D_BSD_SOURCE
+endif
+
+# OS specific
+# TODO: support Windows
+ifeq ($(filter $(UNAME_S),Linux Darwin DragonFly FreeBSD NetBSD OpenBSD Haiku),$(UNAME_S))
+	CFLAGS   += -pthread
+	CXXFLAGS += -pthread
+endif
+
+# detect Windows
+ifneq ($(findstring _NT,$(UNAME_S)),)
+	_WIN32 := 1
+endif
+
+# Windows Sockets 2 (Winsock) for network-capable apps
+ifeq ($(_WIN32),1)
+	LWINSOCK2 := -lws2_32
+endif
+
+# Architecture specific
+# TODO: probably these flags need to be tweaked on some architectures
+#       feel free to update the Makefile for your architecture and send a pull request or issue
+ifeq ($(UNAME_M),$(filter $(UNAME_M),x86_64 i686 amd64))
+	ifeq ($(UNAME_S),Darwin)
+		CPUINFO_CMD := sysctl machdep.cpu.features machdep.cpu.leaf7_features
+	else ifeq ($(UNAME_S),Linux)
+		CPUINFO_CMD := cat /proc/cpuinfo
+	else ifneq (,$(filter MINGW32_NT% MINGW64_NT% MSYS_NT%,$(UNAME_S)))
+		CPUINFO_CMD := cat /proc/cpuinfo
+	else ifneq (,$(filter DragonFly FreeBSD,$(UNAME_S)))
+		CPUINFO_CMD := grep Features /var/run/dmesg.boot
+	else ifeq ($(UNAME_S),Haiku)
+		CPUINFO_CMD := sysinfo -cpu
+	endif
+
+	# x86 ISA extensions (chronological order)
+	ifdef CPUINFO_CMD
+		SSE3_M := $(shell $(CPUINFO_CMD) | grep -iwE 'PNI|SSE3')
+		SSSE3_M := $(shell $(CPUINFO_CMD) | grep -iw 'SSSE3')
+		AVX_M := $(shell $(CPUINFO_CMD) | grep -iwE 'AVX|AVX1.0')
+		F16C_M := $(shell $(CPUINFO_CMD) | grep -iw 'F16C')
+		FMA_M := $(shell $(CPUINFO_CMD) | grep -iw 'FMA')
+		AVX2_M := $(shell $(CPUINFO_CMD) | grep -iw 'AVX2')
+		AVX512F_M := $(shell $(CPUINFO_CMD) | grep -iw 'AVX512F')
+		AVX512VBMI_M := $(shell $(CPUINFO_CMD) | grep -iw 'AVX512VBMI')
+		AVX512VNNI_M := $(shell $(CPUINFO_CMD) | grep -iwE 'AVX512_VNNI|AVX512VNNI')
+
+		# AVX-512 has many subsets, so let's make it easy to disable them all
+		ifneq ($(filter-out 0,$(WHISPER_NO_AVX512)),)
+			AVX512F_M :=
+			AVX512VBMI_M :=
+			AVX512VNNI_M :=
+		endif
+
+		ifneq (,$(SSE3_M))
+			CFLAGS   += -msse3
+			CXXFLAGS += -msse3
+		endif
+
+		ifneq (,$(SSSE3_M))
+			CFLAGS   += -mssse3
+			CXXFLAGS += -mssse3
+		endif
+
+		ifneq (,$(AVX_M))
+			CFLAGS   += -mavx
+			CXXFLAGS += -mavx
+		endif
+
+		ifneq (,$(F16C_M))
+			CFLAGS   += -mf16c
+			CXXFLAGS += -mf16c
+		endif
+
+		ifneq (,$(FMA_M))
+			CFLAGS   += -mfma
+			CXXFLAGS += -mfma
+		endif
+
+		ifneq (,$(AVX2_M))
+			CFLAGS   += -mavx2
+			CXXFLAGS += -mavx2
+		endif
+
+		ifneq (,$(AVX512F_M))
+			CFLAGS   += -mavx512f -mavx512cd -mavx512vl -mavx512dq -mavx512bw
+			CXXFLAGS += -mavx512f -mavx512cd -mavx512vl -mavx512dq -mavx512bw
+		endif
+
+		ifneq (,$(AVX512VBMI_M))
+			CFLAGS   += -mavx512vbmi
+			CXXFLAGS += -mavx512vbmi
+		endif
+
+		ifneq (,$(AVX512VNNI_M))
+			CFLAGS   += -mavx512vnni
+			CXXFLAGS += -mavx512vnni
+		endif
+	endif
+endif
+
+ifneq ($(filter ppc64%,$(UNAME_M)),)
+	POWER9_M := $(shell grep "POWER9" /proc/cpuinfo)
+	ifneq (,$(findstring POWER9,$(POWER9_M)))
+		CFLAGS += -mpower9-vector
+	endif
+	# Require c++23's std::byteswap for big-endian support.
+	ifeq ($(UNAME_M),ppc64)
+		CXXFLAGS += -std=c++23 -DGGML_BIG_ENDIAN
+	endif
+endif
+
+ifndef WHISPER_NO_ACCELERATE
+	# Mac M1 - include Accelerate framework
+	ifeq ($(UNAME_S),Darwin)
+		CFLAGS  += -DGGML_USE_ACCELERATE
+		CFLAGS  += -DACCELERATE_NEW_LAPACK
+		CFLAGS  += -DACCELERATE_LAPACK_ILP64
+		LDFLAGS += -framework Accelerate
+	endif
+endif
+
+ifdef WHISPER_COREML
+	CXXFLAGS += -DWHISPER_USE_COREML
+	LDFLAGS  += -framework Foundation -framework CoreML
+
+ifdef WHISPER_COREML_ALLOW_FALLBACK
+	CXXFLAGS += -DWHISPER_COREML_ALLOW_FALLBACK
+endif
+endif
+
+ifndef WHISPER_NO_METAL
+	ifeq ($(UNAME_S),Darwin)
+		WHISPER_METAL := 1
+
+		CFLAGS   += -DGGML_USE_METAL
+		CXXFLAGS += -DGGML_USE_METAL
+		LDFLAGS  += -framework Foundation -framework Metal -framework MetalKit
+	endif
+endif
+
+ifneq ($(filter-out 0,$(WHISPER_OPENBLAS)),) # OpenBLAS
+	WHISPER_OPENBLAS_INTERFACE64 ?= 0 # use 32-bit interface by default
+	ifneq ($(filter-out 0,$(WHISPER_OPENBLAS_INTERFACE64)),)
+		WHISPER_BLAS_LIB := openblas64
+	else
+		WHISPER_BLAS_LIB := openblas
+	endif
+	ifneq ($(OPENBLAS_PATH),)
+		WHISPER_BLAS_CFLAGS  := -I$(OPENBLAS_PATH)/include
+		WHISPER_BLAS_LDFLAGS := -L$(OPENBLAS_PATH)/lib -l$(WHISPER_BLAS_LIB)
+	else
+		WHISPER_BLAS_LIB_PC_EXISTS := $(shell pkg-config --exists $(WHISPER_BLAS_LIB) && echo 1)
+		ifneq ($(filter-out 0,$(WHISPER_BLAS_LIB_PC_EXISTS)),)
+			WHISPER_BLAS_CFLAGS  := $(shell pkg-config --cflags $(WHISPER_BLAS_LIB))
+			WHISPER_BLAS_LDFLAGS := $(shell pkg-config --libs   $(WHISPER_BLAS_LIB))
+		else
+			WHISPER_BLAS_CFLAGS  := -I/usr/include/openblas
+			WHISPER_BLAS_LDFLAGS := -l$(WHISPER_BLAS_LIB)
+		endif
+	endif
+	CFLAGS  += $(WHISPER_BLAS_CFLAGS) -DGGML_USE_OPENBLAS
+	LDFLAGS += $(WHISPER_BLAS_LDFLAGS)
+endif
+
+ifdef WHISPER_CUBLAS
+# WHISPER_CUBLAS is deprecated and will be removed in the future
+	WHISPER_CUDA := 1
+endif
+
+ifdef WHISPER_CUDA
+	ifeq ($(shell expr $(NVCC_VERSION) \>= 11.6), 1)
+		CUDA_ARCH_FLAG ?= native
+	else
+		CUDA_ARCH_FLAG ?= all
+	endif
+
+	CFLAGS      += -DGGML_USE_CUDA -I/usr/local/cuda/include -I/opt/cuda/include -I$(CUDA_PATH)/targets/$(UNAME_M)-linux/include
+	CXXFLAGS    += -DGGML_USE_CUDA -I/usr/local/cuda/include -I/opt/cuda/include -I$(CUDA_PATH)/targets/$(UNAME_M)-linux/include
+	LDFLAGS     += -lcuda -lcublas -lculibos -lcudart -lcublasLt -lcufft -lpthread -ldl -lrt -L/usr/local/cuda/lib64 -L/opt/cuda/lib64 -L$(CUDA_PATH)/targets/$(UNAME_M)-linux/lib -L/usr/lib/wsl/lib
+	WHISPER_OBJ += ggml-cuda.o whisper-mel-cuda.o
+	WHISPER_OBJ += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/*.cu))
+	NVCC        = nvcc
+	NVCCFLAGS   = --forward-unknown-to-host-compiler -arch=$(CUDA_ARCH_FLAG)
+
+ggml-cuda/%.o: ggml-cuda/%.cu ggml-cuda/%.cuh ggml.h ggml-common.h ggml-cuda/common.cuh
+	$(NVCC) $(NVCCFLAGS) $(CXXFLAGS) -c $< -o $@
+
+ggml-cuda.o: ggml-cuda.cu ggml-cuda.h ggml.h ggml-backend.h ggml-backend-impl.h ggml-common.h $(wildcard ggml-cuda/*.cuh)
+	$(NVCC) $(NVCCFLAGS) $(CXXFLAGS) -Wno-pedantic -c $< -o $@
+endif
+
+whisper-mel-cuda.o: whisper-mel-cuda.cu whisper.h ggml.h ggml-backend.h whisper-mel.hpp whisper-mel-cuda.hpp
+	$(NVCC) $(NVCCFLAGS) $(CXXFLAGS) -Wno-pedantic -c $< -o $@
+
+ifdef WHISPER_HIPBLAS
+	ROCM_PATH   ?= /opt/rocm
+	HIPCC       ?= $(ROCM_PATH)/bin/hipcc
+	GPU_TARGETS ?= $(shell $(ROCM_PATH)/llvm/bin/amdgpu-arch)
+	CFLAGS      += -DGGML_USE_HIPBLAS -DGGML_USE_CUDA
+	CXXFLAGS    += -DGGML_USE_HIPBLAS -DGGML_USE_CUDA
+	LDFLAGS     += -L$(ROCM_PATH)/lib -Wl,-rpath=$(ROCM_PATH)/lib
+	LDFLAGS     += -lhipblas -lamdhip64 -lrocblas
+	HIPFLAGS    += $(addprefix --offload-arch=,$(GPU_TARGETS))
+	WHISPER_OBJ += ggml-cuda.o
+	WHISPER_OBJ += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/*.cu))
+
+ggml-cuda/%.o: ggml-cuda/%.cu ggml-cuda/%.cuh ggml.h ggml-common.h ggml-cuda/common.cuh
+	$(HIPCC) $(CXXFLAGS) $(HIPFLAGS) -x hip -c -o $@ $<
+
+ggml-cuda.o: ggml-cuda.cu ggml-cuda.h ggml.h ggml-backend.h ggml-backend-impl.h ggml-common.h $(wildcard ggml-cuda/*.cuh)
+	$(HIPCC) $(CXXFLAGS) $(HIPFLAGS) -x hip -c -o $@ $<
+endif
+
+ifdef WHISPER_CLBLAST
+	CFLAGS 		+= -DGGML_USE_CLBLAST
+	CXXFLAGS 	+= -DGGML_USE_CLBLAST
+	LDFLAGS	 	+= -lclblast
+	ifeq ($(UNAME_S),Darwin)
+		LDFLAGS	 	+= -framework OpenCL
+	else
+		LDFLAGS	    += -lOpenCL
+	endif
+	WHISPER_OBJ	+= ggml-opencl.o
+
+ggml-opencl.o: ggml-opencl.cpp ggml-opencl.h
+	$(CXX) $(CXXFLAGS) -c $< -o $@
+endif
+
+ifdef WHISPER_GPROF
+	CFLAGS   += -pg
+	CXXFLAGS += -pg
+endif
+
+ifneq ($(filter aarch64%,$(UNAME_M)),)
+	CFLAGS   += -mcpu=native
+	CXXFLAGS += -mcpu=native
+endif
+
+ifneq ($(filter armv6%,$(UNAME_M)),)
+	# 32-bit Raspberry Pi 1, 2, 3
+	CFLAGS += -mfpu=neon -mfp16-format=ieee -mno-unaligned-access
+endif
+
+ifneq ($(filter armv7%,$(UNAME_M)),)
+	# 32-bit ARM, for example on Armbian or possibly raspbian
+	#CFLAGS   += -mfpu=neon -mfp16-format=ieee -funsafe-math-optimizations -mno-unaligned-access
+	#CXXFLAGS += -mfpu=neon -mfp16-format=ieee -funsafe-math-optimizations -mno-unaligned-access
+
+	# 64-bit ARM on 32-bit OS, use these (TODO: auto-detect 64-bit)
+	CFLAGS   += -mfpu=neon-fp-armv8 -mfp16-format=ieee -funsafe-math-optimizations -mno-unaligned-access
+	CXXFLAGS += -mfpu=neon-fp-armv8 -mfp16-format=ieee -funsafe-math-optimizations -mno-unaligned-access
+endif
+
+ifneq ($(filter armv8%,$(UNAME_M)),)
+	# Raspberry Pi 4
+	CFLAGS   += -mfpu=neon-fp-armv8 -mfp16-format=ieee -funsafe-math-optimizations -mno-unaligned-access
+	CXXFLAGS += -mfpu=neon-fp-armv8 -mfp16-format=ieee -funsafe-math-optimizations -mno-unaligned-access
+endif
+
+#
+# Print build information
+#
+
+$(info I whisper.cpp build info: )
+$(info I UNAME_S:  $(UNAME_S))
+$(info I UNAME_P:  $(UNAME_P))
+$(info I UNAME_M:  $(UNAME_M))
+$(info I CFLAGS:   $(CFLAGS))
+$(info I CXXFLAGS: $(CXXFLAGS))
+$(info I LDFLAGS:  $(LDFLAGS))
+$(info I CC:       $(CCV))
+$(info I CXX:      $(CXXV))
+$(info )
+
+ifdef WHISPER_CUBLAS
+$(info !!!!)
+$(info WHISPER_CUBLAS is deprecated and will be removed in the future. Use WHISPER_CUDA instead.)
+$(info !!!!)
+$(info )
+endif
+
+#
+# Build library
+#
+
+ggml.o: ggml.c ggml.h ggml-cuda.h
+	$(CC)  $(CFLAGS)   -c $< -o $@
+
+ggml-alloc.o: ggml-alloc.c ggml.h ggml-alloc.h
+	$(CC)  $(CFLAGS)   -c $< -o $@
+
+ggml-backend.o: ggml-backend.c ggml.h ggml-backend.h
+	$(CC)  $(CFLAGS)   -c $< -o $@
+
+ggml-quants.o: ggml-quants.c ggml.h ggml-quants.h
+	$(CC)  $(CFLAGS)   -c $< -o $@
+
+WHISPER_OBJ += ggml.o ggml-alloc.o ggml-backend.o ggml-quants.o
+
+whisper.o: whisper.cpp whisper.h whisper-mel.hpp ggml.h ggml-cuda.h
+	$(CXX) $(CXXFLAGS) -c $< -o $@
+
+ifndef WHISPER_COREML
+WHISPER_OBJ += whisper.o
+else
+whisper-encoder.o: coreml/whisper-encoder.mm coreml/whisper-encoder.h
+	$(CXX) -O3 -I . -fobjc-arc -c coreml/whisper-encoder.mm -o whisper-encoder.o
+
+whisper-encoder-impl.o: coreml/whisper-encoder-impl.m coreml/whisper-encoder-impl.h
+	$(CXX) -O3 -I . -fobjc-arc -c coreml/whisper-encoder-impl.m -o whisper-encoder-impl.o
+
+WHISPER_OBJ += whisper.o whisper-encoder.o whisper-encoder-impl.o
+endif
+
+ifdef WHISPER_METAL
+ggml-metal.o: ggml-metal.m ggml-metal.h
+	$(CC) $(CFLAGS) -c $< -o $@
+
+WHISPER_OBJ += ggml-metal.o
+
+ifdef WHISPER_METAL_EMBED_LIBRARY
+CFLAGS += -DGGML_METAL_EMBED_LIBRARY
+
+ggml-metal-embed.o: ggml-metal.metal ggml-common.h
+	@echo "Embedding Metal library"
+	$(eval TEMP_ASSEMBLY=$(shell mktemp))
+	$(eval TEMP_METALLIB=$(shell mktemp))
+	@sed "/^#include \"ggml-common.h\"/{r ggml-common.h"$$'\n'"d;}" ggml-metal.metal > $(TEMP_METALLIB)
+	@echo ".section __DATA, __ggml_metallib" > $(TEMP_ASSEMBLY)
+	@echo ".globl _ggml_metallib_start" >> $(TEMP_ASSEMBLY)
+	@echo "_ggml_metallib_start:" >> $(TEMP_ASSEMBLY)
+	@echo ".incbin \"$(TEMP_METALLIB)\"" >> $(TEMP_ASSEMBLY)
+	@echo ".globl _ggml_metallib_end" >> $(TEMP_ASSEMBLY)
+	@echo "_ggml_metallib_end:" >> $(TEMP_ASSEMBLY)
+	@$(AS) $(TEMP_ASSEMBLY) -o $@
+	@rm -f $(TEMP_ASSEMBLY) $(TEMP_METALLIB)
+
+WHISPER_OBJ += ggml-metal-embed.o
+endif
+endif
+
+libwhisper.a: $(WHISPER_OBJ)
+	$(AR) rcs libwhisper.a $(WHISPER_OBJ)
+
+libwhisper.so: $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) -shared -o libwhisper.so $(WHISPER_OBJ) $(LDFLAGS)
+
+clean:
+	rm -f *.o main stream command talk talk-llama bench quantize server lsp libwhisper.a libwhisper.so
+
+#
+# Examples
+#
+
+CC_SDL=`sdl2-config --cflags --libs`
+
+SRC_COMMON     = examples/common.cpp examples/common-ggml.cpp examples/grammar-parser.cpp
+SRC_COMMON_SDL = examples/common-sdl.cpp
+
+main: examples/main/main.cpp $(SRC_COMMON) $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) examples/main/main.cpp $(SRC_COMMON) $(WHISPER_OBJ) -o main $(LDFLAGS)
+	./main -h
+
+bench: examples/bench/bench.cpp $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) examples/bench/bench.cpp $(WHISPER_OBJ) -o bench $(LDFLAGS)
+
+quantize: examples/quantize/quantize.cpp $(WHISPER_OBJ) $(SRC_COMMON)
+	$(CXX) $(CXXFLAGS) examples/quantize/quantize.cpp $(SRC_COMMON) $(WHISPER_OBJ) -o quantize $(LDFLAGS)
+
+server: examples/server/server.cpp $(SRC_COMMON) $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) examples/server/server.cpp $(SRC_COMMON) $(WHISPER_OBJ) -o server $(LDFLAGS) $(LWINSOCK2)
+
+stream: examples/stream/stream.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) examples/stream/stream.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ) -o stream $(CC_SDL) $(LDFLAGS)
+
+command: examples/command/command.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) examples/command/command.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ) -o command $(CC_SDL) $(LDFLAGS)
+
+lsp: examples/lsp/lsp.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) examples/lsp/lsp.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ) -o lsp $(CC_SDL) $(LDFLAGS)
+
+talk: examples/talk/talk.cpp examples/talk/gpt-2.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) examples/talk/talk.cpp examples/talk/gpt-2.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ) -o talk $(CC_SDL) $(LDFLAGS)
+
+talk-llama: examples/talk-llama/talk-llama.cpp examples/talk-llama/llama.cpp examples/talk-llama/unicode.cpp examples/talk-llama/unicode-data.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ)
+	$(CXX) $(CXXFLAGS) examples/talk-llama/talk-llama.cpp examples/talk-llama/llama.cpp examples/talk-llama/unicode.cpp examples/talk-llama/unicode-data.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) $(WHISPER_OBJ) -o talk-llama $(CC_SDL) $(LDFLAGS)
+
 #
 # Audio samples
 #

-.PHONY: build
-build:
-	cmake -B build $(CMAKE_ARGS)
-	cmake --build build --config Release
-
 # download a few audio samples into folder "./samples":
 .PHONY: samples
 samples:
@ -18,6 +510,17 @@ samples:
 	@wget --quiet --show-progress -O samples/mm1.wav https://cdn.openai.com/whisper/draft-20220913a/micro-machines.wav
 	@wget --quiet --show-progress -O samples/a13.mp3 https://upload.wikimedia.org/wikipedia/commons/transcoded/6/6f/Apollo13-wehaveaproblem.ogg/Apollo13-wehaveaproblem.ogg.mp3
 	@wget --quiet --show-progress -O samples/diffusion2023-07-03.flac https://archive.org/download/diffusion2023-07-03/diffusion2023-07-03.flac
+	@echo "Converting to 16-bit WAV ..."
+	@ffmpeg -loglevel -0 -y -i samples/gb0.ogg -ar 16000 -ac 1 -c:a pcm_s16le samples/gb0.wav
+	@ffmpeg -loglevel -0 -y -i samples/gb1.ogg -ar 16000 -ac 1 -c:a pcm_s16le samples/gb1.wav
+	@ffmpeg -loglevel -0 -y -i samples/hp0.ogg -ar 16000 -ac 1 -c:a pcm_s16le samples/hp0.wav
+	@rm samples/*.ogg
+	@ffmpeg -loglevel -0 -y -i samples/mm1.wav -ar 16000 -ac 1 -c:a pcm_s16le samples/mm0.wav
+	@rm samples/mm1.wav
+	@ffmpeg -loglevel -0 -y -i samples/a13.mp3 -ar 16000 -ac 1 -c:a pcm_s16le -ss 00:00:00 -to 00:00:30 samples/a13.wav
+	@rm samples/a13.mp3
+	@ffmpeg -loglevel -0 -y -i samples/diffusion2023-07-03.flac -ar 16000 -ac 1 -c:a pcm_s16le samples/diffusion2023-07-03.wav
+	@rm samples/diffusion2023-07-03.flac

 #
 # Models
@ -37,22 +540,27 @@ samples:
 .PHONY: large-v1
 .PHONY: large-v2
 .PHONY: large-v3
-.PHONY: large-v3-turbo

-tiny.en tiny base.en base small.en small medium.en medium large-v1 large-v2 large-v3 large-v3-turbo:
+tiny.en tiny base.en base small.en small medium.en medium large-v1 large-v2 large-v3: main
 	bash ./models/download-ggml-model.sh $@
-	cmake -B build $(CMAKE_ARGS)
-	cmake --build build --config Release
 	@echo ""
 	@echo "==============================================="
 	@echo "Running $@ on all samples in ./samples ..."
 	@echo "==============================================="
 	@echo ""
-	@for f in samples/*.{flac,mp3,ogg,wav}; do \
+	@for f in samples/*.wav; do \
 		echo "----------------------------------------------" ; \
 		echo "[+] Running $@ on $$f ... (run 'ffplay $$f' to listen)" ; \
-		echo "----------------------------------------------" ; \
+	    echo "----------------------------------------------" ; \
 		echo "" ; \
-		./build/bin/whisper-cli -m models/ggml-$@.bin -f $$f ; \
+		./main -m models/ggml-$@.bin -f $$f ; \
 		echo "" ; \
 	done
+
+#
+# Tests
+#
+
+.PHONY: tests
+tests:
+	bash ./tests/run-tests.sh $(word 2, $(MAKECMDGOALS))
--- a/Package.swift
+++ b/Package.swift
@ -0,0 +1,61 @@
+// swift-tools-version:5.5
+
+import PackageDescription
+
+let package = Package(
+    name: "whisper",
+    platforms: [
+        .macOS(.v12),
+        .iOS(.v14),
+        .watchOS(.v4),
+        .tvOS(.v14)
+    ],
+    products: [
+        .library(name: "whisper", targets: ["whisper"]),
+    ],
+    targets: [
+        .target(
+            name: "whisper",
+            path: ".",
+            exclude: [
+               "bindings",
+               "cmake",
+               "coreml",
+               "examples",
+               "extra",
+               "models",
+               "samples",
+               "tests",
+               "CMakeLists.txt",
+               "ggml-cuda.cu",
+               "ggml-cuda.h",
+               "Makefile"
+            ],
+            sources: [
+                "ggml.c",
+                "whisper.cpp",
+                "ggml-alloc.c",
+                "ggml-backend.c",
+                "ggml-quants.c",
+                "ggml-metal.m"
+            ],
+            resources: [.process("ggml-metal.metal")],
+            publicHeadersPath: "spm-headers",
+            cSettings: [
+                .unsafeFlags(["-Wno-shorten-64-to-32", "-O3", "-DNDEBUG"]),
+                .define("GGML_USE_ACCELERATE"),
+                .unsafeFlags(["-fno-objc-arc"]),
+                .define("GGML_USE_METAL")
+                // NOTE: NEW_LAPACK will required iOS version 16.4+
+                // We should consider add this in the future when we drop support for iOS 14
+                // (ref: ref: https://developer.apple.com/documentation/accelerate/1513264-cblas_sgemm?language=objc)
+                // .define("ACCELERATE_NEW_LAPACK"),
+                // .define("ACCELERATE_LAPACK_ILP64")
+            ],
+            linkerSettings: [
+                .linkedFramework("Accelerate")
+            ]
+        )
+    ],
+    cxxLanguageStandard: .cxx11
+)
--- a/README.md
+++ b/README.md
@ -2,30 +2,27 @@

 ![whisper.cpp](https://user-images.githubusercontent.com/1991296/235238348-05d0f6a4-da44-4900-a1de-d0707e75b763.jpeg)

-[![Actions Status](https://github.com/ggml-org/whisper.cpp/workflows/CI/badge.svg)](https://github.com/ggml-org/whisper.cpp/actions)
+[![Actions Status](https://github.com/ggerganov/whisper.cpp/workflows/CI/badge.svg)](https://github.com/ggerganov/whisper.cpp/actions)
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](https://opensource.org/licenses/MIT)
 [![Conan Center](https://shields.io/conan/v/whisper-cpp)](https://conan.io/center/whisper-cpp)
 [![npm](https://img.shields.io/npm/v/whisper.cpp.svg)](https://www.npmjs.com/package/whisper.cpp/)

-Stable: [v1.7.5](https://github.com/ggml-org/whisper.cpp/releases/tag/v1.7.5) / [Roadmap](https://github.com/orgs/ggml-org/projects/4/)
+Stable: [v1.6.2](https://github.com/ggerganov/whisper.cpp/releases/tag/v1.6.0) / [Roadmap | F.A.Q.](https://github.com/ggerganov/whisper.cpp/discussions/126)

 High-performance inference of [OpenAI's Whisper](https://github.com/openai/whisper) automatic speech recognition (ASR) model:

 - Plain C/C++ implementation without dependencies
- Apple Silicon first-class citizen - optimized via ARM NEON, Accelerate framework, Metal and [Core ML](#core-ml-support)
+- Apple Silicon first-class citizen - optimized via ARM NEON, Accelerate framework, Metal and [Core ML](https://github.com/ggerganov/whisper.cpp#core-ml-support)
 - AVX intrinsics support for x86 architectures
- [VSX intrinsics support for POWER architectures](#power-vsx-intrinsics)
+- VSX intrinsics support for POWER architectures
 - Mixed F16 / F32 precision
- [Integer quantization support](#quantization)
+- [4-bit and 5-bit integer quantization support](https://github.com/ggerganov/whisper.cpp#quantization)
 - Zero memory allocations at runtime
- [Vulkan support](#vulkan-gpu-support)
 - Support for CPU-only inference
- [Efficient GPU support for NVIDIA](#nvidia-gpu-support)
- [OpenVINO Support](#openvino-support)
- [Ascend NPU Support](#ascend-npu-support)
- [Moore Threads GPU Support](#moore-threads-gpu-support)
- [C-style API](https://github.com/ggml-org/whisper.cpp/blob/master/include/whisper.h)
- [Voice Activity Detection (VAD)](#voice-activity-detection-vad)
+- [Efficient GPU support for NVIDIA](https://github.com/ggerganov/whisper.cpp#nvidia-gpu-support-via-cublas)
+- [Partial OpenCL GPU support via CLBlast](https://github.com/ggerganov/whisper.cpp#opencl-gpu-support-via-clblast)
+- [OpenVINO Support](https://github.com/ggerganov/whisper.cpp#openvino-support)
+- [C-style API](https://github.com/ggerganov/whisper.cpp/blob/master/whisper.h)

 Supported platforms:

@ -33,14 +30,14 @@ Supported platforms:
 - [x] [iOS](examples/whisper.objc)
 - [x] [Android](examples/whisper.android)
 - [x] [Java](bindings/java/README.md)
- [x] Linux / [FreeBSD](https://github.com/ggml-org/whisper.cpp/issues/56#issuecomment-1350920264)
+- [x] Linux / [FreeBSD](https://github.com/ggerganov/whisper.cpp/issues/56#issuecomment-1350920264)
 - [x] [WebAssembly](examples/whisper.wasm)
- [x] Windows ([MSVC](https://github.com/ggml-org/whisper.cpp/blob/master/.github/workflows/build.yml#L117-L144) and [MinGW](https://github.com/ggml-org/whisper.cpp/issues/168))
- [x] [Raspberry Pi](https://github.com/ggml-org/whisper.cpp/discussions/166)
- [x] [Docker](https://github.com/ggml-org/whisper.cpp/pkgs/container/whisper.cpp)
+- [x] Windows ([MSVC](https://github.com/ggerganov/whisper.cpp/blob/master/.github/workflows/build.yml#L117-L144) and [MinGW](https://github.com/ggerganov/whisper.cpp/issues/168)]
+- [x] [Raspberry Pi](https://github.com/ggerganov/whisper.cpp/discussions/166)
+- [x] [docker](https://github.com/ggerganov/whisper.cpp/pkgs/container/whisper.cpp)

-The entire high-level implementation of the model is contained in [whisper.h](include/whisper.h) and [whisper.cpp](src/whisper.cpp).
-The rest of the code is part of the [`ggml`](https://github.com/ggml-org/ggml) machine learning library.
+The entire high-level implementation of the model is contained in [whisper.h](whisper.h) and [whisper.cpp](whisper.cpp).
+The rest of the code is part of the [`ggml`](https://github.com/ggerganov/ggml) machine learning library.

 Having such a lightweight implementation of the model allows to easily integrate it in different platforms and applications.
 As an example, here is a video of running the model on an iPhone 13 device - fully offline, on-device: [whisper.objc](examples/whisper.objc)
@ -53,48 +50,162 @@ https://user-images.githubusercontent.com/1991296/204038393-2f846eae-c255-4099-a

 On Apple Silicon, the inference runs fully on the GPU via Metal:

-https://github.com/ggml-org/whisper.cpp/assets/1991296/c82e8f86-60dc-49f2-b048-d2fdbd6b5225
+https://github.com/ggerganov/whisper.cpp/assets/1991296/c82e8f86-60dc-49f2-b048-d2fdbd6b5225
+
+Or you can even run it straight in the browser: [talk.wasm](examples/talk.wasm)
+
+## Implementation details
+
+- The core tensor operations are implemented in C ([ggml.h](ggml.h) / [ggml.c](ggml.c))
+- The transformer model and the high-level C-style API are implemented in C++ ([whisper.h](whisper.h) / [whisper.cpp](whisper.cpp))
+- Sample usage is demonstrated in [main.cpp](examples/main)
+- Sample real-time audio transcription from the microphone is demonstrated in [stream.cpp](examples/stream)
+- Various other examples are available in the [examples](examples) folder
+
+The tensor operators are optimized heavily for Apple silicon CPUs. Depending on the computation size, Arm Neon SIMD intrinsics or CBLAS Accelerate framework routines are used. The latter are especially effective for bigger sizes since the Accelerate framework utilizes the special-purpose AMX coprocessor available in modern Apple products.

 ## Quick start

 First clone the repository:

 ```bash
-git clone https://github.com/ggml-org/whisper.cpp.git
-```
-
-Navigate into the directory:
-
-```
-cd whisper.cpp
+git clone https://github.com/ggerganov/whisper.cpp.git
 ```

 Then, download one of the Whisper [models](models/README.md) converted in [`ggml` format](#ggml-format). For example:

 ```bash
-sh ./models/download-ggml-model.sh base.en
+bash ./models/download-ggml-model.sh base.en
 ```

-Now build the [whisper-cli](examples/cli) example and transcribe an audio file like this:
+Now build the [main](examples/main) example and transcribe an audio file like this:

 ```bash
-# build the project
-cmake -B build
-cmake --build build --config Release
+# build the main example
+make

 # transcribe an audio file
-./build/bin/whisper-cli -f samples/jfk.wav
+./main -f samples/jfk.wav
 ```

 ---

-For a quick demo, simply run `make base.en`.
+For a quick demo, simply run `make base.en`:
+
+```text
+$ make base.en
+
+cc  -I.              -O3 -std=c11   -pthread -DGGML_USE_ACCELERATE   -c ggml.c -o ggml.o
+c++ -I. -I./examples -O3 -std=c++11 -pthread -c whisper.cpp -o whisper.o
+c++ -I. -I./examples -O3 -std=c++11 -pthread examples/main/main.cpp whisper.o ggml.o -o main  -framework Accelerate
+./main -h
+
+usage: ./main [options] file0.wav file1.wav ...
+
+options:
+  -h,        --help              [default] show this help message and exit
+  -t N,      --threads N         [4      ] number of threads to use during computation
+  -p N,      --processors N      [1      ] number of processors to use during computation
+  -ot N,     --offset-t N        [0      ] time offset in milliseconds
+  -on N,     --offset-n N        [0      ] segment index offset
+  -d  N,     --duration N        [0      ] duration of audio to process in milliseconds
+  -mc N,     --max-context N     [-1     ] maximum number of text context tokens to store
+  -ml N,     --max-len N         [0      ] maximum segment length in characters
+  -sow,      --split-on-word     [false  ] split on word rather than on token
+  -bo N,     --best-of N         [5      ] number of best candidates to keep
+  -bs N,     --beam-size N       [5      ] beam size for beam search
+  -wt N,     --word-thold N      [0.01   ] word timestamp probability threshold
+  -et N,     --entropy-thold N   [2.40   ] entropy threshold for decoder fail
+  -lpt N,    --logprob-thold N   [-1.00  ] log probability threshold for decoder fail
+  -debug,    --debug-mode        [false  ] enable debug mode (eg. dump log_mel)
+  -tr,       --translate         [false  ] translate from source language to english
+  -di,       --diarize           [false  ] stereo audio diarization
+  -tdrz,     --tinydiarize       [false  ] enable tinydiarize (requires a tdrz model)
+  -nf,       --no-fallback       [false  ] do not use temperature fallback while decoding
+  -otxt,     --output-txt        [false  ] output result in a text file
+  -ovtt,     --output-vtt        [false  ] output result in a vtt file
+  -osrt,     --output-srt        [false  ] output result in a srt file
+  -olrc,     --output-lrc        [false  ] output result in a lrc file
+  -owts,     --output-words      [false  ] output script for generating karaoke video
+  -fp,       --font-path         [/System/Library/Fonts/Supplemental/Courier New Bold.ttf] path to a monospace font for karaoke video
+  -ocsv,     --output-csv        [false  ] output result in a CSV file
+  -oj,       --output-json       [false  ] output result in a JSON file
+  -ojf,      --output-json-full  [false  ] include more information in the JSON file
+  -of FNAME, --output-file FNAME [       ] output file path (without file extension)
+  -ps,       --print-special     [false  ] print special tokens
+  -pc,       --print-colors      [false  ] print colors
+  -pp,       --print-progress    [false  ] print progress
+  -nt,       --no-timestamps     [false  ] do not print timestamps
+  -l LANG,   --language LANG     [en     ] spoken language ('auto' for auto-detect)
+  -dl,       --detect-language   [false  ] exit after automatically detecting language
+             --prompt PROMPT     [       ] initial prompt
+  -m FNAME,  --model FNAME       [models/ggml-base.en.bin] model path
+  -f FNAME,  --file FNAME        [       ] input WAV file path
+  -oved D,   --ov-e-device DNAME [CPU    ] the OpenVINO device used for encode inference
+  -ls,       --log-score         [false  ] log best decoder scores of tokens
+  -ng,       --no-gpu            [false  ] disable GPU
+
+
+bash ./models/download-ggml-model.sh base.en
+Downloading ggml model base.en ...
+ggml-base.en.bin               100%[========================>] 141.11M  6.34MB/s    in 24s
+Done! Model 'base.en' saved in 'models/ggml-base.en.bin'
+You can now use it like this:
+
+  $ ./main -m models/ggml-base.en.bin -f samples/jfk.wav
+
+
+===============================================
+Running base.en on all samples in ./samples ...
+===============================================
+
+----------------------------------------------
+[+] Running base.en on samples/jfk.wav ... (run 'ffplay samples/jfk.wav' to listen)
+----------------------------------------------
+
+whisper_init_from_file: loading model from 'models/ggml-base.en.bin'
+whisper_model_load: loading model
+whisper_model_load: n_vocab       = 51864
+whisper_model_load: n_audio_ctx   = 1500
+whisper_model_load: n_audio_state = 512
+whisper_model_load: n_audio_head  = 8
+whisper_model_load: n_audio_layer = 6
+whisper_model_load: n_text_ctx    = 448
+whisper_model_load: n_text_state  = 512
+whisper_model_load: n_text_head   = 8
+whisper_model_load: n_text_layer  = 6
+whisper_model_load: n_mels        = 80
+whisper_model_load: f16           = 1
+whisper_model_load: type          = 2
+whisper_model_load: mem required  =  215.00 MB (+    6.00 MB per decoder)
+whisper_model_load: kv self size  =    5.25 MB
+whisper_model_load: kv cross size =   17.58 MB
+whisper_model_load: adding 1607 extra tokens
+whisper_model_load: model ctx     =  140.60 MB
+whisper_model_load: model size    =  140.54 MB
+
+system_info: n_threads = 4 / 10 | AVX = 0 | AVX2 = 0 | AVX512 = 0 | FMA = 0 | NEON = 1 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 0 | VSX = 0 |
+
+main: processing 'samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, 1 processors, lang = en, task = transcribe, timestamps = 1 ...
+
+
+[00:00:00.000 --> 00:00:11.000]   And so my fellow Americans, ask not what your country can do for you, ask what you can do for your country.
+
+
+whisper_print_timings:     fallbacks =   0 p /   0 h
+whisper_print_timings:     load time =   113.81 ms
+whisper_print_timings:      mel time =    15.40 ms
+whisper_print_timings:   sample time =    11.58 ms /    27 runs (    0.43 ms per run)
+whisper_print_timings:   encode time =   266.60 ms /     1 runs (  266.60 ms per run)
+whisper_print_timings:   decode time =    66.11 ms /    27 runs (    2.45 ms per run)
+whisper_print_timings:    total time =   476.31 ms
+```

 The command downloads the `base.en` model converted to custom `ggml` format and runs the inference on all `.wav` samples in the folder `samples`.

-For detailed usage instructions, run: `./build/bin/whisper-cli -h`
+For detailed usage instructions, run: `./main -h`

-Note that the [whisper-cli](examples/cli) example currently runs only with 16-bit WAV files, so make sure to convert your input before running the tool.
+Note that the [main](examples/main) example currently runs only with 16-bit WAV files, so make sure to convert your input before running the tool.
 For example, you can use `ffmpeg` like this:

 ```bash
@ -106,7 +217,7 @@ ffmpeg -i input.mp3 -ar 16000 -ac 1 -c:a pcm_s16le output.wav
 If you want some extra audio samples to play with, simply run:

 ```
-make -j samples
+make samples
 ```

 This will download a few more audio files from Wikipedia and convert them to 16-bit WAV format via `ffmpeg`.
@ -114,18 +225,17 @@ This will download a few more audio files from Wikipedia and convert them to 16-
 You can download and run the other models as follows:

 ```
-make -j tiny.en
-make -j tiny
-make -j base.en
-make -j base
-make -j small.en
-make -j small
-make -j medium.en
-make -j medium
-make -j large-v1
-make -j large-v2
-make -j large-v3
-make -j large-v3-turbo
+make tiny.en
+make tiny
+make base.en
+make base
+make small.en
+make small
+make medium.en
+make medium
+make large-v1
+make large-v2
+make large-v3
 ```

 ## Memory usage
@ -138,21 +248,6 @@ make -j large-v3-turbo
 | medium | 1.5 GiB | ~2.1 GB |
 | large  | 2.9 GiB | ~3.9 GB |

-## POWER VSX Intrinsics
-
-`whisper.cpp` supports POWER architectures and includes code which
-significantly speeds operation on Linux running on POWER9/10, making it
-capable of faster-than-realtime transcription on underclocked Raptor
-Talos II. Ensure you have a BLAS package installed, and replace the
-standard cmake setup with:
-
-```bash
-# build with GGML_BLAS defined
-cmake -B build -DGGML_BLAS=1
-cmake --build build --config Release
-./build/bin/whisper-cli [ .. etc .. ]
-```
-
 ## Quantization

 `whisper.cpp` supports integer quantization of the Whisper `ggml` models.
@ -162,12 +257,11 @@ Here are the steps for creating and using a quantized model:

 ```bash
 # quantize a model with Q5_0 method
-cmake -B build
-cmake --build build --config Release
-./build/bin/quantize models/ggml-base.en.bin models/ggml-base.en-q5_0.bin q5_0
+make quantize
+./quantize models/ggml-base.en.bin models/ggml-base.en-q5_0.bin q5_0

 # run the examples as usual, specifying the quantized model file
-./build/bin/whisper-cli -m models/ggml-base.en-q5_0.bin ./samples/gb0.wav
+./main -m models/ggml-base.en-q5_0.bin ./samples/gb0.wav
 ```

 ## Core ML support
@ -184,11 +278,11 @@ speed-up - more than x3 faster compared with CPU-only execution. Here are the in
  ```

  - To ensure `coremltools` operates correctly, please confirm that [Xcode](https://developer.apple.com/xcode/) is installed and execute `xcode-select --install` to install the command-line tools.
-  - Python 3.11 is recommended.
+  - Python 3.10 is recommended.
  - MacOS Sonoma (version 14) or newer is recommended, as older versions of MacOS might experience issues with transcription hallucination.
  - [OPTIONAL] It is recommended to utilize a Python version management system, such as [Miniconda](https://docs.conda.io/en/latest/miniconda.html) for this step:
-    - To create an environment, use: `conda create -n py311-whisper python=3.11 -y`
-    - To activate the environment, use: `conda activate py311-whisper`
+    - To create an environment, use: `conda create -n py310-whisper python=3.10 -y`
+    - To activate the environment, use: `conda activate py310-whisper`

 - Generate a Core ML model. For example, to generate a `base.en` model, use:

@ -201,6 +295,10 @@ speed-up - more than x3 faster compared with CPU-only execution. Here are the in
 - Build `whisper.cpp` with Core ML support:

  ```bash
+  # using Makefile
+  make clean
+  WHISPER_COREML=1 make -j
+
  # using CMake
  cmake -B build -DWHISPER_COREML=1
  cmake --build build -j --config Release
@ -209,7 +307,7 @@ speed-up - more than x3 faster compared with CPU-only execution. Here are the in
 - Run the examples as usual. For example:

  ```text
-  $ ./build/bin/whisper-cli -m models/ggml-base.en.bin -f samples/jfk.wav
+  $ ./main -m models/ggml-base.en.bin -f samples/jfk.wav

  ...

@ -225,7 +323,7 @@ speed-up - more than x3 faster compared with CPU-only execution. Here are the in
  The first run on a device is slow, since the ANE service compiles the Core ML model to some device-specific format.
  Next runs are faster.

-For more information about the Core ML implementation please refer to PR [#566](https://github.com/ggml-org/whisper.cpp/pull/566).
+For more information about the Core ML implementation please refer to PR [#566](https://github.com/ggerganov/whisper.cpp/pull/566).

 ## OpenVINO support

@ -267,7 +365,7 @@ This can result in significant speedup in encoder performance. Here are the inst

 - Build `whisper.cpp` with OpenVINO support:

-  Download OpenVINO package from [release page](https://github.com/openvinotoolkit/openvino/releases). The recommended version to use is [2024.6.0](https://github.com/openvinotoolkit/openvino/releases/tag/2024.6.0). Ready to use Binaries of the required libraries can be found in the [OpenVino Archives](https://storage.openvinotoolkit.org/repositories/openvino/packages/2024.6/)
+  Download OpenVINO package from [release page](https://github.com/openvinotoolkit/openvino/releases). The recommended version to use is [2023.0.0](https://github.com/openvinotoolkit/openvino/releases/tag/2023.0.0).

  After downloading & extracting package onto your development system, set up required environment by sourcing setupvars script. For example:

@ -293,7 +391,7 @@ This can result in significant speedup in encoder performance. Here are the inst
 - Run the examples as usual. For example:

  ```text
-  $ ./build/bin/whisper-cli -m models/ggml-base.en.bin -f samples/jfk.wav
+  $ ./main -m models/ggml-base.en.bin -f samples/jfk.wav

  ...

@ -310,7 +408,7 @@ This can result in significant speedup in encoder performance. Here are the inst
  The first time run on an OpenVINO device is slow, since the OpenVINO framework will compile the IR (Intermediate Representation) model to a device-specific 'blob'. This device-specific blob will get
  cached for the next run.

-For more information about the OpenVINO implementation please refer to PR [#1037](https://github.com/ggml-org/whisper.cpp/pull/1037).
+For more information about the Core ML implementation please refer to PR [#1037](https://github.com/ggerganov/whisper.cpp/pull/1037).

 ## NVIDIA GPU support

@ -320,25 +418,31 @@ First, make sure you have installed `cuda`: https://developer.nvidia.com/cuda-do
 Now build `whisper.cpp` with CUDA support:

 ```
-cmake -B build -DGGML_CUDA=1
+make clean
+WHISPER_CUDA=1 make -j
+```
+
+## OpenCL GPU support via CLBlast
+
+For cards and integrated GPUs that support OpenCL, the Encoder processing can be largely offloaded to the GPU through CLBlast. This is especially useful for users with AMD APUs or low end devices for up to ~2x speedup.
+
+First, make sure you have installed `CLBlast` for your OS or Distribution: https://github.com/CNugteren/CLBlast
+
+Now build `whisper.cpp` with CLBlast support:
+
+```
+Makefile:
+cd whisper.cpp
+make clean
+WHISPER_CLBLAST=1 make -j
+
+CMake:
+cd whisper.cpp
+cmake -B build -DWHISPER_CLBLAST=ON
 cmake --build build -j --config Release
 ```

-or for newer NVIDIA GPU's (RTX 5000 series):
-```
-cmake -B build -DGGML_CUDA=1 -DCMAKE_CUDA_ARCHITECTURES="86"
-cmake --build build -j --config Release
-```
-
-## Vulkan GPU support
-Cross-vendor solution which allows you to accelerate workload on your GPU.
-First, make sure your graphics card driver provides support for Vulkan API.
-
-Now build `whisper.cpp` with Vulkan support:
-```
-cmake -B build -DGGML_VULKAN=1
-cmake --build build -j --config Release
-```
+Run all the examples as usual.

 ## BLAS CPU support via OpenBLAS

@ -348,89 +452,23 @@ First, make sure you have installed `openblas`: https://www.openblas.net/
 Now build `whisper.cpp` with OpenBLAS support:

 ```
-cmake -B build -DGGML_BLAS=1
-cmake --build build -j --config Release
+make clean
+WHISPER_OPENBLAS=1 make -j
 ```

-## Ascend NPU support
+## BLAS CPU support via Intel MKL

-Ascend NPU provides inference acceleration via [`CANN`](https://www.hiascend.com/en/software/cann) and AI cores.
+Encoder processing can be accelerated on the CPU via the BLAS compatible interface of Intel's Math Kernel Library.
+First, make sure you have installed Intel's MKL runtime and development packages: https://www.intel.com/content/www/us/en/developer/tools/oneapi/onemkl-download.html

-First, check if your Ascend NPU device is supported:
-
-**Verified devices**
-| Ascend NPU                    | Status  |
-|:-----------------------------:|:-------:|
-| Atlas 300T A2                 | Support |
-
-Then, make sure you have installed [`CANN toolkit`](https://www.hiascend.com/en/software/cann/community) . The lasted version of CANN is recommanded.
-
-Now build `whisper.cpp` with CANN support:
+Now build `whisper.cpp` with Intel MKL BLAS support:

 ```
-cmake -B build -DGGML_CANN=1
-cmake --build build -j --config Release
-```
-
-Run the inference examples as usual, for example:
-
-```
-./build/bin/whisper-cli -f samples/jfk.wav -m models/ggml-base.en.bin -t 8
-```
-
-*Notes:*
-
- If you have trouble with Ascend NPU device, please create a issue with **[CANN]** prefix/tag.
- If you run successfully with your Ascend NPU device, please help update the table `Verified devices`.
-
-## Moore Threads GPU support
-
-With Moore Threads cards the processing of the models is done efficiently on the GPU via muBLAS and custom MUSA kernels.
-First, make sure you have installed `MUSA SDK rc4.0.1`: https://developer.mthreads.com/sdk/download/musa?equipment=&os=&driverVersion=&version=4.0.1
-
-Now build `whisper.cpp` with MUSA support:
-
-```
-cmake -B build -DGGML_MUSA=1
-cmake --build build -j --config Release
-```
-
-or specify the architecture for your Moore Threads GPU. For example, if you have a MTT S80 GPU, you can specify the architecture as follows:
-
-```
-cmake -B build -DGGML_MUSA=1 -DMUSA_ARCHITECTURES="21"
-cmake --build build -j --config Release
-```
-
-## FFmpeg support (Linux only)
-
-If you want to support more audio formats (such as Opus and AAC), you can turn on the `WHISPER_FFMPEG` build flag to enable FFmpeg integration.
-
-First, you need to install required libraries:
-
-```bash
-# Debian/Ubuntu
-sudo apt install libavcodec-dev libavformat-dev libavutil-dev
-
-# RHEL/Fedora
-sudo dnf install libavcodec-free-devel libavformat-free-devel libavutil-free-devel
-```
-
-Then you can build the project as follows:
-
-```bash
-cmake -B build -D WHISPER_FFMPEG=yes
-cmake --build build
-```
-
-Run the following example to confirm it's working:
-
-```bash
-# Convert an audio file to Opus format
-ffmpeg -i samples/jfk.wav jfk.opus
-
-# Transcribe the audio file
-./build/bin/whisper-cli --model models/ggml-base.en.bin --file jfk.opus
+source /opt/intel/oneapi/setvars.sh
+mkdir build
+cd build
+cmake -DWHISPER_MKL=ON ..
+WHISPER_MKL=1 make -j
 ```

 ## Docker
@ -444,9 +482,8 @@ ffmpeg -i samples/jfk.wav jfk.opus

 We have two Docker images available for this project:

-1. `ghcr.io/ggml-org/whisper.cpp:main`: This image includes the main executable file as well as `curl` and `ffmpeg`. (platforms: `linux/amd64`, `linux/arm64`)
-2. `ghcr.io/ggml-org/whisper.cpp:main-cuda`: Same as `main` but compiled with CUDA support. (platforms: `linux/amd64`)
-3. `ghcr.io/ggml-org/whisper.cpp:main-musa`: Same as `main` but compiled with MUSA support. (platforms: `linux/amd64`)
+1. `ghcr.io/ggerganov/whisper.cpp:main`: This image includes the main executable file as well as `curl` and `ffmpeg`. (platforms: `linux/amd64`, `linux/arm64`)
+2. `ghcr.io/ggerganov/whisper.cpp:main-cuda`: Same as `main` but compiled with CUDA support. (platforms: `linux/amd64`)

 ### Usage

@ -459,11 +496,11 @@ docker run -it --rm \
 docker run -it --rm \
  -v path/to/models:/models \
  -v path/to/audios:/audios \
-  whisper.cpp:main "whisper-cli -m /models/ggml-base.bin -f /audios/jfk.wav"
+  whisper.cpp:main "./main -m /models/ggml-base.bin -f /audios/jfk.wav"
 # transcribe an audio file in samples folder
 docker run -it --rm \
  -v path/to/models:/models \
-  whisper.cpp:main "whisper-cli -m /models/ggml-base.bin -f ./samples/jfk.wav"
+  whisper.cpp:main "./main -m /models/ggml-base.bin -f ./samples/jfk.wav"
 ```

 ## Installing with Conan
@ -480,17 +517,98 @@ For detailed instructions on how to use Conan, please refer to the [Conan docume

 - Inference only

+## Another example
+
+Here is another example of transcribing a [3:24 min speech](https://upload.wikimedia.org/wikipedia/commons/1/1f/George_W_Bush_Columbia_FINAL.ogg)
+in about half a minute on a MacBook M1 Pro, using `medium.en` model:
+
+<details>
+  <summary>Expand to see the result</summary>
+
+```text
+$ ./main -m models/ggml-medium.en.bin -f samples/gb1.wav -t 8
+
+whisper_init_from_file: loading model from 'models/ggml-medium.en.bin'
+whisper_model_load: loading model
+whisper_model_load: n_vocab       = 51864
+whisper_model_load: n_audio_ctx   = 1500
+whisper_model_load: n_audio_state = 1024
+whisper_model_load: n_audio_head  = 16
+whisper_model_load: n_audio_layer = 24
+whisper_model_load: n_text_ctx    = 448
+whisper_model_load: n_text_state  = 1024
+whisper_model_load: n_text_head   = 16
+whisper_model_load: n_text_layer  = 24
+whisper_model_load: n_mels        = 80
+whisper_model_load: f16           = 1
+whisper_model_load: type          = 4
+whisper_model_load: mem required  = 1720.00 MB (+   43.00 MB per decoder)
+whisper_model_load: kv self size  =   42.00 MB
+whisper_model_load: kv cross size =  140.62 MB
+whisper_model_load: adding 1607 extra tokens
+whisper_model_load: model ctx     = 1462.35 MB
+whisper_model_load: model size    = 1462.12 MB
+
+system_info: n_threads = 8 / 10 | AVX = 0 | AVX2 = 0 | AVX512 = 0 | FMA = 0 | NEON = 1 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 0 | VSX = 0 |
+
+main: processing 'samples/gb1.wav' (3179750 samples, 198.7 sec), 8 threads, 1 processors, lang = en, task = transcribe, timestamps = 1 ...
+
+
+[00:00:00.000 --> 00:00:08.000]   My fellow Americans, this day has brought terrible news and great sadness to our country.
+[00:00:08.000 --> 00:00:17.000]   At nine o'clock this morning, Mission Control in Houston lost contact with our Space Shuttle Columbia.
+[00:00:17.000 --> 00:00:23.000]   A short time later, debris was seen falling from the skies above Texas.
+[00:00:23.000 --> 00:00:29.000]   The Columbia's lost. There are no survivors.
+[00:00:29.000 --> 00:00:32.000]   On board was a crew of seven.
+[00:00:32.000 --> 00:00:39.000]   Colonel Rick Husband, Lieutenant Colonel Michael Anderson, Commander Laurel Clark,
+[00:00:39.000 --> 00:00:48.000]   Captain David Brown, Commander William McCool, Dr. Kultna Shavla, and Ilan Ramon,
+[00:00:48.000 --> 00:00:52.000]   a colonel in the Israeli Air Force.
+[00:00:52.000 --> 00:00:58.000]   These men and women assumed great risk in the service to all humanity.
+[00:00:58.000 --> 00:01:03.000]   In an age when space flight has come to seem almost routine,
+[00:01:03.000 --> 00:01:07.000]   it is easy to overlook the dangers of travel by rocket
+[00:01:07.000 --> 00:01:12.000]   and the difficulties of navigating the fierce outer atmosphere of the Earth.
+[00:01:12.000 --> 00:01:18.000]   These astronauts knew the dangers, and they faced them willingly,
+[00:01:18.000 --> 00:01:23.000]   knowing they had a high and noble purpose in life.
+[00:01:23.000 --> 00:01:31.000]   Because of their courage and daring and idealism, we will miss them all the more.
+[00:01:31.000 --> 00:01:36.000]   All Americans today are thinking as well of the families of these men and women
+[00:01:36.000 --> 00:01:40.000]   who have been given this sudden shock and grief.
+[00:01:40.000 --> 00:01:45.000]   You're not alone. Our entire nation grieves with you,
+[00:01:45.000 --> 00:01:52.000]   and those you love will always have the respect and gratitude of this country.
+[00:01:52.000 --> 00:01:56.000]   The cause in which they died will continue.
+[00:01:56.000 --> 00:02:04.000]   Mankind is led into the darkness beyond our world by the inspiration of discovery
+[00:02:04.000 --> 00:02:11.000]   and the longing to understand. Our journey into space will go on.
+[00:02:11.000 --> 00:02:16.000]   In the skies today, we saw destruction and tragedy.
+[00:02:16.000 --> 00:02:22.000]   Yet farther than we can see, there is comfort and hope.
+[00:02:22.000 --> 00:02:29.000]   In the words of the prophet Isaiah, "Lift your eyes and look to the heavens
+[00:02:29.000 --> 00:02:35.000]   who created all these. He who brings out the starry hosts one by one
+[00:02:35.000 --> 00:02:39.000]   and calls them each by name."
+[00:02:39.000 --> 00:02:46.000]   Because of His great power and mighty strength, not one of them is missing.
+[00:02:46.000 --> 00:02:55.000]   The same Creator who names the stars also knows the names of the seven souls we mourn today.
+[00:02:55.000 --> 00:03:01.000]   The crew of the shuttle Columbia did not return safely to earth,
+[00:03:01.000 --> 00:03:05.000]   yet we can pray that all are safely home.
+[00:03:05.000 --> 00:03:13.000]   May God bless the grieving families, and may God continue to bless America.
+[00:03:13.000 --> 00:03:19.000]   [Silence]
+
+
+whisper_print_timings:     fallbacks =   1 p /   0 h
+whisper_print_timings:     load time =   569.03 ms
+whisper_print_timings:      mel time =   146.85 ms
+whisper_print_timings:   sample time =   238.66 ms /   553 runs (    0.43 ms per run)
+whisper_print_timings:   encode time = 18665.10 ms /     9 runs ( 2073.90 ms per run)
+whisper_print_timings:   decode time = 13090.93 ms /   549 runs (   23.85 ms per run)
+whisper_print_timings:    total time = 32733.52 ms
+```
+
+</details>
+
 ## Real-time audio input example

 This is a naive example of performing real-time inference on audio from your microphone.
 The [stream](examples/stream) tool samples the audio every half a second and runs the transcription continuously.
-More info is available in [issue #10](https://github.com/ggml-org/whisper.cpp/issues/10).
-You will need to have [sdl2](https://wiki.libsdl.org/SDL2/Installation) installed for it to work properly.
+More info is available in [issue #10](https://github.com/ggerganov/whisper.cpp/issues/10).

 ```bash
-cmake -B build -DWHISPER_SDL2=ON
-cmake --build build --config Release
-./build/bin/whisper-stream -m ./models/ggml-base.en.bin -t 8 --step 500 --length 5000
+make stream
+./stream -m ./models/ggml-base.en.bin -t 8 --step 500 --length 5000
 ```

 https://user-images.githubusercontent.com/1991296/194935793-76afede7-cfa8-48d8-a80f-28ba83be7d09.mp4
@ -501,7 +619,7 @@ Adding the `--print-colors` argument will print the transcribed text using an ex
 to highlight words with high or low confidence:

 ```bash
-./build/bin/whisper-cli -m models/ggml-base.en.bin -f samples/gb0.wav --print-colors
+./main -m models/ggml-base.en.bin -f samples/gb0.wav --print-colors
 ```

 <img width="965" alt="image" src="https://user-images.githubusercontent.com/1991296/197356445-311c8643-9397-4e5e-b46e-0b4b4daa2530.png">
@ -511,7 +629,7 @@ to highlight words with high or low confidence:
 For example, to limit the line length to a maximum of 16 characters, simply add `-ml 16`:

 ```text
-$ ./build/bin/whisper-cli -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -ml 16
+$ ./main -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -ml 16

 whisper_model_load: loading model from './models/ggml-base.en.bin'
 ...
@ -535,7 +653,7 @@ main: processing './samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, 1 pr
 The `--max-len` argument can be used to obtain word-level timestamps. Simply use `-ml 1`:

 ```text
-$ ./build/bin/whisper-cli -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -ml 1
+$ ./main -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -ml 1

 whisper_model_load: loading model from './models/ggml-base.en.bin'
 ...
@ -573,7 +691,7 @@ main: processing './samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, 1 pr

 ## Speaker segmentation via tinydiarize (experimental)

-More information about this approach is available here: https://github.com/ggml-org/whisper.cpp/pull/1058
+More information about this approach is available here: https://github.com/ggerganov/whisper.cpp/pull/1058

 Sample usage:

@ -582,7 +700,7 @@ Sample usage:
 ./models/download-ggml-model.sh small.en-tdrz

 # run as usual, adding the "-tdrz" command-line argument
-./build/bin/whisper-cli -f ./samples/a13.wav -m ./models/ggml-small.en-tdrz.bin -tdrz
+./main -f ./samples/a13.wav -m ./models/ggml-small.en-tdrz.bin -tdrz
 ...
 main: processing './samples/a13.wav' (480000 samples, 30.0 sec), 4 threads, 1 processors, lang = en, task = transcribe, tdrz = 1, timestamps = 1 ...
 ...
@ -599,14 +717,14 @@ main: processing './samples/a13.wav' (480000 samples, 30.0 sec), 4 threads, 1 pr

 ## Karaoke-style movie generation (experimental)

-The [whisper-cli](examples/cli) example provides support for output of karaoke-style movies, where the
-currently pronounced word is highlighted. Use the `-owts` argument and run the generated bash script.
+The [main](examples/main) example provides support for output of karaoke-style movies, where the
+currently pronounced word is highlighted. Use the `-wts` argument and run the generated bash script.
 This requires to have `ffmpeg` installed.

 Here are a few _"typical"_ examples:

 ```bash
-./build/bin/whisper-cli -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -owts
+./main -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -owts
 source ./samples/jfk.wav.wts
 ffplay ./samples/jfk.wav.mp4
 ```
@ -616,7 +734,7 @@ https://user-images.githubusercontent.com/1991296/199337465-dbee4b5e-9aeb-48a3-b
 ---

 ```bash
-./build/bin/whisper-cli -m ./models/ggml-base.en.bin -f ./samples/mm0.wav -owts
+./main -m ./models/ggml-base.en.bin -f ./samples/mm0.wav -owts
 source ./samples/mm0.wav.wts
 ffplay ./samples/mm0.wav.mp4
 ```
@ -626,7 +744,7 @@ https://user-images.githubusercontent.com/1991296/199337504-cc8fd233-0cb7-4920-9
 ---

 ```bash
-./build/bin/whisper-cli -m ./models/ggml-base.en.bin -f ./samples/gb0.wav -owts
+./main -m ./models/ggml-base.en.bin -f ./samples/gb0.wav -owts
 source ./samples/gb0.wav.wts
 ffplay ./samples/gb0.wav.mp4
 ```
@ -637,7 +755,7 @@ https://user-images.githubusercontent.com/1991296/199337538-b7b0c7a3-2753-4a88-a

 ## Video comparison of different models

-Use the [scripts/bench-wts.sh](https://github.com/ggml-org/whisper.cpp/blob/master/scripts/bench-wts.sh) script to generate a video in the following format:
+Use the [scripts/bench-wts.sh](https://github.com/ggerganov/whisper.cpp/blob/master/scripts/bench-wts.sh) script to generate a video in the following format:

 ```bash
 ./scripts/bench-wts.sh samples/jfk.wav
@ -651,12 +769,12 @@ https://user-images.githubusercontent.com/1991296/223206245-2d36d903-cf8e-4f09-8
 ## Benchmarks

 In order to have an objective comparison of the performance of the inference across different system configurations,
-use the [whisper-bench](examples/bench) tool. The tool simply runs the Encoder part of the model and prints how much time it
+use the [bench](examples/bench) tool. The tool simply runs the Encoder part of the model and prints how much time it
 took to execute it. The results are summarized in the following Github issue:

-[Benchmark results](https://github.com/ggml-org/whisper.cpp/issues/89)
+[Benchmark results](https://github.com/ggerganov/whisper.cpp/issues/89)

-Additionally a script to run whisper.cpp with different models and audio files is provided [bench.py](scripts/bench.py).
+Additionally a script to run whisper.cpp with different models and audio files is provided [bench.py](bench.py).

 You can run it with the following command, by default it will run against any standard model in the models folder.

@ -681,141 +799,31 @@ You can download the converted models using the [models/download-ggml-model.sh](
 or manually from here:

 - https://huggingface.co/ggerganov/whisper.cpp
+- https://ggml.ggerganov.com

 For more details, see the conversion script [models/convert-pt-to-ggml.py](models/convert-pt-to-ggml.py) or [models/README.md](models/README.md).

-## [Bindings](https://github.com/ggml-org/whisper.cpp/discussions/categories/bindings)
+## [Bindings](https://github.com/ggerganov/whisper.cpp/discussions/categories/bindings)

- [x] Rust: [tazz4843/whisper-rs](https://github.com/tazz4843/whisper-rs) | [#310](https://github.com/ggml-org/whisper.cpp/discussions/310)
- [x] JavaScript: [bindings/javascript](bindings/javascript) | [#309](https://github.com/ggml-org/whisper.cpp/discussions/309)
+- [x] Rust: [tazz4843/whisper-rs](https://github.com/tazz4843/whisper-rs) | [#310](https://github.com/ggerganov/whisper.cpp/discussions/310)
+- [x] JavaScript: [bindings/javascript](bindings/javascript) | [#309](https://github.com/ggerganov/whisper.cpp/discussions/309)
  - React Native (iOS / Android): [whisper.rn](https://github.com/mybigday/whisper.rn)
- [x] Go: [bindings/go](bindings/go) | [#312](https://github.com/ggml-org/whisper.cpp/discussions/312)
+- [x] Go: [bindings/go](bindings/go) | [#312](https://github.com/ggerganov/whisper.cpp/discussions/312)
 - [x] Java:
  - [GiviMAD/whisper-jni](https://github.com/GiviMAD/whisper-jni)
- [x] Ruby: [bindings/ruby](bindings/ruby) | [#507](https://github.com/ggml-org/whisper.cpp/discussions/507)
- [x] Objective-C / Swift: [ggml-org/whisper.spm](https://github.com/ggml-org/whisper.spm) | [#313](https://github.com/ggml-org/whisper.cpp/discussions/313)
+- [x] Ruby: [bindings/ruby](bindings/ruby) | [#507](https://github.com/ggerganov/whisper.cpp/discussions/507)
+- [x] Objective-C / Swift: [ggerganov/whisper.spm](https://github.com/ggerganov/whisper.spm) | [#313](https://github.com/ggerganov/whisper.cpp/discussions/313)
  - [exPHAT/SwiftWhisper](https://github.com/exPHAT/SwiftWhisper)
- [x] .NET: | [#422](https://github.com/ggml-org/whisper.cpp/discussions/422)
+- [x] .NET: | [#422](https://github.com/ggerganov/whisper.cpp/discussions/422)
  - [sandrohanea/whisper.net](https://github.com/sandrohanea/whisper.net)
  - [NickDarvey/whisper](https://github.com/NickDarvey/whisper)
- [x] Python: | [#9](https://github.com/ggml-org/whisper.cpp/issues/9)
+- [x] Python: | [#9](https://github.com/ggerganov/whisper.cpp/issues/9)
  - [stlukey/whispercpp.py](https://github.com/stlukey/whispercpp.py) (Cython)
  - [AIWintermuteAI/whispercpp](https://github.com/AIWintermuteAI/whispercpp) (Updated fork of aarnphm/whispercpp)
  - [aarnphm/whispercpp](https://github.com/aarnphm/whispercpp) (Pybind11)
-  - [abdeladim-s/pywhispercpp](https://github.com/abdeladim-s/pywhispercpp) (Pybind11)
 - [x] R: [bnosac/audio.whisper](https://github.com/bnosac/audio.whisper)
 - [x] Unity: [macoron/whisper.unity](https://github.com/Macoron/whisper.unity)

-## XCFramework
-The XCFramework is a precompiled version of the library for iOS, visionOS, tvOS,
-and macOS. It can be used in Swift projects without the need to compile the
-library from source. For examples:
-```swift
-// swift-tools-version: 5.10
-// The swift-tools-version declares the minimum version of Swift required to build this package.
-
-import PackageDescription
-
-let package = Package(
-    name: "Whisper",
-    targets: [
-        .executableTarget(
-            name: "Whisper",
-            dependencies: [
-                "WhisperFramework"
-            ]),
-        .binaryTarget(
-            name: "WhisperFramework",
-            url: "https://github.com/ggml-org/whisper.cpp/releases/download/v1.7.5/whisper-v1.7.5-xcframework.zip",
-            checksum: "c7faeb328620d6012e130f3d705c51a6ea6c995605f2df50f6e1ad68c59c6c4a"
-        )
-    ]
-)
-```
-
-## Voice Activity Detection (VAD)
-Support for Voice Activity Detection (VAD) can be enabled using the `--vad`
-argument to `whisper-cli`. In addition to this option a VAD model is also
-required.
-
-The way this works is that first the audio samples are passed through
-the VAD model which will detect speech segments. Using this information the
-only the speech segments that are detected are extracted from the original audio
-input and passed to whisper for processing. This reduces the amount of audio
-data that needs to be processed by whisper and can significantly speed up the
-transcription process.
-
-The following VAD models are currently supported:
-
-### Silero-VAD
-[Silero-vad](https://github.com/snakers4/silero-vad) is a lightweight VAD model
-written in Python that is fast and accurate.
-
-Models can be downloaded by running the following command on Linux or MacOS:
-```console
-$ ./models/download-vad-model.sh silero-v5.1.2
-Downloading ggml model silero-v5.1.2 from 'https://huggingface.co/ggml-org/whisper-vad' ...
-ggml-silero-v5.1.2.bin        100%[==============================================>] 864.35K  --.-KB/s    in 0.04s
-Done! Model 'silero-v5.1.2' saved in '/path/models/ggml-silero-v5.1.2.bin'
-You can now use it like this:
-
-  $ ./build/bin/whisper-cli -vm /path/models/ggml-silero-v5.1.2.bin --vad -f samples/jfk.wav -m models/ggml-base.en.bin
-
-```
-And the following command on Windows:
-```console
-> .\models\download-vad-model.cmd silero-v5.1.2
-Downloading vad model silero-v5.1.2...
-Done! Model silero-v5.1.2 saved in C:\Users\danie\work\ai\whisper.cpp\ggml-silero-v5.1.2.bin
-You can now use it like this:
-
-C:\path\build\bin\Release\whisper-cli.exe -vm C:\path\ggml-silero-v5.1.2.bin --vad -m models/ggml-base.en.bin -f samples\jfk.wav
-
-```
-
-To see a list of all available models, run the above commands without any
-arguments.
-
-This model can be also be converted manually to ggml using the following command:
-```console
-$ python3 -m venv venv && source venv/bin/activate
-$ (venv) pip install silero-vad
-$ (venv) $ python models/convert-silero-vad-to-ggml.py --output models/silero.bin
-Saving GGML Silero-VAD model to models/silero-v5.1.2-ggml.bin
-```
-And it can then be used with whisper as follows:
-```console
-$ ./build/bin/whisper-cli \
-   --file ./samples/jfk.wav \
-   --model ./models/ggml-base.en.bin \
-   --vad \
-   --vad-model ./models/silero-v5.1.2-ggml.bin
-```
-
-### VAD Options
-
-* --vad-threshold: Threshold probability for speech detection. A probability
-for a speech segment/frame above this threshold will be considered as speech.
-
-* --vad-min-speech-duration-ms: Minimum speech duration in milliseconds. Speech
-segments shorter than this value will be discarded to filter out brief noise or
-false positives.
-
-* --vad-min-silence-duration-ms: Minimum silence duration in milliseconds. Silence
-periods must be at least this long to end a speech segment. Shorter silence
-periods will be ignored and included as part of the speech.
-
-* --vad-max-speech-duration-s: Maximum speech duration in seconds. Speech segments
-longer than this will be automatically split into multiple segments at silence
-points exceeding 98ms to prevent excessively long segments.
-
-* --vad-speech-pad-ms: Speech padding in milliseconds. Adds this amount of padding
-before and after each detected speech segment to avoid cutting off speech edges.
-
-* --vad-samples-overlap: Amount of audio to extend from each speech segment into
-the next one, in seconds (e.g., 0.10 = 100ms overlap). This ensures speech isn't
-cut off abruptly between segments when they're concatenated together.
-
 ## Examples

 There are various examples of using the library for different projects in the [examples](examples) folder.
@ -823,24 +831,25 @@ Some of the examples are even ported to run in the browser using WebAssembly. Ch

 | Example                                             | Web                                   | Description                                                                                                                     |
 | --------------------------------------------------- | ------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------- |
-| [whisper-cli](examples/cli)                         | [whisper.wasm](examples/whisper.wasm) | Tool for translating and transcribing audio using Whisper                                                                       |
-| [whisper-bench](examples/bench)                     | [bench.wasm](examples/bench.wasm)     | Benchmark the performance of Whisper on your machine                                                                            |
-| [whisper-stream](examples/stream)                   | [stream.wasm](examples/stream.wasm)   | Real-time transcription of raw microphone capture                                                                               |
-| [whisper-command](examples/command)                 | [command.wasm](examples/command.wasm) | Basic voice assistant example for receiving voice commands from the mic                                                         |
-| [whisper-server](examples/server)                   |                                       | HTTP transcription server with OAI-like API                                                                                     |
-| [whisper-talk-llama](examples/talk-llama)           |                                       | Talk with a LLaMA bot                                                                                                           |
+| [main](examples/main)                               | [whisper.wasm](examples/whisper.wasm) | Tool for translating and transcribing audio using Whisper                                                                       |
+| [bench](examples/bench)                             | [bench.wasm](examples/bench.wasm)     | Benchmark the performance of Whisper on your machine                                                                            |
+| [stream](examples/stream)                           | [stream.wasm](examples/stream.wasm)   | Real-time transcription of raw microphone capture                                                                               |
+| [command](examples/command)                         | [command.wasm](examples/command.wasm) | Basic voice assistant example for receiving voice commands from the mic                                                         |
+| [wchess](examples/wchess)                           | [wchess.wasm](examples/wchess)        | Voice-controlled chess                                                                                                          |
+| [talk](examples/talk)                               | [talk.wasm](examples/talk.wasm)       | Talk with a GPT-2 bot                                                                                                           |
+| [talk-llama](examples/talk-llama)                   |                                       | Talk with a LLaMA bot                                                                                                           |
 | [whisper.objc](examples/whisper.objc)               |                                       | iOS mobile application using whisper.cpp                                                                                        |
 | [whisper.swiftui](examples/whisper.swiftui)         |                                       | SwiftUI iOS / macOS application using whisper.cpp                                                                               |
 | [whisper.android](examples/whisper.android)         |                                       | Android mobile application using whisper.cpp                                                                                    |
 | [whisper.nvim](examples/whisper.nvim)               |                                       | Speech-to-text plugin for Neovim                                                                                                |
 | [generate-karaoke.sh](examples/generate-karaoke.sh) |                                       | Helper script to easily [generate a karaoke video](https://youtu.be/uj7hVta4blM) of raw audio capture                           |
-| [livestream.sh](examples/livestream.sh)             |                                       | [Livestream audio transcription](https://github.com/ggml-org/whisper.cpp/issues/185)                                            |
+| [livestream.sh](examples/livestream.sh)             |                                       | [Livestream audio transcription](https://github.com/ggerganov/whisper.cpp/issues/185)                                           |
 | [yt-wsp.sh](examples/yt-wsp.sh)                     |                                       | Download + transcribe and/or translate any VOD [(original)](https://gist.github.com/DaniruKun/96f763ec1a037cc92fe1a059b643b818) |
-| [wchess](examples/wchess)                           | [wchess.wasm](examples/wchess)        | Voice-controlled chess                                                                                                          |
+| [server](examples/server)                           |                                       | HTTP transcription server with OAI-like API                                                                                     |

-## [Discussions](https://github.com/ggml-org/whisper.cpp/discussions)
+## [Discussions](https://github.com/ggerganov/whisper.cpp/discussions)

 If you have any kind of feedback about this project feel free to use the Discussions section and open a new topic.
-You can use the [Show and tell](https://github.com/ggml-org/whisper.cpp/discussions/categories/show-and-tell) category
+You can use the [Show and tell](https://github.com/ggerganov/whisper.cpp/discussions/categories/show-and-tell) category
 to share your own projects that use `whisper.cpp`. If you have a question, make sure to check the
-[Frequently asked questions (#126)](https://github.com/ggml-org/whisper.cpp/discussions/126) discussion.
+[Frequently asked questions (#126)](https://github.com/ggerganov/whisper.cpp/discussions/126) discussion.
--- a/README_sycl.md
+++ b/README_sycl.md
@ -1,249 +1,249 @@
-# whisper.cpp for SYCL
-
-[Background](#background)
-
-[OS](#os)
-
-[Intel GPU](#intel-gpu)
-
-[Linux](#linux)
-
-[Environment Variable](#environment-variable)
-
-[Known Issue](#known-issue)
-
-[Todo](#todo)
-
-## Background
-
-SYCL is a higher-level programming model to improve programming productivity on various hardware accelerators—such as CPUs, GPUs, and FPGAs. It is a single-source embedded domain-specific language based on pure C++17.
-
-oneAPI is a specification that is open and standards-based, supporting multiple architecture types including but not limited to GPU, CPU, and FPGA. The spec has both direct programming and API-based programming paradigms.
-
-Intel uses the SYCL as direct programming language to support CPU, GPUs and FPGAs.
-
-To avoid  re-inventing the wheel, this code refers other code paths in llama.cpp (like OpenBLAS, cuBLAS, CLBlast). We use a open-source tool [SYCLomatic](https://github.com/oneapi-src/SYCLomatic) (Commercial release [Intel® DPC++ Compatibility Tool](https://www.intel.com/content/www/us/en/developer/tools/oneapi/dpc-compatibility-tool.html)) migrate to SYCL.
-
-The whisper.cpp for SYCL is used to support Intel GPUs.
-
-For Intel CPU, recommend to use whisper.cpp for X86 (Intel MKL build).
-
-## OS
-
-|OS|Status|Verified|
-|-|-|-|
-|Linux|Support|Ubuntu 22.04|
-|Windows|Ongoing| |
-
-
-## Intel GPU
-
-|Intel GPU| Status | Verified Model|
-|-|-|-|
-|Intel Data Center Max Series| Support| Max 1550|
-|Intel Data Center Flex Series| Support| Flex 170|
-|Intel Arc Series| Support| Arc 770|
-|Intel built-in Arc GPU| Support| built-in Arc GPU in Meteor Lake|
-|Intel iGPU| Support| iGPU in i5-1250P, i7-1165G7|
-
-
-## Linux
-
-### Setup Environment
-
-1. Install Intel GPU driver.
-
-a. Please install Intel GPU driver by official guide: [Install GPU Drivers](https://dgpu-docs.intel.com/driver/installation.html).
-
-Note: for iGPU, please install the client GPU driver.
-
-b. Add user to group: video, render.
-
-```
-sudo usermod -aG render username
-sudo usermod -aG video username
-```
-
-Note: re-login to enable it.
-
-c. Check
-
-```
-sudo apt install clinfo
-sudo clinfo -l
-```
-
-Output (example):
-
-```
-Platform #0: Intel(R) OpenCL Graphics
- `-- Device #0: Intel(R) Arc(TM) A770 Graphics
-
-
-Platform #0: Intel(R) OpenCL HD Graphics
- `-- Device #0: Intel(R) Iris(R) Xe Graphics [0x9a49]
-```
-
-2. Install Intel® oneAPI Base toolkit.
-
-
-a. Please follow the procedure in [Get the Intel® oneAPI Base Toolkit ](https://www.intel.com/content/www/us/en/developer/tools/oneapi/base-toolkit.html).
-
-Recommend to install to default folder: **/opt/intel/oneapi**.
-
-Following guide use the default folder as example. If you use other folder, please modify the following guide info with your folder.
-
-b. Check
-
-```
-source /opt/intel/oneapi/setvars.sh
-
-sycl-ls
-```
-
-There should be one or more level-zero devices. Like **[ext_oneapi_level_zero:gpu:0]**.
-
-Output (example):
-```
-[opencl:acc:0] Intel(R) FPGA Emulation Platform for OpenCL(TM), Intel(R) FPGA Emulation Device OpenCL 1.2  [2023.16.10.0.17_160000]
-[opencl:cpu:1] Intel(R) OpenCL, 13th Gen Intel(R) Core(TM) i7-13700K OpenCL 3.0 (Build 0) [2023.16.10.0.17_160000]
-[opencl:gpu:2] Intel(R) OpenCL Graphics, Intel(R) Arc(TM) A770 Graphics OpenCL 3.0 NEO  [23.30.26918.50]
-[ext_oneapi_level_zero:gpu:0] Intel(R) Level-Zero, Intel(R) Arc(TM) A770 Graphics 1.3 [1.3.26918]
-
-```
-
-2. Build locally:
-
-```
-mkdir -p build
-cd build
-source /opt/intel/oneapi/setvars.sh
-
-#for FP16
-#cmake .. -DWHISPER_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DWHISPER_SYCL_F16=ON 
-
-#for FP32
-cmake .. -DWHISPER_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx
-
-#build example/main only
-#cmake --build . --config Release --target main
-
-#build all binary
-cmake --build . --config Release -v
-
-```
-
-or
-
-```
-./examples/sycl/build.sh
-```
-
-Note:
-
- By default, it will build for all binary files. It will take more time. To reduce the time, we recommend to build for **example/main** only.
-
-### Run
-
-1. Put model file to folder **models**
-
-2. Enable oneAPI running environment
-
-```
-source /opt/intel/oneapi/setvars.sh
-```
-
-3. List device ID
-
-Run without parameter:
-
-```
-./build/bin/ls-sycl-device
-
-or
-
-./build/bin/main
-```
-
-Check the ID in startup log, like:
-
-```
-found 4 SYCL devices:
-  Device 0: Intel(R) Arc(TM) A770 Graphics,	compute capability 1.3,
-    max compute_units 512,	max work group size 1024,	max sub group size 32,	global mem size 16225243136
-  Device 1: Intel(R) FPGA Emulation Device,	compute capability 1.2,
-    max compute_units 24,	max work group size 67108864,	max sub group size 64,	global mem size 67065057280
-  Device 2: 13th Gen Intel(R) Core(TM) i7-13700K,	compute capability 3.0,
-    max compute_units 24,	max work group size 8192,	max sub group size 64,	global mem size 67065057280
-  Device 3: Intel(R) Arc(TM) A770 Graphics,	compute capability 3.0,
-    max compute_units 512,	max work group size 1024,	max sub group size 32,	global mem size 16225243136
-
-```
-
-|Attribute|Note|
-|-|-|
-|compute capability 1.3|Level-zero running time, recommended |
-|compute capability 3.0|OpenCL running time, slower than level-zero in most cases|
-
-4. Set device ID and execute whisper.cpp
-
-Set device ID = 0 by **GGML_SYCL_DEVICE=0**
-
-```
-GGML_SYCL_DEVICE=0 ./build/bin/main -m models/ggml-base.en.bin -f samples/jfk.wav
-```
-or run by script:
-
-```
-./examples/sycl/run_whisper.sh
-```
-
-
-
-5. Check the device ID in output
-
-Like:
-```
-Using device **0** (Intel(R) Arc(TM) A770 Graphics) as main device
-```
-
-
-## Environment Variable
-
-#### Build
-
-|Name|Value|Function|
-|-|-|-|
-|WHISPER_SYCL|ON (mandatory)|Enable build with SYCL code path. <br>For FP32/FP16, WHISPER_SYCL=ON is mandatory.|
-|WHISPER_SYCL_F16|ON (optional)|Enable FP16 build with SYCL code path.For FP32, do not set it.|
-|CMAKE_C_COMPILER|icx|Use icx compiler for SYCL code path|
-|CMAKE_CXX_COMPILER|icpx|use icpx for SYCL code path|
-
-#### Running
-
-
-|Name|Value|Function|
-|-|-|-|
-|GGML_SYCL_DEVICE|0 (default) or 1|Set the device id used. Check the device ids by default running output|
-|GGML_SYCL_DEBUG|0 (default) or 1|Enable log function by macro: GGML_SYCL_DEBUG|
-
-## Known Issue
-
- Error:  `error while loading shared libraries: libsycl.so.7: cannot open shared object file: No such file or directory`.
-
-  Miss to enable oneAPI running environment.
-
-  Install oneAPI base toolkit and enable it by: `source /opt/intel/oneapi/setvars.sh`.
-
-
- Hang during startup
-
-  llama.cpp use mmap as default way to read model file and copy to GPU. In some system, memcpy will be abnormal and block.
-
-  Solution: add **--no-mmap**.
-
-## Todo
-
- Support to build in Windows.
-
- Support multiple cards.
+# whisper.cpp for SYCL
+
+[Background](#background)
+
+[OS](#os)
+
+[Intel GPU](#intel-gpu)
+
+[Linux](#linux)
+
+[Environment Variable](#environment-variable)
+
+[Known Issue](#known-issue)
+
+[Todo](#todo)
+
+## Background
+
+SYCL is a higher-level programming model to improve programming productivity on various hardware accelerators<EFBFBD>such as CPUs, GPUs, and FPGAs. It is a single-source embedded domain-specific language based on pure C++17.
+
+oneAPI is a specification that is open and standards-based, supporting multiple architecture types including but not limited to GPU, CPU, and FPGA. The spec has both direct programming and API-based programming paradigms.
+
+Intel uses the SYCL as direct programming language to support CPU, GPUs and FPGAs.
+
+To avoid  re-inventing the wheel, this code refers other code paths in llama.cpp (like OpenBLAS, cuBLAS, CLBlast). We use a open-source tool [SYCLomatic](https://github.com/oneapi-src/SYCLomatic) (Commercial release [Intel<EFBFBD> DPC++ Compatibility Tool](https://www.intel.com/content/www/us/en/developer/tools/oneapi/dpc-compatibility-tool.html)) migrate to SYCL.
+
+The whisper.cpp for SYCL is used to support Intel GPUs.
+
+For Intel CPU, recommend to use whisper.cpp for X86 (Intel MKL build).
+
+## OS
+
+|OS|Status|Verified|
+|-|-|-|
+|Linux|Support|Ubuntu 22.04|
+|Windows|Ongoing| |
+
+
+## Intel GPU
+
+|Intel GPU| Status | Verified Model|
+|-|-|-|
+|Intel Data Center Max Series| Support| Max 1550|
+|Intel Data Center Flex Series| Support| Flex 170|
+|Intel Arc Series| Support| Arc 770|
+|Intel built-in Arc GPU| Support| built-in Arc GPU in Meteor Lake|
+|Intel iGPU| Support| iGPU in i5-1250P, i7-1165G7|
+
+
+## Linux
+
+### Setup Environment
+
+1. Install Intel GPU driver.
+
+a. Please install Intel GPU driver by official guide: [Install GPU Drivers](https://dgpu-docs.intel.com/driver/installation.html).
+
+Note: for iGPU, please install the client GPU driver.
+
+b. Add user to group: video, render.
+
+```
+sudo usermod -aG render username
+sudo usermod -aG video username
+```
+
+Note: re-login to enable it.
+
+c. Check
+
+```
+sudo apt install clinfo
+sudo clinfo -l
+```
+
+Output (example):
+
+```
+Platform #0: Intel(R) OpenCL Graphics
+ `-- Device #0: Intel(R) Arc(TM) A770 Graphics
+
+
+Platform #0: Intel(R) OpenCL HD Graphics
+ `-- Device #0: Intel(R) Iris(R) Xe Graphics [0x9a49]
+```
+
+2. Install Intel<EFBFBD> oneAPI Base toolkit.
+
+
+a. Please follow the procedure in [Get the Intel<EFBFBD> oneAPI Base Toolkit ](https://www.intel.com/content/www/us/en/developer/tools/oneapi/base-toolkit.html).
+
+Recommend to install to default folder: **/opt/intel/oneapi**.
+
+Following guide use the default folder as example. If you use other folder, please modify the following guide info with your folder.
+
+b. Check
+
+```
+source /opt/intel/oneapi/setvars.sh
+
+sycl-ls
+```
+
+There should be one or more level-zero devices. Like **[ext_oneapi_level_zero:gpu:0]**.
+
+Output (example):
+```
+[opencl:acc:0] Intel(R) FPGA Emulation Platform for OpenCL(TM), Intel(R) FPGA Emulation Device OpenCL 1.2  [2023.16.10.0.17_160000]
+[opencl:cpu:1] Intel(R) OpenCL, 13th Gen Intel(R) Core(TM) i7-13700K OpenCL 3.0 (Build 0) [2023.16.10.0.17_160000]
+[opencl:gpu:2] Intel(R) OpenCL Graphics, Intel(R) Arc(TM) A770 Graphics OpenCL 3.0 NEO  [23.30.26918.50]
+[ext_oneapi_level_zero:gpu:0] Intel(R) Level-Zero, Intel(R) Arc(TM) A770 Graphics 1.3 [1.3.26918]
+
+```
+
+2. Build locally:
+
+```
+mkdir -p build
+cd build
+source /opt/intel/oneapi/setvars.sh
+
+#for FP16
+#cmake .. -DWHISPER_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DWHISPER_SYCL_F16=ON 
+
+#for FP32
+cmake .. -DWHISPER_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx
+
+#build example/main only
+#cmake --build . --config Release --target main
+
+#build all binary
+cmake --build . --config Release -v
+
+```
+
+or
+
+```
+./examples/sycl/build.sh
+```
+
+Note:
+
+- By default, it will build for all binary files. It will take more time. To reduce the time, we recommend to build for **example/main** only.
+
+### Run
+
+1. Put model file to folder **models**
+
+2. Enable oneAPI running environment
+
+```
+source /opt/intel/oneapi/setvars.sh
+```
+
+3. List device ID
+
+Run without parameter:
+
+```
+./build/bin/ls-sycl-device
+
+or
+
+./build/bin/main
+```
+
+Check the ID in startup log, like:
+
+```
+found 4 SYCL devices:
+  Device 0: Intel(R) Arc(TM) A770 Graphics,	compute capability 1.3,
+    max compute_units 512,	max work group size 1024,	max sub group size 32,	global mem size 16225243136
+  Device 1: Intel(R) FPGA Emulation Device,	compute capability 1.2,
+    max compute_units 24,	max work group size 67108864,	max sub group size 64,	global mem size 67065057280
+  Device 2: 13th Gen Intel(R) Core(TM) i7-13700K,	compute capability 3.0,
+    max compute_units 24,	max work group size 8192,	max sub group size 64,	global mem size 67065057280
+  Device 3: Intel(R) Arc(TM) A770 Graphics,	compute capability 3.0,
+    max compute_units 512,	max work group size 1024,	max sub group size 32,	global mem size 16225243136
+
+```
+
+|Attribute|Note|
+|-|-|
+|compute capability 1.3|Level-zero running time, recommended |
+|compute capability 3.0|OpenCL running time, slower than level-zero in most cases|
+
+4. Set device ID and execute whisper.cpp
+
+Set device ID = 0 by **GGML_SYCL_DEVICE=0**
+
+```
+GGML_SYCL_DEVICE=0 ./build/bin/main -m models/ggml-base.en.bin -f samples/jfk.wav
+```
+or run by script:
+
+```
+./examples/sycl/run_whisper.sh
+```
+
+
+
+5. Check the device ID in output
+
+Like:
+```
+Using device **0** (Intel(R) Arc(TM) A770 Graphics) as main device
+```
+
+
+## Environment Variable
+
+#### Build
+
+|Name|Value|Function|
+|-|-|-|
+|WHISPER_SYCL|ON (mandatory)|Enable build with SYCL code path. <br>For FP32/FP16, WHISPER_SYCL=ON is mandatory.|
+|WHISPER_SYCL_F16|ON (optional)|Enable FP16 build with SYCL code path.For FP32, do not set it.|
+|CMAKE_C_COMPILER|icx|Use icx compiler for SYCL code path|
+|CMAKE_CXX_COMPILER|icpx|use icpx for SYCL code path|
+
+#### Running
+
+
+|Name|Value|Function|
+|-|-|-|
+|GGML_SYCL_DEVICE|0 (default) or 1|Set the device id used. Check the device ids by default running output|
+|GGML_SYCL_DEBUG|0 (default) or 1|Enable log function by macro: GGML_SYCL_DEBUG|
+
+## Known Issue
+
+- Error:  `error while loading shared libraries: libsycl.so.7: cannot open shared object file: No such file or directory`.
+
+  Miss to enable oneAPI running environment.
+
+  Install oneAPI base toolkit and enable it by: `source /opt/intel/oneapi/setvars.sh`.
+
+
+- Hang during startup
+
+  llama.cpp use mmap as default way to read model file and copy to GPU. In some system, memcpy will be abnormal and block.
+
+  Solution: add **--no-mmap**.
+
+## Todo
+
+- Support to build in Windows.
+
+- Support multiple cards.
--- a/bindings/go/Makefile
+++ b/bindings/go/Makefile
@ -11,16 +11,11 @@ UNAME_M := $(shell uname -m)
 endif

 GGML_METAL_PATH_RESOURCES := $(abspath ../..)
-BUILD_DIR := build_go
+BUILD_DIR := build
 MODELS_DIR := models
 EXAMPLES_DIR := $(wildcard examples/*)
-INCLUDE_PATH := $(abspath ../../include):$(abspath ../../ggml/include)
-LIBRARY_PATH := $(abspath ../../${BUILD_DIR}/src:$(abspath ../../${BUILD_DIR}/ggml/src))
-
-ifeq ($(GGML_CUDA),1)
-	LIBRARY_PATH := $(LIBRARY_PATH):$(CUDA_PATH)/targets/$(UNAME_M)-linux/lib/
-	BUILD_FLAGS := -ldflags "-extldflags '-lcudart -lcuda -lcublas'"
-endif
+INCLUDE_PATH := $(abspath ../..)
+LIBRARY_PATH := $(abspath ../..)

 ifeq ($(UNAME_S),Darwin)
 	EXT_LDFLAGS := -framework Foundation -framework Metal -framework MetalKit
@ -29,10 +24,8 @@ endif
 all: clean whisper examples

 whisper: mkdir
-	cmake -S ../.. -B ../../${BUILD_DIR} \
-		-DCMAKE_BUILD_TYPE=Release \
-		-DBUILD_SHARED_LIBS=OFF
-	cmake --build ../../${BUILD_DIR} --target whisper
+	@echo Build whisper
+	@${MAKE} -C ../.. libwhisper.a

 test: model-small whisper modtidy
 ifeq ($(UNAME_S),Darwin)
--- a/bindings/go/README.md
+++ b/bindings/go/README.md
@ -31,7 +31,7 @@ func main() {
 	if err != nil {
 		panic(err)
 	}
-	if err := context.Process(samples, nil, nil, nil); err != nil {
+	if err := context.Process(samples, nil, nil); err != nil {
 		return err
 	}

@ -51,7 +51,7 @@ func main() {
 In order to build, you need to have the Go compiler installed. You can get it from [here](https://golang.org/dl/). Run the tests with:

 ```bash
-git clone https://github.com/ggml-org/whisper.cpp.git
+git clone https://github.com/ggerganov/whisper.cpp.git
 cd whisper.cpp/bindings/go
 make test
 ```
@ -62,12 +62,6 @@ This will compile a static `libwhisper.a` in a `build` folder, download a model
 make examples
 ```

-To build using cuda support add `GGML_CUDA=1`:
-
-```bash
-GGML_CUDA=1 make examples
-```
-
 The examples are placed in the `build` directory. Once built, you can download all the models with the following command:

 ```bash
@ -98,7 +92,7 @@ The API Documentation:

 Getting help:

-  * Follow the discussion for the go bindings [here](https://github.com/ggml-org/whisper.cpp/discussions/312)
+  * Follow the discussion for the go bindings [here](https://github.com/ggerganov/whisper.cpp/discussions/312)

 ## License

--- a/bindings/go/doc.go
+++ b/bindings/go/doc.go
@ -1,5 +1,5 @@
 /*
-github.com/ggml-org/whisper.cpp/bindings/go
+github.com/ggerganov/whisper.cpp/bindings/go
 provides a speech-to-text service bindings for the Go programming language.
 */
 package whisper
--- a/bindings/go/examples/go-model-download/context.go
+++ b/bindings/go/examples/go-model-download/context.go
@ -9,23 +9,22 @@ import (
 // ContextForSignal returns a context object which is cancelled when a signal
 // is received. It returns nil if no signal parameter is provided
 func ContextForSignal(signals ...os.Signal) context.Context {
-    if len(signals) == 0 {
-        return nil
-    }
+	if len(signals) == 0 {
+		return nil
+	}

-    ch := make(chan os.Signal, 1) // Buffered channel with space for 1 signal
-    ctx, cancel := context.WithCancel(context.Background())
+	ch := make(chan os.Signal)
+	ctx, cancel := context.WithCancel(context.Background())

-    // Send message on channel when signal received
-    signal.Notify(ch, signals...)
+	// Send message on channel when signal received
+	signal.Notify(ch, signals...)

-    // When any signal is received, call cancel
-    go func() {
-        <-ch
-        cancel()
-    }()
+	// When any signal received, call cancel
+	go func() {
+		<-ch
+		cancel()
+	}()

-    // Return success
-    return ctx
+	// Return success
+	return ctx
 }
-
--- a/bindings/go/examples/go-model-download/main.go
+++ b/bindings/go/examples/go-model-download/main.go
@ -9,7 +9,6 @@ import (
 	"net/url"
 	"os"
 	"path/filepath"
-	"strings"
 	"syscall"
 	"time"
 )
@ -18,27 +17,14 @@ import (
 // CONSTANTS

 const (
-	srcUrl  = "https://huggingface.co/ggerganov/whisper.cpp/resolve/main/" // The location of the models
-	srcExt  = ".bin"                                                       // Filename extension
-	bufSize = 1024 * 64                                                    // Size of the buffer used for downloading the model
+	srcUrl  = "https://huggingface.co/ggerganov/whisper.cpp/resolve/main" // The location of the models
+	srcExt  = ".bin"                                                      // Filename extension
+	bufSize = 1024 * 64                                                   // Size of the buffer used for downloading the model
 )

 var (
 	// The models which will be downloaded, if no model is specified as an argument
-	modelNames = []string{
-		"tiny", "tiny-q5_1", "tiny-q8_0",
-		"tiny.en", "tiny.en-q5_1", "tiny.en-q8_0",
-		"base", "base-q5_1", "base-q8_0",
-		"base.en", "base.en-q5_1", "base.en-q8_0",
-		"small", "small-q5_1", "small-q8_0",
-		"small.en", "small.en-q5_1", "small.en-q8_0",
-		"medium", "medium-q5_0", "medium-q8_0",
-		"medium.en", "medium.en-q5_0", "medium.en-q8_0",
-		"large-v1",
-		"large-v2", "large-v2-q5_0", "large-v2-q8_0",
-		"large-v3", "large-v3-q5_0",
-		"large-v3-turbo", "large-v3-turbo-q5_0", "large-v3-turbo-q8_0",
-	}
+	modelNames = []string{"ggml-tiny.en", "ggml-tiny", "ggml-base.en", "ggml-base", "ggml-small.en", "ggml-small", "ggml-medium.en", "ggml-medium", "ggml-large-v1", "ggml-large-v2", "ggml-large-v3"}
 )

 var (
@ -58,25 +44,7 @@ var (
 func main() {
 	flag.Usage = func() {
 		name := filepath.Base(flag.CommandLine.Name())
-		fmt.Fprintf(flag.CommandLine.Output(), `
-			Usage: %s [options] [<model>...]
-
-			Options:
-  			-out string     Specify the output folder where models will be saved.
-                  			Default: Current working directory.
-  			-timeout duration Set the maximum duration for downloading a model.
-            			      Example: 10m, 1h (default: 30m0s).
-  			-quiet           Suppress all output except errors.
-
-			Examples:
-  			1. Download a specific model:
-     			%s -out ./models tiny-q8_0
-
-			  2. Download all models:
-     			%s -out ./models
-
-			`, name, name, name)
-
+		fmt.Fprintf(flag.CommandLine.Output(), "Usage: %s [options] <model>\n\n", name)
 		flag.PrintDefaults()
 	}
 	flag.Parse()
@ -146,87 +114,23 @@ func GetOut() (string, error) {
 // GetModels returns the list of models to download
 func GetModels() []string {
 	if flag.NArg() == 0 {
-		fmt.Println("No model specified.")
-		fmt.Println("Preparing to download all models...")
-
-		// Calculate total download size
-		fmt.Println("Calculating total download size...")
-		totalSize, err := CalculateTotalDownloadSize(modelNames)
-		if err != nil {
-			fmt.Println("Error calculating download sizes:", err)
-			os.Exit(1)
-		}
-
-		fmt.Println("View available models: https://huggingface.co/ggerganov/whisper.cpp/tree/main")
-		fmt.Printf("Total download size: %.2f GB\n", float64(totalSize)/(1024*1024*1024))
-		fmt.Println("Would you like to download all models? (y/N)")
-
-		// Prompt for user input
-		var response string
-		fmt.Scanln(&response)
-		if response != "y" && response != "Y" {
-			fmt.Println("Aborting. Specify a model to download.")
-			os.Exit(0)
-		}
-
-		return modelNames // Return all models if confirmed
+		return modelNames
+	} else {
+		return flag.Args()
 	}
-	return flag.Args() // Return specific models if arguments are provided
-}
-
-func CalculateTotalDownloadSize(models []string) (int64, error) {
-	var totalSize int64
-	client := http.Client{}
-
-	for _, model := range models {
-		modelURL, err := URLForModel(model)
-		if err != nil {
-			return 0, err
-		}
-
-		// Issue a HEAD request to get the file size
-		req, err := http.NewRequest("HEAD", modelURL, nil)
-		if err != nil {
-			return 0, err
-		}
-
-		resp, err := client.Do(req)
-		if err != nil {
-			return 0, err
-		}
-		resp.Body.Close()
-
-		if resp.StatusCode != http.StatusOK {
-			fmt.Printf("Warning: Unable to fetch size for %s (HTTP %d)\n", model, resp.StatusCode)
-			continue
-		}
-
-		size := resp.ContentLength
-		totalSize += size
-	}
-	return totalSize, nil
 }

 // URLForModel returns the URL for the given model on huggingface.co
 func URLForModel(model string) (string, error) {
-	// Ensure "ggml-" prefix is added only once
-	if !strings.HasPrefix(model, "ggml-") {
-		model = "ggml-" + model
-	}
-
-	// Ensure ".bin" extension is added only once
 	if filepath.Ext(model) != srcExt {
 		model += srcExt
 	}
-
-	// Parse the base URL
 	url, err := url.Parse(srcUrl)
 	if err != nil {
 		return "", err
+	} else {
+		url.Path = filepath.Join(url.Path, model)
 	}
-
-	// Ensure no trailing slash in the base URL
-	url.Path = fmt.Sprintf("%s/%s", strings.TrimSuffix(url.Path, "/"), model)
 	return url.String(), nil
 }

--- a/bindings/go/examples/go-whisper/process.go
+++ b/bindings/go/examples/go-whisper/process.go
@ -67,7 +67,7 @@ func Process(model whisper.Model, path string, flags *Flags) error {
 	// Process the data
 	fmt.Fprintf(flags.Output(), "  ...processing %q\n", path)
 	context.ResetTimings()
-	if err := context.Process(data, nil, cb, nil); err != nil {
+	if err := context.Process(data, cb, nil); err != nil {
 		return err
 	}

--- a/bindings/go/go.mod
+++ b/bindings/go/go.mod
@ -1,10 +1,10 @@
 module github.com/ggerganov/whisper.cpp/bindings/go

-go 1.23
+go 1.19

 require (
 	github.com/go-audio/wav v1.1.0
-	github.com/stretchr/testify v1.9.0
+	github.com/stretchr/testify v1.8.1
 )

 require (
--- a/bindings/go/go.sum
+++ b/bindings/go/go.sum
@ -1,3 +1,4 @@
+github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/go-audio/audio v1.0.0 h1:zS9vebldgbQqktK4H0lUqWrG8P0NxCJVqcj7ZpNnwd4=
@ -8,9 +9,15 @@ github.com/go-audio/wav v1.1.0 h1:jQgLtbqBzY7G+BM8fXF7AHUk1uHUviWS4X39d5rsL2g=
 github.com/go-audio/wav v1.1.0/go.mod h1:mpe9qfwbScEbkd8uybLuIpTgHyrISw/OTuvjUW2iGtE=
 github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
-github.com/stretchr/testify v1.9.0 h1:HtqpIVDClZ4nwg75+f6Lvsy/wHu+3BoSGCbBAcpTsTg=
-github.com/stretchr/testify v1.9.0/go.mod h1:r2ic/lqez/lEtzL7wO/rwa5dbSLXVDPFyf8C91i36aY=
+github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
+github.com/stretchr/objx v0.4.0/go.mod h1:YvHI0jy2hoMjB+UWwv71VJQ9isScKT/TqJzVSSt89Yw=
+github.com/stretchr/objx v0.5.0/go.mod h1:Yh+to48EsGEfYuaHDzXPcE3xhTkx73EhmCGUpEOglKo=
+github.com/stretchr/testify v1.7.1/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg=
+github.com/stretchr/testify v1.8.0/go.mod h1:yNjHg4UonilssWZ8iaSj1OCr/vHnekPRkoO+kdMU+MU=
+github.com/stretchr/testify v1.8.1 h1:w7B6lhMri9wdJUVmEZPGGhZzrYTPvgJArz7wNPgYKsk=
+github.com/stretchr/testify v1.8.1/go.mod h1:w2LPCIKwWwSfY2zedu0+kehJoqGctiVI29o6fzry7u4=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
+gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
 gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
 gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
--- a/bindings/go/params.go
+++ b/bindings/go/params.go
@ -119,28 +119,6 @@ func (p *Params) SetAudioCtx(n int) {
 	p.audio_ctx = C.int(n)
 }

-func (p *Params) SetMaxContext(n int) {
-	p.n_max_text_ctx = C.int(n)
-}
-
-func (p *Params) SetBeamSize(n int) {
-	p.beam_search.beam_size = C.int(n)
-}
-
-func (p *Params) SetEntropyThold(t float32) {
-	p.entropy_thold = C.float(t)
-}
-
-func (p *Params) SetTemperature(t float32) {
-	p.temperature = C.float(t)
-}
-
-// Sets the fallback temperature incrementation
-// Pass -1.0 to disable this feature
-func (p *Params) SetTemperatureFallback(t float32) {
-	p.temperature_inc = C.float(t)
-}
-
 // Set initial prompt
 func (p *Params) SetInitialPrompt(prompt string) {
 	p.initial_prompt = C.CString(prompt)
@ -171,10 +149,6 @@ func (p *Params) String() string {
 	str += fmt.Sprintf(" duration_ms=%d", p.duration_ms)
 	str += fmt.Sprintf(" audio_ctx=%d", p.audio_ctx)
 	str += fmt.Sprintf(" initial_prompt=%s", C.GoString(p.initial_prompt))
-	str += fmt.Sprintf(" entropy_thold=%f", p.entropy_thold)
-	str += fmt.Sprintf(" temperature=%f", p.temperature)
-	str += fmt.Sprintf(" temperature_inc=%f", p.temperature_inc)
-	str += fmt.Sprintf(" beam_size=%d", p.beam_search.beam_size)
 	if p.translate {
 		str += " translate"
 	}
--- a/bindings/go/pkg/whisper/context.go
+++ b/bindings/go/pkg/whisper/context.go
@ -71,10 +71,6 @@ func (context *context) Language() string {
 	return whisper.Whisper_lang_str(context.params.Language())
 }

-func (context *context) DetectedLanguage() string {
-	return whisper.Whisper_lang_str(context.model.ctx.Whisper_full_lang_id())
-}
-
 // Set translate flag
 func (context *context) SetTranslate(v bool) {
 	context.params.SetTranslate(v)
@ -129,32 +125,6 @@ func (context *context) SetAudioCtx(n uint) {
 	context.params.SetAudioCtx(int(n))
 }

-// Set maximum number of text context tokens to store
-func (context *context) SetMaxContext(n int) {
-	context.params.SetMaxContext(n)
-}
-
-// Set Beam Size
-func (context *context) SetBeamSize(n int) {
-	context.params.SetBeamSize(n)
-}
-
-// Set Entropy threshold
-func (context *context) SetEntropyThold(t float32) {
-	context.params.SetEntropyThold(t)
-}
-
-// Set Temperature
-func (context *context) SetTemperature(t float32) {
-	context.params.SetTemperature(t)
-}
-
-// Set the fallback temperature incrementation
-// Pass -1.0 to disable this feature
-func (context *context) SetTemperatureFallback(t float32) {
-	context.params.SetTemperatureFallback(t)
-}
-
 // Set initial prompt
 func (context *context) SetInitialPrompt(prompt string) {
 	context.params.SetInitialPrompt(prompt)
@ -193,7 +163,6 @@ func (context *context) WhisperLangAutoDetect(offset_ms int, n_threads int) ([]f
 // Process new sample data and return any errors
 func (context *context) Process(
 	data []float32,
-	callEncoderBegin EncoderBeginCallback,
 	callNewSegment SegmentCallback,
 	callProgress ProgressCallback,
 ) error {
@ -208,20 +177,7 @@ func (context *context) Process(
 	// We don't do parallel processing at the moment
 	processors := 0
 	if processors > 1 {
-		if err := context.model.ctx.Whisper_full_parallel(context.params, data, processors, callEncoderBegin,
-			func(new int) {
-				if callNewSegment != nil {
-					num_segments := context.model.ctx.Whisper_full_n_segments()
-					s0 := num_segments - new
-					for i := s0; i < num_segments; i++ {
-						callNewSegment(toSegment(context.model.ctx, i))
-					}
-				}
-			}); err != nil {
-			return err
-		}
-	} else if err := context.model.ctx.Whisper_full(context.params, data, callEncoderBegin,
-		func(new int) {
+		if err := context.model.ctx.Whisper_full_parallel(context.params, data, processors, nil, func(new int) {
 			if callNewSegment != nil {
 				num_segments := context.model.ctx.Whisper_full_n_segments()
 				s0 := num_segments - new
@ -229,11 +185,22 @@ func (context *context) Process(
 					callNewSegment(toSegment(context.model.ctx, i))
 				}
 			}
-		}, func(progress int) {
-			if callProgress != nil {
-				callProgress(progress)
-			}
 		}); err != nil {
+			return err
+		}
+	} else if err := context.model.ctx.Whisper_full(context.params, data, nil, func(new int) {
+		if callNewSegment != nil {
+			num_segments := context.model.ctx.Whisper_full_n_segments()
+			s0 := num_segments - new
+			for i := s0; i < num_segments; i++ {
+				callNewSegment(toSegment(context.model.ctx, i))
+			}
+		}
+	}, func(progress int) {
+		if callProgress != nil {
+			callProgress(progress)
+		}
+	}); err != nil {
 		return err
 	}

--- a/bindings/go/pkg/whisper/context_test.go
+++ b/bindings/go/pkg/whisper/context_test.go
@ -4,121 +4,52 @@ import (
 	"os"
 	"testing"

-	"github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
-	"github.com/go-audio/wav"
+	// Packages
+	whisper "github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
 	assert "github.com/stretchr/testify/assert"
 )

-func TestSetLanguage(t *testing.T) {
-	assert := assert.New(t)
+const (
+	ModelPath  = "../../models/ggml-tiny.bin"
+	SamplePath = "../../samples/jfk.wav"
+)

+func Test_Whisper_000(t *testing.T) {
+	assert := assert.New(t)
+	if _, err := os.Stat(ModelPath); os.IsNotExist(err) {
+		t.Skip("Skipping test, model not found:", ModelPath)
+	}
+	if _, err := os.Stat(SamplePath); os.IsNotExist(err) {
+		t.Skip("Skipping test, sample not found:", SamplePath)
+	}
+
+	// Load model
+	model, err := whisper.New(ModelPath)
+	assert.NoError(err)
+	assert.NotNil(model)
+	assert.NoError(model.Close())
+
+	t.Log("languages=", model.Languages())
+}
+
+func Test_Whisper_001(t *testing.T) {
+	assert := assert.New(t)
+	if _, err := os.Stat(ModelPath); os.IsNotExist(err) {
+		t.Skip("Skipping test, model not found:", ModelPath)
+	}
+	if _, err := os.Stat(SamplePath); os.IsNotExist(err) {
+		t.Skip("Skipping test, sample not found:", SamplePath)
+	}
+
+	// Load model
 	model, err := whisper.New(ModelPath)
 	assert.NoError(err)
 	assert.NotNil(model)
 	defer model.Close()

-	context, err := model.NewContext()
+	// Get context for decoding
+	ctx, err := model.NewContext()
 	assert.NoError(err)
+	assert.NotNil(ctx)

-	// This returns an error since
-	// the model 'models/ggml-small.en.bin'
-	// that is loaded is not multilingual
-	err = context.SetLanguage("en")
-	assert.Error(err)
-}
-
-func TestContextModelIsMultilingual(t *testing.T) {
-	assert := assert.New(t)
-
-	model, err := whisper.New(ModelPath)
-	assert.NoError(err)
-	assert.NotNil(model)
-	defer model.Close()
-
-	context, err := model.NewContext()
-	assert.NoError(err)
-
-	isMultilingual := context.IsMultilingual()
-
-	// This returns false since
-	// the model 'models/ggml-small.en.bin'
-	// that is loaded is not multilingual
-	assert.False(isMultilingual)
-}
-
-func TestLanguage(t *testing.T) {
-	assert := assert.New(t)
-
-	model, err := whisper.New(ModelPath)
-	assert.NoError(err)
-	assert.NotNil(model)
-	defer model.Close()
-
-	context, err := model.NewContext()
-	assert.NoError(err)
-
-	// This always returns en since
-	// the model 'models/ggml-small.en.bin'
-	// that is loaded is not multilingual
-	expectedLanguage := "en"
-	actualLanguage := context.Language()
-	assert.Equal(expectedLanguage, actualLanguage)
-}
-
-func TestProcess(t *testing.T) {
-	assert := assert.New(t)
-
-	fh, err := os.Open(SamplePath)
-	assert.NoError(err)
-	defer fh.Close()
-
-	// Decode the WAV file - load the full buffer
-	dec := wav.NewDecoder(fh)
-	buf, err := dec.FullPCMBuffer()
-	assert.NoError(err)
-	assert.Equal(uint16(1), dec.NumChans)
-
-	data := buf.AsFloat32Buffer().Data
-
-	model, err := whisper.New(ModelPath)
-	assert.NoError(err)
-	assert.NotNil(model)
-	defer model.Close()
-
-	context, err := model.NewContext()
-	assert.NoError(err)
-
-	err = context.Process(data, nil, nil, nil)
-	assert.NoError(err)
-}
-
-func TestDetectedLanguage(t *testing.T) {
-	assert := assert.New(t)
-
-	fh, err := os.Open(SamplePath)
-	assert.NoError(err)
-	defer fh.Close()
-
-	// Decode the WAV file - load the full buffer
-	dec := wav.NewDecoder(fh)
-	buf, err := dec.FullPCMBuffer()
-	assert.NoError(err)
-	assert.Equal(uint16(1), dec.NumChans)
-
-	data := buf.AsFloat32Buffer().Data
-
-	model, err := whisper.New(ModelPath)
-	assert.NoError(err)
-	assert.NotNil(model)
-	defer model.Close()
-
-	context, err := model.NewContext()
-	assert.NoError(err)
-
-	err = context.Process(data, nil, nil, nil)
-	assert.NoError(err)
-
-	expectedLanguage := "en"
-	actualLanguage := context.DetectedLanguage()
-	assert.Equal(expectedLanguage, actualLanguage)
 }
--- a/bindings/go/pkg/whisper/interface.go
+++ b/bindings/go/pkg/whisper/interface.go
@ -16,10 +16,6 @@ type SegmentCallback func(Segment)
 // processing. It is called during the Process function
 type ProgressCallback func(int)

-// EncoderBeginCallback is the callback function for checking if we want to
-// continue processing. It is called during the Process function
-type EncoderBeginCallback func() bool
-
 // Model is the interface to a whisper model. Create a new model with the
 // function whisper.New(string)
 type Model interface {
@ -35,35 +31,29 @@ type Model interface {
 	Languages() []string
 }

-// Context is the speech recognition context.
+// Context is the speach recognition context.
 type Context interface {
 	SetLanguage(string) error // Set the language to use for speech recognition, use "auto" for auto detect language.
 	SetTranslate(bool)        // Set translate flag
 	IsMultilingual() bool     // Return true if the model is multilingual.
 	Language() string         // Get language
-	DetectedLanguage() string // Get detected language

-	SetOffset(time.Duration)          // Set offset
-	SetDuration(time.Duration)        // Set duration
-	SetThreads(uint)                  // Set number of threads to use
-	SetSplitOnWord(bool)              // Set split on word flag
-	SetTokenThreshold(float32)        // Set timestamp token probability threshold
-	SetTokenSumThreshold(float32)     // Set timestamp token sum probability threshold
-	SetMaxSegmentLength(uint)         // Set max segment length in characters
-	SetTokenTimestamps(bool)          // Set token timestamps flag
-	SetMaxTokensPerSegment(uint)      // Set max tokens per segment (0 = no limit)
-	SetAudioCtx(uint)                 // Set audio encoder context
-	SetMaxContext(n int)              // Set maximum number of text context tokens to store
-	SetBeamSize(n int)                // Set Beam Size
-	SetEntropyThold(t float32)        // Set Entropy threshold
-	SetInitialPrompt(prompt string)   // Set initial prompt
-	SetTemperature(t float32)         // Set temperature
-	SetTemperatureFallback(t float32) // Set temperature incrementation
+	SetOffset(time.Duration)        // Set offset
+	SetDuration(time.Duration)      // Set duration
+	SetThreads(uint)                // Set number of threads to use
+	SetSplitOnWord(bool)            // Set split on word flag
+	SetTokenThreshold(float32)      // Set timestamp token probability threshold
+	SetTokenSumThreshold(float32)   // Set timestamp token sum probability threshold
+	SetMaxSegmentLength(uint)       // Set max segment length in characters
+	SetTokenTimestamps(bool)        // Set token timestamps flag
+	SetMaxTokensPerSegment(uint)    // Set max tokens per segment (0 = no limit)
+	SetAudioCtx(uint)               // Set audio encoder context
+	SetInitialPrompt(prompt string) // Set initial prompt

 	// Process mono audio data and return any errors.
 	// If defined, newly generated segments are passed to the
 	// callback function during processing.
-	Process([]float32, EncoderBeginCallback, SegmentCallback, ProgressCallback) error
+	Process([]float32, SegmentCallback, ProgressCallback) error

 	// After process is called, return segments until the end of the stream
 	// is reached, when io.EOF is returned.
--- a/bindings/go/pkg/whisper/model_test.go
+++ b/bindings/go/pkg/whisper/model_test.go
@ -1,91 +0,0 @@
-package whisper_test
-
-import (
-	"testing"
-
-	"github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
-	assert "github.com/stretchr/testify/assert"
-)
-
-func TestNew(t *testing.T) {
-	assert := assert.New(t)
-	t.Run("valid model path", func(t *testing.T) {
-		model, err := whisper.New(ModelPath)
-		assert.NoError(err)
-		assert.NotNil(model)
-		defer model.Close()
-
-	})
-
-	t.Run("invalid model path", func(t *testing.T) {
-		invalidModelPath := "invalid-model-path.bin"
-		model, err := whisper.New(invalidModelPath)
-		assert.Error(err)
-		assert.Nil(model)
-	})
-}
-
-func TestClose(t *testing.T) {
-	assert := assert.New(t)
-
-	model, err := whisper.New(ModelPath)
-	assert.NoError(err)
-	assert.NotNil(model)
-
-	err = model.Close()
-	assert.NoError(err)
-}
-
-func TestNewContext(t *testing.T) {
-	assert := assert.New(t)
-
-	model, err := whisper.New(ModelPath)
-	assert.NoError(err)
-	assert.NotNil(model)
-	defer model.Close()
-
-	context, err := model.NewContext()
-	assert.NoError(err)
-	assert.NotNil(context)
-}
-
-func TestIsMultilingual(t *testing.T) {
-	assert := assert.New(t)
-
-	model, err := whisper.New(ModelPath)
-	assert.NoError(err)
-	assert.NotNil(model)
-	defer model.Close()
-
-	isMultilingual := model.IsMultilingual()
-
-	// This returns false since
-	// the model 'models/ggml-small.en.bin'
-	// that is loaded is not multilingual
-	assert.False(isMultilingual)
-}
-
-func TestLanguages(t *testing.T) {
-	assert := assert.New(t)
-
-	model, err := whisper.New(ModelPath)
-	assert.NoError(err)
-	assert.NotNil(model)
-	defer model.Close()
-
-	expectedLanguages := []string{
-		"en", "zh", "de", "es", "ru", "ko", "fr", "ja", "pt", "tr", "pl",
-		"ca", "nl", "ar", "sv", "it", "id", "hi", "fi", "vi", "he", "uk",
-		"el", "ms", "cs", "ro", "da", "hu", "ta", "no", "th", "ur", "hr",
-		"bg", "lt", "la", "mi", "ml", "cy", "sk", "te", "fa", "lv", "bn",
-		"sr", "az", "sl", "kn", "et", "mk", "br", "eu", "is", "hy", "ne",
-		"mn", "bs", "kk", "sq", "sw", "gl", "mr", "pa", "si", "km", "sn",
-		"yo", "so", "af", "oc", "ka", "be", "tg", "sd", "gu", "am", "yi",
-		"lo", "uz", "fo", "ht", "ps", "tk", "nn", "mt", "sa", "lb", "my",
-		"bo", "tl", "mg", "as", "tt", "haw", "ln", "ha", "ba", "jw", "su",
-	}
-
-	actualLanguages := model.Languages()
-
-	assert.Equal(expectedLanguages, actualLanguages)
-}
--- a/bindings/go/pkg/whisper/util_test.go
+++ b/bindings/go/pkg/whisper/util_test.go
@ -1,6 +0,0 @@
-package whisper_test
-
-const (
-	ModelPath  = "../../models/ggml-small.en.bin"
-	SamplePath = "../../samples/jfk.wav"
-)
--- a/bindings/go/whisper.go
+++ b/bindings/go/whisper.go
@ -9,7 +9,7 @@ import (
 // CGO

 /*
-#cgo LDFLAGS: -lwhisper -lggml -lggml-base -lggml-cpu  -lm -lstdc++ -fopenmp
+#cgo LDFLAGS: -lwhisper -lm -lstdc++
 #cgo darwin LDFLAGS: -framework Accelerate -framework Metal -framework Foundation -framework CoreGraphics
 #include <whisper.h>
 #include <stdlib.h>
--- a/bindings/ios
+++ b/bindings/ios
--- a/bindings/java/README.md
+++ b/bindings/java/README.md
@ -31,10 +31,10 @@ public class Example {
            var whisperParams = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY);
            // custom configuration if required
            whisperParams.temperature_inc = 0f;
-
+            
            var samples = readAudio(); // divide each value by 32767.0f
            whisper.fullTranscribe(whisperParams, samples);
-
+            
            int segmentCount = whisper.getTextSegmentCount(context);
            for (int i = 0; i < segmentCount; i++) {
                String text = whisper.getTextSegment(context, i);
@ -52,7 +52,7 @@ public class Example {
 In order to build, you need to have the JDK 8 or higher installed. Run the tests with:

 ```bash
-git clone https://github.com/ggml-org/whisper.cpp.git
+git clone https://github.com/ggerganov/whisper.cpp.git
 cd whisper.cpp/bindings/java

 ./gradlew build
@ -67,5 +67,5 @@ copy /y ..\..\build\bin\Release\whisper.dll build\generated\resources\main\win32

 ## License

-The license for the Java bindings is the same as the license for the rest of the whisper.cpp project, which is the MIT License. See the `LICENSE` file for more details.
+The license for the Go bindings is the same as the license for the rest of the whisper.cpp project, which is the MIT License. See the `LICENSE` file for more details.

--- a/bindings/java/build.gradle
+++ b/bindings/java/build.gradle
@ -25,43 +25,25 @@ sourceSets {
 }

 tasks.register('copyLibwhisperDynlib', Copy) {
-    from '../../build/src'
-    include 'libwhisper.dylib'
-    into 'build/generated/resources/main'
+    from '../../build'
+    include 'libwhisper.dynlib'
+    into 'build/generated/resources/main/darwin'
 }

 tasks.register('copyLibwhisperSo', Copy) {
-    from '../../build/src'
+    from '../../build'
    include 'libwhisper.so'
-    into 'build/generated/resources/main'
+    into 'build/generated/resources/main/linux-x86-64'
 }

-tasks.register('copyWhisperDLL', Copy) {
-    from '../../build/bin/Release'
+tasks.register('copyWhisperDll', Copy) {
+    from '../../build/Release'
    include 'whisper.dll'
-    into 'build/generated/resources/main'
-}
-
-tasks.register('copyGGML_BASE_DLL', Copy) {
-    from '../../build/bin/Release'
-    include 'ggml-base.dll'
-    into 'build/generated/resources/main'
-}
-
-tasks.register('copyGGML_DLL', Copy) {
-    from '../../build/bin/Release'
-    include 'ggml.dll'
-    into 'build/generated/resources/main'
-}
-
-tasks.register('copyGGML_CPU_DLL', Copy) {
-    from '../../build/bin/Release'
-    include 'ggml-cpu.dll'
-    into 'build/generated/resources/main'
+    into 'build/generated/resources/main/windows-x86-64'
 }

 tasks.register('copyLibs') {
-    dependsOn copyLibwhisperDynlib, copyLibwhisperSo, copyWhisperDLL, copyGGML_BASE_DLL, copyGGML_DLL, copyGGML_CPU_DLL
+    dependsOn copyLibwhisperDynlib, copyLibwhisperSo, copyWhisperDll
 }

 test {
@ -73,12 +55,7 @@ java {
    withJavadocJar()
 }

-sourcesJar() {
-    dependsOn copyLibs
-}
-
 jar {
-    dependsOn copyLibs
    exclude '**/whisper_java.exp', '**/whisper_java.lib'
 }

@ -90,9 +67,6 @@ tasks.withType(Test) {
    useJUnitPlatform()
 }

-test.dependsOn copyLibs
-processResources.dependsOn copyLibs
-
 dependencies {
    implementation "net.java.dev.jna:jna:5.13.0"
    testImplementation "org.junit.jupiter:junit-jupiter:5.9.2"
--- a/bindings/java/gradlew
+++ b/bindings/java/gradlew
--- a/bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperConstants.java
+++ b/bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperConstants.java
@ -1,24 +0,0 @@
-package io.github.ggerganov.whispercpp;
-
-/**
- * Presets for alignment heads in DTW token timestamps
- */
-public class WhisperConstants {
-    // Alignment heads presets
-    public static final int WHISPER_AHEADS_NONE = 0;
-    public static final int WHISPER_AHEADS_TINY_EN = 1;
-    public static final int WHISPER_AHEADS_TINY = 2;
-    public static final int WHISPER_AHEADS_BASE_EN = 3;
-    public static final int WHISPER_AHEADS_BASE = 4;
-    public static final int WHISPER_AHEADS_SMALL_EN = 5;
-    public static final int WHISPER_AHEADS_SMALL = 6;
-    public static final int WHISPER_AHEADS_MEDIUM_EN = 7;
-    public static final int WHISPER_AHEADS_MEDIUM = 8;
-    public static final int WHISPER_AHEADS_LARGE_V1 = 9;
-    public static final int WHISPER_AHEADS_LARGE_V2 = 10;
-    public static final int WHISPER_AHEADS_LARGE_V3 = 11;
-    public static final int WHISPER_AHEADS_LARGE_V3_TURBO = 12;
-    public static final int WHISPER_AHEADS_CUSTOM = 13;
-    public static final int WHISPER_AHEADS_N_TOP_MOST = 14;
-    public static final int WHISPER_AHEADS_COUNT = 15;
-}
--- a/bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperContext.java
+++ b/bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperContext.java
@ -1,9 +1,7 @@
 package io.github.ggerganov.whispercpp;

-import com.sun.jna.NativeLong;
 import com.sun.jna.Structure;
 import com.sun.jna.ptr.PointerByReference;
-import com.sun.jna.Pointer;
 import io.github.ggerganov.whispercpp.ggml.GgmlType;
 import io.github.ggerganov.whispercpp.WhisperModel;
 import io.github.ggerganov.whispercpp.params.WhisperContextParams;
@ -11,26 +9,33 @@ import io.github.ggerganov.whispercpp.params.WhisperContextParams;
 import java.util.List;

 public class WhisperContext extends Structure {
-    public NativeLong t_load_us;
-    public NativeLong t_start_us;
+    int t_load_us = 0;
+    int t_start_us = 0;

    /** weight type (FP32 / FP16 / QX) */
-    public GgmlType wtype = GgmlType.GGML_TYPE_F16;
+    GgmlType wtype = GgmlType.GGML_TYPE_F16;
    /** intermediate type (FP32 or FP16) */
-    public GgmlType itype = GgmlType.GGML_TYPE_F16;
+    GgmlType itype = GgmlType.GGML_TYPE_F16;

-    public WhisperContextParams.ByValue params;
-
-    public Pointer model;
-    public Pointer vocab;
-    public Pointer state;
+//    WhisperModel model;
+    public PointerByReference model;
+//    whisper_vocab vocab;
+//    whisper_state * state = nullptr;
+    public PointerByReference vocab;
+    public PointerByReference state;

    /** populated by whisper_init_from_file_with_params() */
-    public Pointer path_model;
+    String path_model;
+    WhisperContextParams params;

-    @Override
-    protected List<String> getFieldOrder() {
-        return List.of("t_load_us", "t_start_us", "wtype", "itype",
-                "params", "model", "vocab", "state", "path_model");
-    }
+//    public static class ByReference extends WhisperContext implements Structure.ByReference {
+//    }
+//
+//    public static class ByValue extends WhisperContext implements Structure.ByValue {
+//    }
+//
+//    @Override
+//    protected List<String> getFieldOrder() {
+//        return List.of("t_load_us", "t_start_us", "wtype", "itype", "model", "vocab", "state", "path_model");
+//    }
 }
--- a/bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperCpp.java
+++ b/bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperCpp.java
@ -43,11 +43,11 @@ public class WhisperCpp implements AutoCloseable {
     * @param modelPath - absolute path, or just the name (eg: "base", "base-en" or "base.en")
     * @param params - params to use when initialising the context
     */
-    public void initContext(String modelPath, WhisperContextParams.ByValue params) throws FileNotFoundException {
+    public void initContext(String modelPath, WhisperContextParams params) throws FileNotFoundException {
        initContextImpl(modelPath, params);
    }

-    private void initContextImpl(String modelPath, WhisperContextParams.ByValue params) throws FileNotFoundException {
+    private void initContextImpl(String modelPath, WhisperContextParams params) throws FileNotFoundException {
        if (ctx != null) {
            lib.whisper_free(ctx);
        }
@ -69,13 +69,15 @@ public class WhisperCpp implements AutoCloseable {

    /**
     * Provides default params which can be used with `whisper_init_from_file_with_params()` etc.
-     * Returns a ByValue instance to ensure proper parameter passing to native code.
+     * Because this function allocates memory for the params, the caller must call either:
+     * - call `whisper_free_context_params()`
+     * - `Native.free(Pointer.nativeValue(pointer));`
     */
-    public WhisperContextParams.ByValue getContextDefaultParams() {
-        WhisperContextParams.ByValue valueParams = new WhisperContextParams.ByValue(
-            lib.whisper_context_default_params_by_ref());
-        valueParams.read();
-        return valueParams;
+    public WhisperContextParams getContextDefaultParams() {
+        paramsPointer = lib.whisper_context_default_params_by_ref();
+        WhisperContextParams params = new WhisperContextParams(paramsPointer);
+        params.read();
+        return params;
    }
    
    /**
@ -86,7 +88,7 @@ public class WhisperCpp implements AutoCloseable {
     *
     * @param strategy - GREEDY
     */
-    public WhisperFullParams.ByValue getFullDefaultParams(WhisperSamplingStrategy strategy) {
+    public WhisperFullParams getFullDefaultParams(WhisperSamplingStrategy strategy) {
        Pointer pointer;

        // whisper_full_default_params_by_ref allocates memory which we need to delete, so only create max 1 pointer for each strategy.
@ -102,7 +104,7 @@ public class WhisperCpp implements AutoCloseable {
            pointer = beamParamsPointer;
        }

-        WhisperFullParams.ByValue params = new WhisperFullParams.ByValue(pointer);
+        WhisperFullParams params = new WhisperFullParams(pointer);
        params.read();
        return params;
    }
@ -136,21 +138,15 @@ public class WhisperCpp implements AutoCloseable {
    }

    /**
-     * Run the entire model: PCM -&gt; log mel spectrogram -&gt; encoder -&gt; decoder -&gt; text.
+     * Run the entire model: PCM -> log mel spectrogram -> encoder -> decoder -> text.
     * Not thread safe for same context
     * Uses the specified decoding strategy to obtain the text.
     */
-    public String fullTranscribe(WhisperFullParams.ByValue whisperParams, float[] audioData) throws IOException {
+    public String fullTranscribe(WhisperFullParams whisperParams, float[] audioData) throws IOException {
        if (ctx == null) {
            throw new IllegalStateException("Model not initialised");
        }

-        /*
-        WhisperFullParams.ByValue valueParams = new WhisperFullParams.ByValue(
-            lib.whisper_full_default_params_by_ref(WhisperSamplingStrategy.WHISPER_SAMPLING_BEAM_SEARCH.ordinal()));
-        valueParams.read();
-        */
-
        if (lib.whisper_full(ctx, whisperParams, audioData, audioData.length) != 0) {
            throw new IOException("Failed to process audio");
        }
@ -167,16 +163,7 @@ public class WhisperCpp implements AutoCloseable {

        return str.toString().trim();
    }
-
-    /**
-     * Full transcribe with time list.
-     *
-     * @param whisperParams the whisper params
-     * @param audioData     the audio data
-     * @return the list
-     * @throws IOException the io exception
-     */
-    public List<WhisperSegment> fullTranscribeWithTime(WhisperFullParams.ByValue whisperParams, float[] audioData) throws IOException {
+    public List<WhisperSegment> fullTranscribeWithTime(WhisperFullParams whisperParams, float[] audioData) throws IOException {
        if (ctx == null) {
            throw new IllegalStateException("Model not initialised");
        }
@ -188,6 +175,7 @@ public class WhisperCpp implements AutoCloseable {
        int nSegments = lib.whisper_full_n_segments(ctx);
        List<WhisperSegment> segments= new ArrayList<>(nSegments);

+
        for (int i = 0; i < nSegments; i++) {
            long t0 = lib.whisper_full_get_segment_t0(ctx, i);
            String text = lib.whisper_full_get_segment_text(ctx, i);
--- a/bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperCppJnaLibrary.java
+++ b/bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperCppJnaLibrary.java
@ -9,7 +9,6 @@ import io.github.ggerganov.whispercpp.params.WhisperContextParams;
 import io.github.ggerganov.whispercpp.params.WhisperFullParams;

 public interface WhisperCppJnaLibrary extends Library {
-
    WhisperCppJnaLibrary instance = Native.load("whisper", WhisperCppJnaLibrary.class);

    String whisper_print_system_info();
@ -39,7 +38,7 @@ public interface WhisperCppJnaLibrary extends Library {
     * @param params     Pointer to whisper_context_params
     * @return Whisper context on success, null on failure
     */
-    Pointer whisper_init_from_file_with_params(String path_model, WhisperContextParams.ByValue params);
+    Pointer whisper_init_from_file_with_params(String path_model, WhisperContextParams params);

    /**
     * Allocate (almost) all memory needed for the model by loading from a buffer.
@ -181,12 +180,12 @@ public interface WhisperCppJnaLibrary extends Library {
    /**
     * @return the id of the specified language, returns -1 if not found.
     * Examples:
-     *   "de" -&gt; 2
-     *   "german" -&gt; 2
+     *   "de" -> 2
+     *   "german" -> 2
     */
    int whisper_lang_id(String lang);

-    /** @return the short string of the specified language id (e.g. 2 -&gt; "de"), returns nullptr if not found */
+    /** @return the short string of the specified language id (e.g. 2 -> "de"), returns nullptr if not found */
    String whisper_lang_str(int id);

    /**
@ -269,21 +268,20 @@ public interface WhisperCppJnaLibrary extends Library {
    void whisper_free_params(Pointer params);

    /**
-     * Run the entire model: PCM -&gt; log mel spectrogram -&gt; encoder -&gt; decoder -&gt; text
+     * Run the entire model: PCM -> log mel spectrogram -> encoder -> decoder -> text
     * Not thread safe for same context
     * Uses the specified decoding strategy to obtain the text.
     */
-    int whisper_full(Pointer ctx, WhisperFullParams.ByValue params, final float[] samples, int n_samples);
+    int whisper_full(Pointer ctx, WhisperFullParams params, final float[] samples, int n_samples);

-    public int whisper_full_with_state(Pointer ctx, Pointer state, WhisperFullParams.ByValue params, float[] samples, int n_samples);
-    //int whisper_full_with_state(Pointer ctx, Pointer state, WhisperFullParams params, final float[] samples, int n_samples);
+    int whisper_full_with_state(Pointer ctx, Pointer state, WhisperFullParams params, final float[] samples, int n_samples);

    // Split the input audio in chunks and process each chunk separately using whisper_full_with_state()
    // Result is stored in the default state of the context
    // Not thread safe if executed in parallel on the same context.
    // It seems this approach can offer some speedup in some cases.
    // However, the transcription accuracy can be worse at the beginning and end of each chunk.
-    int whisper_full_parallel(Pointer ctx, WhisperFullParams.ByValue params, final float[] samples, int n_samples, int n_processors);
+    int whisper_full_parallel(Pointer ctx, WhisperFullParams params, final float[] samples, int n_samples, int n_processors);

    /**
     * Number of generated text segments.
--- a/bindings/java/src/main/java/io/github/ggerganov/whispercpp/callbacks/GgmlAbortCallback.java
+++ b/bindings/java/src/main/java/io/github/ggerganov/whispercpp/callbacks/GgmlAbortCallback.java
@ -1,17 +0,0 @@
-package io.github.ggerganov.whispercpp.callbacks;
-
-import com.sun.jna.Callback;
-
-/**
- * Callback for aborting GGML computation
- * Maps to the C typedef: bool (*ggml_abort_callback)(void * data)
- */
-public interface GgmlAbortCallback extends Callback {
-    /**
-     * Return true to abort the computation, false to continue
-     *
-     * @param data User data passed to the callback
-     * @return true to abort, false to continue
-     */
-    boolean invoke(com.sun.jna.Pointer data);
-}
--- a/bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperAhead.java
+++ b/bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperAhead.java
@ -1,30 +0,0 @@
-package io.github.ggerganov.whispercpp.params;
-import com.sun.jna.*;
-import java.util.Arrays;
-import java.util.List;
-
-public class WhisperAhead extends Structure {
-
-    public int n_text_layer;
-
-    public int n_head;
-
-    public WhisperAhead() {
-        super();
-    }
-
-    public WhisperAhead(int textLayer, int head) {
-        super();
-        this.n_text_layer = textLayer;
-        this.n_head = head;
-    }
-
-    @Override
-    protected List<String> getFieldOrder() {
-        return Arrays.asList("n_text_layer", "n_head");
-    }
-
-    public static class ByReference extends WhisperAhead implements Structure.ByReference {}
-
-    public static class ByValue extends WhisperAhead implements Structure.ByValue {}
-}
--- a/bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperAheads.java
+++ b/bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperAheads.java
@ -1,41 +0,0 @@
-package io.github.ggerganov.whispercpp.params;
-import com.sun.jna.*;
-import java.util.Arrays;
-import java.util.List;
-
-public class WhisperAheads extends Structure {
-    public NativeLong n_heads;
-
-    public Pointer heads;
-
-    public WhisperAheads() {
-        super();
-    }
-
-    /**
-     * Create alignment heads from an array of WhisperAhead objects
-     */
-    public void setHeads(WhisperAhead[] aheadsArray) {
-        this.n_heads = new NativeLong(aheadsArray.length);
-
-        int structSize = aheadsArray[0].size();
-        Memory mem = new Memory(structSize * aheadsArray.length);
-
-        for (int i = 0; i < aheadsArray.length; i++) {
-            aheadsArray[i].write();
-            byte[] buffer = aheadsArray[i].getPointer().getByteArray(0, structSize);
-            mem.write(i * structSize, buffer, 0, buffer.length);
-        }
-
-        this.heads = mem;
-    }
-
-    @Override
-    protected List<String> getFieldOrder() {
-        return Arrays.asList("n_heads", "heads");
-    }
-
-    public static class ByReference extends WhisperAheads implements Structure.ByReference {}
-
-    public static class ByValue extends WhisperAheads implements Structure.ByValue {}
-}
--- a/bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperContextParams.java
+++ b/bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperContextParams.java
@ -1,5 +1,7 @@
 package io.github.ggerganov.whispercpp.params;
+
 import com.sun.jna.*;
+
 import java.util.Arrays;
 import java.util.List;

@ -9,73 +11,21 @@ import java.util.List;
 * whisper_context_default_params()
 */
 public class WhisperContextParams extends Structure {
+
    public WhisperContextParams(Pointer p) {
        super(p);
    }

-    public WhisperContextParams() {
-        super();
-    }
-
-    /** Use GPU for inference (default = true) */
+    /** Use GPU for inference Number (default = true) */
    public CBool use_gpu;

-    /** Use flash attention (default = false) */
-    public CBool flash_attn;
-
-    /** CUDA device to use (default = 0) */
-    public int gpu_device;
-
-    /** [EXPERIMENTAL] Enable token-level timestamps with DTW (default = false) */
-    public CBool dtw_token_timestamps;
-
-    /** [EXPERIMENTAL] Alignment heads preset for DTW */
-    public int dtw_aheads_preset;
-
-    /** Number of top layers to use for DTW when using WHISPER_AHEADS_N_TOP_MOST preset */
-    public int dtw_n_top;
-
-    public WhisperAheads.ByValue dtw_aheads;
-
-    /** DTW memory size (internal use) */
-    public NativeLong dtw_mem_size;
-
-    /** Use GPU for inference */
+    /** Use GPU for inference Number (default = true) */
    public void useGpu(boolean enable) {
        use_gpu = enable ? CBool.TRUE : CBool.FALSE;
    }

-    /** Use flash attention */
-    public void useFlashAttn(boolean enable) {
-        flash_attn = enable ? CBool.TRUE : CBool.FALSE;
-    }
-
-    /** Enable DTW token-level timestamps */
-    public void enableDtwTokenTimestamps(boolean enable) {
-        dtw_token_timestamps = enable ? CBool.TRUE : CBool.FALSE;
-    }
-
-    /** Set DTW alignment heads preset */
-    public void setDtwAheadsPreset(int preset) {
-        dtw_aheads_preset = preset;
-    }
-
    @Override
    protected List<String> getFieldOrder() {
-        return Arrays.asList(
-            "use_gpu",
-            "flash_attn",
-            "gpu_device",
-            "dtw_token_timestamps",
-            "dtw_aheads_preset",
-            "dtw_n_top",
-            "dtw_aheads",
-            "dtw_mem_size"
-        );
-    }
-
-    public static class ByValue extends WhisperContextParams implements Structure.ByValue {
-        public ByValue() { super(); }
-        public ByValue(Pointer p) { super(p); }
+        return Arrays.asList("use_gpu");
    }
 }
--- a/bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperFullParams.java
+++ b/bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperFullParams.java
@ -5,7 +5,6 @@ import io.github.ggerganov.whispercpp.callbacks.WhisperEncoderBeginCallback;
 import io.github.ggerganov.whispercpp.callbacks.WhisperLogitsFilterCallback;
 import io.github.ggerganov.whispercpp.callbacks.WhisperNewSegmentCallback;
 import io.github.ggerganov.whispercpp.callbacks.WhisperProgressCallback;
-import io.github.ggerganov.whispercpp.callbacks.GgmlAbortCallback;

 import java.util.Arrays;
 import java.util.List;
@ -17,12 +16,10 @@ import java.util.List;
 */
 public class WhisperFullParams extends Structure {

-    public WhisperFullParams() {
-        super();
-    }
-
    public WhisperFullParams(Pointer p) {
        super(p);
+//        super(p, ALIGN_MSVC);
+//        super(p, ALIGN_GNUC);
    }

    /** Sampling strategy for whisper_full() function. */
@ -72,10 +69,10 @@ public class WhisperFullParams extends Structure {
        single_segment = single ? CBool.TRUE : CBool.FALSE;
    }

-    /** Flag to print special tokens (e.g., &lt;SOT&gt;, &lt;EOT&gt;, &lt;BEG&gt;, etc.). (default = false) */
+    /** Flag to print special tokens (e.g., &lt;SOT>, &lt;EOT>, &lt;BEG>, etc.). (default = false) */
    public CBool print_special;

-    /** Flag to print special tokens (e.g., &lt;SOT&gt;, &lt;EOT&gt;, &lt;BEG&gt;, etc.). (default = false) */
+    /** Flag to print special tokens (e.g., &lt;SOT>, &lt;EOT>, &lt;BEG>, etc.). (default = false) */
    public void printSpecial(boolean enable) {
        print_special = enable ? CBool.TRUE : CBool.FALSE;
    }
@ -132,14 +129,6 @@ public class WhisperFullParams extends Structure {
    /** Maximum tokens per segment (0, default = no limit) */
    public int max_tokens;

-    /** [EXPERIMENTAL] Enable debug mode for extra info */
-    public CBool debug_mode;
-
-    /** Enable debug mode */
-    public void enableDebugMode(boolean enable) {
-        debug_mode = enable ? CBool.TRUE : CBool.FALSE;
-    }
-
    /** Overwrite the audio context size (0 = use default). */
    public int audio_ctx;

@ -192,11 +181,11 @@ public class WhisperFullParams extends Structure {
    }

    /** Flag to suppress non-speech tokens. */
-    public CBool suppress_nst;
+    public CBool suppress_non_speech_tokens;

    /** Flag to suppress non-speech tokens. */
    public void suppressNonSpeechTokens(boolean enable) {
-        suppress_nst = enable ? CBool.TRUE : CBool.FALSE;
+        suppress_non_speech_tokens = enable ? CBool.TRUE : CBool.FALSE;
    }

    /** Initial decoding temperature. */
@ -285,16 +274,6 @@ public class WhisperFullParams extends Structure {
     */
    public Pointer encoder_begin_callback_user_data;

-    /** Callback used to abort GGML computation */
-    public Pointer abort_callback;
-
-    /** User data for the abort_callback */
-    public Pointer abort_callback_user_data;
-
-    public void setAbortCallback(GgmlAbortCallback callback) {
-        abort_callback = CallbackReference.getFunctionPointer(callback);
-    }
-
    /**
     * Callback by each decoder to filter obtained logits.
     * WhisperLogitsFilterCallback
@ -331,28 +310,17 @@ public class WhisperFullParams extends Structure {

    @Override
    protected List<String> getFieldOrder() {
-        return Arrays.asList("strategy", "n_threads", "n_max_text_ctx",
-                "offset_ms", "duration_ms", "translate", "no_context",
-                "no_timestamps", "single_segment", "print_special",
-                "print_progress", "print_realtime", "print_timestamps",
-                "token_timestamps", "thold_pt", "thold_ptsum", "max_len",
-                "split_on_word", "max_tokens", "debug_mode", "audio_ctx", 
-                "tdrz_enable", "suppress_regex", "initial_prompt",
-                "prompt_tokens", "prompt_n_tokens", "language", "detect_language",
-                "suppress_blank", "suppress_nst", "temperature",
-                "max_initial_ts", "length_penalty", "temperature_inc",
-                "entropy_thold", "logprob_thold", "no_speech_thold", "greedy",
-                "beam_search", "new_segment_callback", "new_segment_callback_user_data",
+        return Arrays.asList("strategy", "n_threads", "n_max_text_ctx", "offset_ms", "duration_ms", "translate",
+                "no_context", "single_segment", "no_timestamps",
+                "print_special", "print_progress", "print_realtime", "print_timestamps",  "token_timestamps",
+                "thold_pt", "thold_ptsum", "max_len", "split_on_word", "max_tokens", "audio_ctx",
+                "tdrz_enable", "suppress_regex", "initial_prompt", "prompt_tokens", "prompt_n_tokens", "language", "detect_language",
+                "suppress_blank", "suppress_non_speech_tokens", "temperature", "max_initial_ts", "length_penalty",
+                "temperature_inc", "entropy_thold", "logprob_thold", "no_speech_thold", "greedy", "beam_search",
+                "new_segment_callback", "new_segment_callback_user_data",
                "progress_callback", "progress_callback_user_data",
                "encoder_begin_callback", "encoder_begin_callback_user_data",
-                "abort_callback", "abort_callback_user_data",
                "logits_filter_callback", "logits_filter_callback_user_data",
                "grammar_rules", "n_grammar_rules", "i_start_rule", "grammar_penalty");
    }
-
-    public static class ByValue extends WhisperFullParams implements Structure.ByValue {
-        public ByValue() { super(); }
-        public ByValue(Pointer p) { super(p); }
-    }
-
 }
--- a/bindings/java/src/test/java/io/github/ggerganov/whispercpp/WhisperCppTest.java
+++ b/bindings/java/src/test/java/io/github/ggerganov/whispercpp/WhisperCppTest.java
@ -76,7 +76,7 @@ class WhisperCppTest {
        float[] floats = new float[b.length / 2];

        //WhisperFullParams params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY);
-        WhisperFullParams.ByValue params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_BEAM_SEARCH);
+        WhisperFullParams params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_BEAM_SEARCH);
        params.setProgressCallback((ctx, state, progress, user_data) -> System.out.println("progress: " + progress));
        params.print_progress = CBool.FALSE;
        //params.initial_prompt = "and so my fellow Americans um, like";
@ -118,7 +118,7 @@ class WhisperCppTest {
        float[] floats = new float[b.length / 2];

        //WhisperFullParams params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY);
-        WhisperFullParams.ByValue params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_BEAM_SEARCH);
+        WhisperFullParams params = whisper.getFullDefaultParams(WhisperSamplingStrategy.WHISPER_SAMPLING_BEAM_SEARCH);
        params.setProgressCallback((ctx, state, progress, user_data) -> System.out.println("progress: " + progress));
        params.print_progress = CBool.FALSE;
        //params.initial_prompt = "and so my fellow Americans um, like";
--- a/bindings/javascript/README.md
+++ b/bindings/javascript/README.md
@ -33,9 +33,6 @@ mkdir build-em && cd build-em
 emcmake cmake .. && make -j

 # run test
-node ../tests/test-whisper.js
-
-# For Node.js versions prior to v16.4.0, experimental features need to be enabled:
 node --experimental-wasm-threads --experimental-wasm-simd ../tests/test-whisper.js

 # publish npm package
--- a/bindings/javascript/package.json
+++ b/bindings/javascript/package.json
@ -1,6 +1,6 @@
 {
  "name": "whisper.cpp",
-  "version": "1.7.5",
+  "version": "1.6.2",
  "description": "Whisper speech recognition",
  "main": "whisper.js",
  "scripts": {
--- a/bindings/ruby/.gitignore
+++ b/bindings/ruby/.gitignore
@ -1,9 +0,0 @@
-LICENSE
-pkg/
-lib/whisper.*
-ext/examples/
-ext/ggml/
-ext/include/
-ext/scripts/
-ext/src/
-test/fixtures/
--- a/bindings/ruby/README.md
+++ b/bindings/ruby/README.md
@ -1,349 +0,0 @@
-whispercpp
-==========
-
-![whisper.cpp](https://user-images.githubusercontent.com/1991296/235238348-05d0f6a4-da44-4900-a1de-d0707e75b763.jpeg)
-
-Ruby bindings for [whisper.cpp][], an interface of automatic speech recognition model.
-
-Installation
------------
-
-Install the gem and add to the application's Gemfile by executing:
-
-    $ bundle add whispercpp
-
-If bundler is not being used to manage dependencies, install the gem by executing:
-
-    $ gem install whispercpp
-
-You can pass build options for whisper.cpp, for instance:
-
-    $ bundle config build.whispercpp --enable-ggml-cuda
-
-or,
-
-    $ gem install whispercpp -- --enable-ggml-cuda
-
-See whisper.cpp's [README](https://github.com/ggml-org/whisper.cpp/blob/master/README.md) for available options. You need convert options present the README to Ruby-style options, for example:
-
-Boolean options:
-
-* `-DGGML_BLAS=1` -> `--enable-ggml-blas`
-* `-DWHISER_COREML=OFF` -> `--disable-whisper-coreml`
-
-Argument options:
-
-* `-DGGML_CUDA_COMPRESSION_MODE=size` -> `--ggml-cuda-compression-mode=size`
-
-Combination:
-
-* `-DGGML_CUDA=1 -DCMAKE_CUDA_ARCHITECTURES="86"` -> `--enable-ggml-cuda --cmake_cuda-architectures="86"`
-
-For boolean options like `GGML_CUDA`, the README says `-DGGML_CUDA=1`. You need strip `-D`, prepend `--enable-` for `1` or `ON` (`--disable-` for `0` or `OFF`) and make it kebab-case: `--enable-ggml-cuda`.  
-For options which require arguments like `CMAKE_CUDA_ARCHITECTURES`, the README says `-DCMAKE_CUDA_ARCHITECTURES="86"`. You need strip `-D`, prepend `--`, make it kebab-case, append `=` and append argument: `--cmake-cuda-architectures="86"`.
-
-Usage
-----
-
-```ruby
-require "whisper"
-
-whisper = Whisper::Context.new("base")
-
-params = Whisper::Params.new(
-  language: "en",
-  offset: 10_000,
-  duration: 60_000,
-  max_text_tokens: 300,
-  translate: true,
-  print_timestamps: false,
-  initial_prompt: "Initial prompt here."
-)
-
-whisper.transcribe("path/to/audio.wav", params) do |whole_text|
-  puts whole_text
-end
-
-```
-
-### Preparing model ###
-
-Some models are prepared up-front:
-
-You also can use shorthand for pre-converted models:
-
-```ruby
-whisper = Whisper::Context.new("base.en")
-```
-
-You can see the list of prepared model names by `Whisper::Model.pre_converted_models.keys`:
-
-```ruby
-puts Whisper::Model.pre_converted_models.keys
-# tiny
-# tiny.en
-# tiny-q5_1
-# tiny.en-q5_1
-# tiny-q8_0
-# base
-# base.en
-# base-q5_1
-# base.en-q5_1
-# base-q8_0
-#   :
-#   :
-```
-
-You can also retrieve each model:
-
-```ruby
-base_en = Whisper::Model.pre_converted_models["base.en"]
-whisper = Whisper::Context.new(base_en)
-```
-
-At first time you use a model, it is downloaded automatically. After that, downloaded cached file is used. To clear cache, call `#clear_cache`:
-
-```ruby
-Whisper::Model.pre_converted_models["base"].clear_cache
-```
-
-You can also use local model files you prepared:
-
-```ruby
-whisper = Whisper::Context.new("path/to/your/model.bin")
-```
-
-Or, you can download model files:
-
-```ruby
-whisper = Whisper::Context.new("https://example.net/uri/of/your/model.bin")
-# Or
-whisper = Whisper::Context.new(URI("https://example.net/uri/of/your/model.bin"))
-```
-
-See [models][] page for details.
-
-### Preparing audio file ###
-
-Currently, whisper.cpp accepts only 16-bit WAV files.
-
-### Voice Activity Detection (VAD) ###
-
-Support for Voice Activity Detection (VAD) can be enabled by setting `Whisper::Params`'s `vad` argument to `true` and specifying VAD model:
-
-```ruby
-Whisper::Params.new(
-  vad: true,
-  vad_model_path: "silero-v5.1.2",
-  # other arguments...
-)
-```
-
-When you pass the model name (`"silero-v5.1.2"`) or URI (`https://huggingface.co/ggml-org/whisper-vad/resolve/main/ggml-silero-v5.1.2.bin`), it will be downloaded automatically.
-Currently, "silero-v5.1.2" is registered as pre-converted model like ASR models. You also specify file path or URI of model.
-
-If you need configure VAD behavior, pass params for that:
-
-```ruby
-Whisper::Params.new(
-  vad: true,
-  vad_model_path: "silero-v5.1.2",
-  vad_params: Whisper::VAD::Params.new(
-    threshold: 1.0, # defaults to 0.5
-    min_speech_duration_ms: 500, # defaults to 250
-    min_silence_duration_ms: 200, # defaults to 100
-    max_speech_duration_s: 30000, # default is FLT_MAX,
-    speech_pad_ms: 50, # defaults to 30
-    samples_overlap: 0.5 # defaults to 0.1
-  ),
-  # other arguments...
-)
-```
-
-For details on VAD, see [whisper.cpp's README](https://github.com/ggml-org/whisper.cpp?tab=readme-ov-file#voice-activity-detection-vad).
-
-### Output ###
-
-whispercpp supports SRT and WebVTT output:
-
-```ruby
-puts whisper.transcribe("path/to/audio.wav", Whisper::Params.new).to_webvtt
-# =>
-WEBVTT
-
-1
-00:00:00.000 --> 00:00:03.860
- My thought I have nobody by a beauty and will as you poured.
-
-2
-00:00:03.860 --> 00:00:09.840
- Mr. Rochester is sub in that so-don't find simplest, and devoted about, to let might in
-
-3
-00:00:09.840 --> 00:00:09.940
- a
-
-```
-
-You may call `#to_srt`, too
-
-
-API
---
-
-### Transcription ###
-
-By default, `Whisper::Context#transcribe` works in a single thread. You can make it work in parallel by passing `n_processors` option:
-
-```ruby
-whisper.transcribe("path/to/audio.wav", params, n_processors: Etc.nprocessors)
-```
-
-Note that transcription occasionally might be low accuracy when it works in parallel.
-
-### Segments ###
-
-Once `Whisper::Context#transcribe` called, you can retrieve segments by `#each_segment`:
-
-```ruby
-def format_time(time_ms)
-  sec, decimal_part = time_ms.divmod(1000)
-  min, sec = sec.divmod(60)
-  hour, min = min.divmod(60)
-  "%02d:%02d:%02d.%03d" % [hour, min, sec, decimal_part]
-end
-
-whisper
-  .transcribe("path/to/audio.wav", params)
-  .each_segment.with_index do |segment, index|
-    line = "[%{nth}: %{st} --> %{ed}] %{text}" % {
-      nth: index + 1,
-      st: format_time(segment.start_time),
-      ed: format_time(segment.end_time),
-      text: segment.text
-    }
-    line << " (speaker turned)" if segment.speaker_turn_next?
-    puts line
-  end
-
-```
-
-You can also add hook to params called on new segment:
-
-```ruby
-# Add hook before calling #transcribe
-params.on_new_segment do |segment|
-  line = "[%{st} --> %{ed}] %{text}" % {
-    st: format_time(segment.start_time),
-    ed: format_time(segment.end_time),
-    text: segment.text
-  }
-  line << " (speaker turned)" if segment.speaker_turn_next?
-  puts line
-end
-
-whisper.transcribe("path/to/audio.wav", params)
-
-```
-
-### Models ###
-
-You can see model information:
-
-```ruby
-whisper = Whisper::Context.new("base")
-model = whisper.model
-
-model.n_vocab # => 51864
-model.n_audio_ctx # => 1500
-model.n_audio_state # => 512
-model.n_audio_head # => 8
-model.n_audio_layer # => 6
-model.n_text_ctx # => 448
-model.n_text_state # => 512
-model.n_text_head # => 8
-model.n_text_layer # => 6
-model.n_mels # => 80
-model.ftype # => 1
-model.type # => "base"
-
-```
-
-### Logging ###
-
-You can set log callback:
-
-```ruby
-prefix = "[MyApp] "
-log_callback = ->(level, buffer, user_data) {
-  case level
-  when Whisper::LOG_LEVEL_NONE
-    puts "#{user_data}none: #{buffer}"
-  when Whisper::LOG_LEVEL_INFO
-    puts "#{user_data}info: #{buffer}"
-  when Whisper::LOG_LEVEL_WARN
-    puts "#{user_data}warn: #{buffer}"
-  when Whisper::LOG_LEVEL_ERROR
-    puts "#{user_data}error: #{buffer}"
-  when Whisper::LOG_LEVEL_DEBUG
-    puts "#{user_data}debug: #{buffer}"
-  when Whisper::LOG_LEVEL_CONT
-    puts "#{user_data}same to previous: #{buffer}"
-  end
-}
-Whisper.log_set log_callback, prefix
-```
-
-Using this feature, you are also able to suppress log:
-
-```ruby
-Whisper.log_set ->(level, buffer, user_data) {
-  # do nothing
-}, nil
-Whisper::Context.new("base")
-```
-
-### Low-level API to transcribe ###
-
-You can also call `Whisper::Context#full` and `#full_parallel` with a Ruby array as samples. Although `#transcribe` with audio file path is recommended because it extracts PCM samples in C++ and is fast, `#full` and `#full_parallel` give you flexibility.
-
-```ruby
-require "whisper"
-require "wavefile"
-
-reader = WaveFile::Reader.new("path/to/audio.wav", WaveFile::Format.new(:mono, :float, 16000))
-samples = reader.enum_for(:each_buffer).map(&:samples).flatten
-
-whisper = Whisper::Context.new("base")
-whisper
-  .full(Whisper::Params.new, samples)
-  .each_segment do |segment|
-    puts segment.text
-  end
-```
-
-The second argument `samples` may be an array, an object with `length` and `each` method, or a MemoryView. If you can prepare audio data as C array and export it as a MemoryView, whispercpp accepts and works with it with zero copy.
-
-Development
-----------
-
-    % git clone https://github.com/ggml-org/whisper.cpp.git
-    % cd whisper.cpp/bindings/ruby
-    % rake test
-
-First call of `rake test` builds an extension and downloads a model for testing. After that, you add tests in `tests` directory and modify `ext/ruby_whisper.cpp`.
-
-If something seems wrong on build, running `rake clean` solves some cases.
-
-### Need help ###
-
-* Windows support
-* Refinement of C/C++ code, especially memory management
-
-License
-------
-
-The same to [whisper.cpp][].
-
-[whisper.cpp]: https://github.com/ggml-org/whisper.cpp
-[models]: https://github.com/ggml-org/whisper.cpp/tree/master/models
--- a/bindings/ruby/Rakefile
+++ b/bindings/ruby/Rakefile
@ -1,96 +1,12 @@
 require 'rake/clean'
-require "bundler/gem_tasks"
-require "rake/testtask"
-require_relative "extsources"
+  require 'rubygems/package'

-SOURCES_DIR = "ext/sources"
-
-SOURCES = FileList[]
-
-EXTSOURCES.each do |src|
-  basename = src.pathmap("%f")
-  dest = basename == "LICENSE" ? basename
-                               : src.pathmap("%{\\.\\./\\.\\.,#{SOURCES_DIR}}p")
-                                    .pathmap("%{\\.\\./javascript,#{SOURCES_DIR}/bindings/javascript}p")
-  dir = dest.pathmap("%d")
-  file src
-  directory dir
-  file dest => [src, dir] do |t|
-    cp t.source, t.name
-  end
-  SOURCES.include dest
+desc 'Build gem'
+task :package do
+  spec_source = File.read File.join(File.dirname(__FILE__),'whispercpp.gemspec')
+  spec = nil
+  # see: http://gist.github.com/16215
+  Thread.new { spec = eval("#{spec_source}") }.join
+  spec.validate
+  Gem::Package.build(spec)
 end
-
-CLEAN.include SOURCES
-
-SRC = FileList["ext/*.{c,cpp,h}"]
-
-task build: SOURCES
-
-directory "pkg"
-CLOBBER.include "pkg"
-
-LIB_NAME = "whisper".ext(RbConfig::CONFIG["DLEXT"])
-SO_FILE = File.join("ext", LIB_NAME)
-LIB_FILE = File.join("lib", LIB_NAME)
-
-file "ext/Makefile" => SRC + ["ext/extconf.rb"] + SOURCES do |t|
-  chdir "ext" do
-    ruby "extconf.rb"
-  end
-end
-if File.exist? "ext/Makefile"
-  task :make_clean do
-    cd "ext" do
-      sh "make", "clean"
-    end
-  end
-  task clean: :make_clean
-  task :make_distclean do
-    cd "ext" do
-      sh "make", "distclean"
-    end
-  end
-  task clobber: :make_distclean
-end
-
-file SO_FILE => "ext/Makefile" do |t|
-  chdir "ext" do
-    sh "make"
-  end
-end
-CLEAN.include SO_FILE
-
-directory "lib"
-file LIB_FILE => [SO_FILE, "lib"] do |t|
-  copy t.source, t.name
-end
-CLEAN.include LIB_FILE
-
-Rake::TestTask.new
-
-TEST_FIXTURE_AUDIO = "test/fixtures/jfk.wav"
-TEST_FIXTURE_AUDIO_SRC = File.expand_path(File.join(__dir__, "..", "..", "samples", "jfk.wav"))
-TEST_FIXTURE_AUDIO_DIR = TEST_FIXTURE_AUDIO.pathmap("%d")
-directory TEST_FIXTURE_AUDIO_DIR
-if File.exist? TEST_FIXTURE_AUDIO_SRC
-  file TEST_FIXTURE_AUDIO => [TEST_FIXTURE_AUDIO_SRC, TEST_FIXTURE_AUDIO_DIR] do |t|
-    symlink t.source, t.name
-  end
-else
-  require "open-uri"
-  file TEST_FIXTURE_AUDIO => TEST_FIXTURE_AUDIO_DIR do |t|
-    File.write t.name, URI("https://github.com/ggml-org/whisper.cpp/raw/refs/heads/master/samples/jfk.wav").read
-  end
-end
-
-TEST_MEMORY_VIEW = "test/jfk_reader/jfk_reader.#{RbConfig::CONFIG['DLEXT']}"
-file TEST_MEMORY_VIEW => "test/jfk_reader/jfk_reader.c" do |t|
-  chdir "test/jfk_reader" do
-    ruby "extconf.rb"
-    sh "make"
-  end
-end
-CLEAN.include TEST_MEMORY_VIEW
-
-task test: [LIB_FILE, TEST_MEMORY_VIEW, TEST_FIXTURE_AUDIO]
--- a/bindings/ruby/ext/.gitignore
+++ b/bindings/ruby/ext/.gitignore
@ -1,9 +1,9 @@
 Makefile
-whisper.so
+ggml.c
+ggml.h
+ggml-alloc.c
+ggml-alloc.h
 whisper.bundle
-whisper.dll
-*.o
-*.a
-sources/*
-!sources/CMakeGraphVizOptions.cmake
-mkmf.log
+whisper.cpp
+whisper.h
+dr_wav.h
--- a/bindings/ruby/ext/dependencies.rb
+++ b/bindings/ruby/ext/dependencies.rb
@ -1,73 +0,0 @@
-require "tsort"
-
-class Dependencies
-  include TSort
-
-  def initialize(cmake, options)
-    @cmake = cmake
-    @options = options
-    @static_lib_shape = nil
-    @nodes = {}
-    @graph = Hash.new {|h, k| h[k] = []}
-
-    generate_dot
-    parse_dot
-  end
-
-  def libs
-    tsort.filter_map {|node|
-      label, shape = @nodes[node]
-      if shape == @static_lib_shape
-        label.gsub(/\\n\([^)]+\)/, '')
-      else
-        nil
-      end
-    }.reverse.collect {|lib| "lib#{lib}.a"}
-  end
-
-  def to_s
-    libs.join(" ")
-  end
-
-  private
-
-  def dot_path
-    File.join(__dir__, "build", "whisper.cpp.dot")
-  end
-
-  def generate_dot
-    args = ["-S", "sources", "-B", "build", "--graphviz", dot_path, "-D", "BUILD_SHARED_LIBS=OFF"]
-    args << @options.to_s unless @options.to_s.empty?
-    system @cmake, *args, exception: true
-  end
-
-  def parse_dot
-    File.open(dot_path).each_line do |line|
-      case line
-      when /\[\s*label\s*=\s*"Static Library"\s*,\s*shape\s*=\s*(?<shape>\w+)\s*\]/
-        @static_lib_shape = $~[:shape]
-      when /\A\s*"(?<node>\w+)"\s*\[\s*label\s*=\s*"(?<label>\S+)"\s*,\s*shape\s*=\s*(?<shape>\w+)\s*\]\s*;\s*\z/
-        node = $~[:node]
-        label = $~[:label]
-        shape = $~[:shape]
-        @nodes[node] = [label, shape]
-      when /\A\s*"(?<depender>\w+)"\s*->\s*"(?<dependee>\w+)"/
-        depender = $~[:depender]
-        dependee = $~[:dependee]
-        @graph[depender] << dependee
-      end
-    end
-  end
-
-  def tsort_each_node
-    @nodes.each_key do |node|
-      yield node
-    end
-  end
-
-  def tsort_each_child(node)
-    @graph[node].each do |child|
-      yield child
-    end
-  end
-end
--- a/bindings/ruby/ext/extconf.rb
+++ b/bindings/ruby/ext/extconf.rb
@ -1,22 +1,31 @@
-require "mkmf"
-require_relative "options"
-require_relative "dependencies"
+require 'mkmf'
+system("cp #{File.join(File.dirname(__FILE__),'..','..','..','whisper.cpp')} .")
+system("cp #{File.join(File.dirname(__FILE__),'..','..','..','whisper.h')} .")
+system("cp #{File.join(File.dirname(__FILE__),'..','..','..','whisper-mel.hpp')} .")
+system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml.h')} .")
+system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml.c')} .")
+system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml-impl.h')} .")
+system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml-alloc.h')} .")
+system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml-alloc.c')} .")
+system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml-backend-impl.h')} .")
+system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml-backend.h')} .")
+system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml-backend.c')} .")
+system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml-common.h')} .")
+system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml-quants.h')} .")
+system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml-quants.c')} .")
+system("cp #{File.join(File.dirname(__FILE__),'..','..','..','examples','dr_wav.h')} .")

-cmake = find_executable("cmake") || abort
-options = Options.new(cmake)
-have_library("gomp") rescue nil
-libs = Dependencies.new(cmake, options)

-$INCFLAGS << " -Isources/include -Isources/ggml/include -Isources/examples"
-$LOCAL_LIBS << " #{libs}"
-$cleanfiles << " build #{libs}"
-
-create_makefile "whisper" do |conf|
-  conf << <<~EOF
-    $(TARGET_SO): #{libs}
-    #{libs}: cmake-targets
-    cmake-targets:
-    #{"\t"}#{cmake} -S sources -B build -D BUILD_SHARED_LIBS=OFF -D CMAKE_ARCHIVE_OUTPUT_DIRECTORY=#{__dir__} -D CMAKE_POSITION_INDEPENDENT_CODE=ON #{options}
-    #{"\t"}#{cmake} --build build --config Release --target common whisper
-  EOF
+# need to use c++ compiler flags
+$CXXFLAGS << ' -std=c++11'
+# Set to true when building binary gems
+if enable_config('static-stdlib', false)
+  $LDFLAGS << ' -static-libgcc -static-libstdc++'
 end
+
+if enable_config('march-tune-native', false)
+  $CFLAGS << ' -march=native -mtune=native'
+  $CXXFLAGS << ' -march=native -mtune=native'
+end
+
+create_makefile('whisper')
--- a/bindings/ruby/ext/ggml-backend-impl.h
+++ b/bindings/ruby/ext/ggml-backend-impl.h
@ -0,0 +1,141 @@
+#pragma once
+
+// ggml-backend internal header
+
+#include "ggml-backend.h"
+
+#ifdef  __cplusplus
+extern "C" {
+#endif
+
+    //
+    // Backend buffer
+    //
+
+    // buffer type
+    typedef void * ggml_backend_buffer_type_context_t;
+
+    struct ggml_backend_buffer_type_i {
+        const char *          (*GGML_CALL get_name)        (ggml_backend_buffer_type_t buft);
+        ggml_backend_buffer_t (*GGML_CALL alloc_buffer)    (ggml_backend_buffer_type_t buft, size_t size);
+        size_t                (*GGML_CALL get_alignment)   (ggml_backend_buffer_type_t buft); // tensor alignment
+        size_t                (*GGML_CALL get_max_size)    (ggml_backend_buffer_type_t buft); // allocation max size
+        size_t                (*GGML_CALL get_alloc_size)  (ggml_backend_buffer_type_t buft, const struct ggml_tensor * tensor); // data size needed to allocate the tensor, including padding
+        bool                  (*GGML_CALL supports_backend)(ggml_backend_buffer_type_t buft, ggml_backend_t backend); // check if the buffer type is usable by the backend
+        // check if tensor data is in host memory
+        // should be equivalent to supports_backend(buft, ggml_backend_cpu_init())
+        bool                  (*GGML_CALL is_host)         (ggml_backend_buffer_type_t buft);
+    };
+
+    struct ggml_backend_buffer_type {
+        struct ggml_backend_buffer_type_i  iface;
+        ggml_backend_buffer_type_context_t context;
+    };
+
+    // buffer
+    typedef void * ggml_backend_buffer_context_t;
+
+    struct ggml_backend_buffer_i {
+        const char * (*GGML_CALL get_name)   (ggml_backend_buffer_t buffer);
+        void         (*GGML_CALL free_buffer)(ggml_backend_buffer_t buffer);
+        void *       (*GGML_CALL get_base)   (ggml_backend_buffer_t buffer);
+        void         (*GGML_CALL init_tensor)(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
+        void         (*GGML_CALL set_tensor) (ggml_backend_buffer_t buffer,       struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
+        void         (*GGML_CALL get_tensor) (ggml_backend_buffer_t buffer, const struct ggml_tensor * tensor,       void * data, size_t offset, size_t size);
+        bool         (*GGML_CALL cpy_tensor) (ggml_backend_buffer_t buffer, const struct ggml_tensor * src, struct ggml_tensor * dst); // dst is in the buffer, src may be in any buffer
+        void         (*GGML_CALL clear)      (ggml_backend_buffer_t buffer, uint8_t value);
+        void         (*GGML_CALL reset)      (ggml_backend_buffer_t buffer); // reset any internal state due to tensor initialization, such as tensor extras
+    };
+
+    struct ggml_backend_buffer {
+        struct ggml_backend_buffer_i  iface;
+        ggml_backend_buffer_type_t    buft;
+        ggml_backend_buffer_context_t context;
+        size_t size;
+        enum ggml_backend_buffer_usage usage;
+    };
+
+    GGML_CALL ggml_backend_buffer_t ggml_backend_buffer_init(
+                   ggml_backend_buffer_type_t      buft,
+            struct ggml_backend_buffer_i           iface,
+                   ggml_backend_buffer_context_t   context,
+                   size_t                          size);
+
+    // do not use directly, use ggml_backend_tensor_copy instead
+    bool ggml_backend_buffer_copy_tensor(const struct ggml_tensor * src, struct ggml_tensor * dst);
+
+    // buffer that contains a collection of buffers
+    GGML_CALL ggml_backend_buffer_t ggml_backend_multi_buffer_alloc_buffer(ggml_backend_buffer_t * buffers, size_t n_buffers);
+    GGML_CALL bool                  ggml_backend_buffer_is_multi_buffer(ggml_backend_buffer_t buffer);
+    GGML_CALL void                  ggml_backend_multi_buffer_set_usage(ggml_backend_buffer_t buffer, enum ggml_backend_buffer_usage usage);
+
+    //
+    // Backend
+    //
+
+    typedef void * ggml_backend_context_t;
+
+    struct ggml_backend_i {
+        const char * (*GGML_CALL get_name)(ggml_backend_t backend);
+
+        void (*GGML_CALL free)(ggml_backend_t backend);
+
+        // buffer allocation
+        ggml_backend_buffer_type_t (*GGML_CALL get_default_buffer_type)(ggml_backend_t backend);
+
+        // (optional) asynchronous tensor data access
+        void (*GGML_CALL set_tensor_async)(ggml_backend_t backend,       struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
+        void (*GGML_CALL get_tensor_async)(ggml_backend_t backend, const struct ggml_tensor * tensor,       void * data, size_t offset, size_t size);
+        bool (*GGML_CALL cpy_tensor_async)(ggml_backend_t backend_src, ggml_backend_t backend_dst, const struct ggml_tensor * src, struct ggml_tensor * dst);
+
+        // (optional) complete all pending operations
+        void (*GGML_CALL synchronize)(ggml_backend_t backend);
+
+        // compute graph with a plan (not used currently)
+        ggml_backend_graph_plan_t (*GGML_CALL graph_plan_create) (ggml_backend_t backend, const struct ggml_cgraph * cgraph);
+        void                      (*GGML_CALL graph_plan_free)   (ggml_backend_t backend, ggml_backend_graph_plan_t plan);
+
+        // compute graph with a plan
+        enum ggml_status (*GGML_CALL graph_plan_compute)(ggml_backend_t backend, ggml_backend_graph_plan_t plan);
+        // compute graph without a plan (async)
+        enum ggml_status (*GGML_CALL graph_compute)     (ggml_backend_t backend, struct ggml_cgraph * cgraph);
+
+        // check if the backend supports an operation
+        bool (*GGML_CALL supports_op)(ggml_backend_t backend, const struct ggml_tensor * op);
+
+        // check if the backend wants to run an operation, even if the weights are allocated in a CPU buffer
+        // these should be expensive operations with large batch sizes that may benefit from running on this backend
+        // even if the weight has to be copied from the CPU temporarily
+        bool (*GGML_CALL offload_op)(ggml_backend_t backend, const struct ggml_tensor * op);
+
+        // (optional) event synchronization
+        ggml_backend_event_t (*GGML_CALL event_new)         (ggml_backend_t backend);
+        void                 (*GGML_CALL event_free)        (ggml_backend_event_t event);
+        void                 (*GGML_CALL event_record)      (ggml_backend_event_t event);
+        void                 (*GGML_CALL event_wait)        (ggml_backend_t backend, ggml_backend_event_t event);
+        void                 (*GGML_CALL event_synchronize) (ggml_backend_event_t event);
+    };
+
+    struct ggml_backend {
+        ggml_guid_t guid;
+
+        struct ggml_backend_i iface;
+        ggml_backend_context_t context;
+    };
+
+    struct ggml_backend_event {
+        ggml_backend_t backend;
+        void * context;
+    };
+
+    //
+    // Backend registry
+    //
+
+    typedef ggml_backend_t (*GGML_CALL ggml_backend_init_fn)(const char * params, void * user_data);
+
+    GGML_CALL void ggml_backend_register(const char * name, ggml_backend_init_fn init_fn, ggml_backend_buffer_type_t default_buffer_type, void * user_data);
+
+#ifdef  __cplusplus
+}
+#endif
--- a/bindings/ruby/ext/ggml-backend.c
+++ b/bindings/ruby/ext/ggml-backend.c
--- a/bindings/ruby/ext/ggml-backend.h
+++ b/bindings/ruby/ext/ggml-backend.h
@ -0,0 +1,233 @@
+#pragma once
+
+#include "ggml.h"
+#include "ggml-alloc.h"
+
+#ifdef  __cplusplus
+extern "C" {
+#endif
+
+    typedef struct ggml_backend_buffer_type * ggml_backend_buffer_type_t;
+    typedef struct ggml_backend_buffer * ggml_backend_buffer_t;
+    typedef struct ggml_backend_event * ggml_backend_event_t;
+    typedef struct ggml_backend * ggml_backend_t;
+    typedef void * ggml_backend_graph_plan_t;
+
+    //
+    // Backend buffer
+    //
+
+    // buffer type
+    GGML_API           const char *          ggml_backend_buft_name            (ggml_backend_buffer_type_t buft);
+    GGML_API GGML_CALL ggml_backend_buffer_t ggml_backend_buft_alloc_buffer    (ggml_backend_buffer_type_t buft, size_t size);
+    GGML_API           size_t                ggml_backend_buft_get_alignment   (ggml_backend_buffer_type_t buft);
+    GGML_API           size_t                ggml_backend_buft_get_max_size    (ggml_backend_buffer_type_t buft);
+    GGML_API GGML_CALL size_t                ggml_backend_buft_get_alloc_size  (ggml_backend_buffer_type_t buft, struct ggml_tensor * tensor);
+    GGML_API           bool                  ggml_backend_buft_supports_backend(ggml_backend_buffer_type_t buft, ggml_backend_t backend);
+    GGML_API           bool                  ggml_backend_buft_is_host         (ggml_backend_buffer_type_t buft);
+
+    // buffer
+    enum ggml_backend_buffer_usage {
+        GGML_BACKEND_BUFFER_USAGE_ANY = 0,
+        GGML_BACKEND_BUFFER_USAGE_WEIGHTS = 1,
+    };
+
+    GGML_API           const char *               ggml_backend_buffer_name          (ggml_backend_buffer_t buffer);
+    GGML_API           void                       ggml_backend_buffer_free          (ggml_backend_buffer_t buffer);
+    GGML_API           void *                     ggml_backend_buffer_get_base      (ggml_backend_buffer_t buffer);
+    GGML_API           size_t                     ggml_backend_buffer_get_size      (ggml_backend_buffer_t buffer);
+    GGML_API GGML_CALL void                       ggml_backend_buffer_init_tensor   (ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
+    GGML_API           size_t                     ggml_backend_buffer_get_alignment (ggml_backend_buffer_t buffer);
+    GGML_API           size_t                     ggml_backend_buffer_get_max_size  (ggml_backend_buffer_t buffer);
+    GGML_API           size_t                     ggml_backend_buffer_get_alloc_size(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
+    GGML_API           void                       ggml_backend_buffer_clear         (ggml_backend_buffer_t buffer, uint8_t value);
+    GGML_API           bool                       ggml_backend_buffer_is_host       (ggml_backend_buffer_t buffer);
+    GGML_API           void                       ggml_backend_buffer_set_usage     (ggml_backend_buffer_t buffer, enum ggml_backend_buffer_usage usage);
+    GGML_API           ggml_backend_buffer_type_t ggml_backend_buffer_get_type      (ggml_backend_buffer_t buffer);
+    GGML_API           void                       ggml_backend_buffer_reset         (ggml_backend_buffer_t buffer);
+
+    //
+    // Backend
+    //
+
+    GGML_API ggml_guid_t  ggml_backend_guid(ggml_backend_t backend);
+    GGML_API const char * ggml_backend_name(ggml_backend_t backend);
+    GGML_API void         ggml_backend_free(ggml_backend_t backend);
+
+    GGML_API ggml_backend_buffer_type_t ggml_backend_get_default_buffer_type(ggml_backend_t backend);
+    GGML_API ggml_backend_buffer_t      ggml_backend_alloc_buffer(ggml_backend_t backend, size_t size);
+    GGML_API size_t                     ggml_backend_get_alignment(ggml_backend_t backend);
+    GGML_API size_t                     ggml_backend_get_max_size(ggml_backend_t backend);
+
+    GGML_API void ggml_backend_tensor_set_async(ggml_backend_t backend,       struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
+    GGML_API void ggml_backend_tensor_get_async(ggml_backend_t backend, const struct ggml_tensor * tensor,       void * data, size_t offset, size_t size);
+
+    GGML_API GGML_CALL void ggml_backend_tensor_set(      struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
+    GGML_API GGML_CALL void ggml_backend_tensor_get(const struct ggml_tensor * tensor,       void * data, size_t offset, size_t size);
+
+    GGML_API void ggml_backend_synchronize(ggml_backend_t backend);
+
+    GGML_API ggml_backend_graph_plan_t ggml_backend_graph_plan_create(ggml_backend_t backend, struct ggml_cgraph * cgraph);
+    GGML_API void                      ggml_backend_graph_plan_free  (ggml_backend_t backend, ggml_backend_graph_plan_t plan);
+
+    GGML_API enum ggml_status ggml_backend_graph_plan_compute (ggml_backend_t backend, ggml_backend_graph_plan_t plan);
+    GGML_API enum ggml_status ggml_backend_graph_compute      (ggml_backend_t backend, struct ggml_cgraph * cgraph);
+    GGML_API enum ggml_status ggml_backend_graph_compute_async(ggml_backend_t backend, struct ggml_cgraph * cgraph);
+    GGML_API bool ggml_backend_supports_op(ggml_backend_t backend, const struct ggml_tensor * op);
+    GGML_API bool ggml_backend_offload_op(ggml_backend_t backend, const struct ggml_tensor * op);
+
+    // tensor copy between different backends
+    GGML_API void ggml_backend_tensor_copy(struct ggml_tensor * src, struct ggml_tensor * dst);
+
+    // asynchronous copy
+    // the copy is performed after all the currently queued operations in backend_src
+    // backend_dst will wait for the copy to complete before performing other operations
+    // automatic fallback to sync copy if async is not supported
+    GGML_API void ggml_backend_tensor_copy_async(ggml_backend_t backend_src, ggml_backend_t backend_dst, struct ggml_tensor * src, struct ggml_tensor * dst);
+
+    // events
+    GGML_API ggml_backend_event_t   ggml_backend_event_new        (ggml_backend_t backend);
+    GGML_API void                   ggml_backend_event_free       (ggml_backend_event_t event);
+    GGML_API void                   ggml_backend_event_record     (ggml_backend_event_t event);
+    GGML_API void                   ggml_backend_event_synchronize(ggml_backend_event_t event);
+    GGML_API void                   ggml_backend_event_wait       (ggml_backend_t backend, ggml_backend_event_t event); // wait async on event
+
+    //
+    // CPU backend
+    //
+
+    GGML_API ggml_backend_t ggml_backend_cpu_init(void);
+
+    GGML_API GGML_CALL bool ggml_backend_is_cpu                (ggml_backend_t backend);
+    GGML_API           void ggml_backend_cpu_set_n_threads     (ggml_backend_t backend_cpu, int n_threads);
+    GGML_API           void ggml_backend_cpu_set_abort_callback(ggml_backend_t backend_cpu, ggml_abort_callback abort_callback, void * abort_callback_data);
+
+    // Create a backend buffer from an existing pointer
+    GGML_API GGML_CALL ggml_backend_buffer_t ggml_backend_cpu_buffer_from_ptr(void * ptr, size_t size);
+
+    GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_cpu_buffer_type(void);
+
+#ifdef GGML_USE_CPU_HBM
+    GGML_API ggml_backend_buffer_type_t ggml_backend_cpu_hbm_buffer_type(void);
+#endif
+
+    //
+    // Backend registry
+    //
+
+    // The backend registry is a registry of all the available backends, and allows initializing backends in a generic way
+
+    GGML_API size_t                     ggml_backend_reg_get_count(void);
+    GGML_API size_t                     ggml_backend_reg_find_by_name(const char * name);
+    GGML_API ggml_backend_t             ggml_backend_reg_init_backend_from_str(const char * backend_str); // str is name[:params]
+    GGML_API const char *               ggml_backend_reg_get_name(size_t i);
+    GGML_API ggml_backend_t             ggml_backend_reg_init_backend(size_t i, const char * params); // params is backend-specific
+    GGML_API ggml_backend_buffer_type_t ggml_backend_reg_get_default_buffer_type(size_t i);
+    GGML_API ggml_backend_buffer_t      ggml_backend_reg_alloc_buffer(size_t i, size_t size);
+
+    //
+    // Backend scheduler
+    //
+
+    // The backend scheduler allows for multiple backends to be used together
+    // Handles compute buffer allocation, assignment of tensors to backends, and copying of tensors between backends
+    // The backends are selected based on:
+    // - the backend that supports the operation
+    // - the location of the pre-allocated tensors (e.g. the weights)
+    /*
+      Example usage:
+
+        // operations that use tensors allocated in a buffer with USAGE_WEIGHTS will be assigned
+        // preferrably to run on the same backend as the buffer
+        ggml_backend_buffer_set_usage(buf_weights, GGML_BACKEND_BUFFER_USAGE_WEIGHTS);
+
+        sched = ggml_backend_sched_new({backend_gpu, backend_gpu2, backend_cpu}, NULL, num_backends, GGML_DEFAULT_GRAPH_SIZE, false);
+
+        // initialize buffers from a max size graph (optional)
+        reserve_graph = build_graph(sched, max_batch_size);
+
+        // manually assign nodes to a backend (optional, should not be needed in most cases)
+        struct ggml_tensor * node = ggml_mul_mat(ctx, ...);
+        ggml_backend_sched_set_tensor_backend(sched, node, backend_gpu);
+
+        ggml_backend_sched_reserve(sched, reserve_graph);
+
+        // compute
+        graph = build_graph(sched);
+        ggml_backend_sched_graph_compute(sched, graph);
+
+        // if there are graph inputs:
+        ggml_backend_sched_reset(sched);
+        ggml_backend_sched_alloc_graph(sched, graph);
+        ggml_backend_tensor_set(input_tensor, ...);
+        ggml_backend_sched_graph_compute(sched, graph);
+    }
+    */
+
+    struct ggml_backend_sched;
+    typedef struct ggml_backend_sched * ggml_backend_sched_t;
+
+    // when ask == true, the scheduler wants to know if the user wants to observe this node
+    // this allows the scheduler to batch nodes together in order to evaluate them in a single call
+    //
+    // when ask == false, the scheduler is passing the node tensor to the user for observation
+    // if the user returns false, the scheduler will cancel the graph compute
+    //
+    typedef bool (*ggml_backend_sched_eval_callback)(struct ggml_tensor * t, bool ask, void * user_data);
+
+    // Initialize a backend scheduler
+    GGML_API ggml_backend_sched_t ggml_backend_sched_new(ggml_backend_t * backends, ggml_backend_buffer_type_t * bufts, int n_backends, size_t graph_size, bool parallel);
+    GGML_API void                 ggml_backend_sched_free(ggml_backend_sched_t sched);
+
+    // Initialize backend buffers from a measure graph
+    GGML_API bool                 ggml_backend_sched_reserve(ggml_backend_sched_t sched, struct ggml_cgraph * measure_graph);
+
+    // Get the number of splits of the last graph
+    GGML_API int                  ggml_backend_sched_get_n_splits(ggml_backend_sched_t sched);
+    GGML_API int                  ggml_backend_sched_get_n_copies(ggml_backend_sched_t sched);
+
+    GGML_API size_t               ggml_backend_sched_get_buffer_size(ggml_backend_sched_t sched, ggml_backend_t backend);
+
+    GGML_API void                 ggml_backend_sched_set_tensor_backend(ggml_backend_sched_t sched, struct ggml_tensor * node, ggml_backend_t backend);
+    GGML_API ggml_backend_t       ggml_backend_sched_get_tensor_backend(ggml_backend_sched_t sched, struct ggml_tensor * node);
+
+    // Allocate and compute graph on the backend scheduler
+    GGML_API bool                 ggml_backend_sched_alloc_graph(ggml_backend_sched_t sched, struct ggml_cgraph * graph);
+    GGML_API enum ggml_status     ggml_backend_sched_graph_compute(ggml_backend_sched_t sched, struct ggml_cgraph * graph);
+    GGML_API enum ggml_status     ggml_backend_sched_graph_compute_async(ggml_backend_sched_t sched, struct ggml_cgraph * graph);
+    GGML_API void                 ggml_backend_sched_synchronize(ggml_backend_sched_t sched);
+
+    // Reset all assignments and allocators - must be called before changing the node backends
+    GGML_API void                 ggml_backend_sched_reset(ggml_backend_sched_t sched);
+
+    // Set a callback to be called for each resulting node during graph compute
+    GGML_API void                 ggml_backend_sched_set_eval_callback(ggml_backend_sched_t sched, ggml_backend_sched_eval_callback callback, void * user_data);
+
+    //
+    // Utils
+    //
+
+    struct ggml_backend_graph_copy {
+        ggml_backend_buffer_t buffer;
+        struct ggml_context * ctx_allocated;
+        struct ggml_context * ctx_unallocated;
+        struct ggml_cgraph * graph;
+    };
+
+    // Copy a graph to a different backend
+    GGML_API struct ggml_backend_graph_copy ggml_backend_graph_copy(ggml_backend_t backend, struct ggml_cgraph * graph);
+    GGML_API void                           ggml_backend_graph_copy_free(struct ggml_backend_graph_copy copy);
+
+    typedef bool (*GGML_CALL ggml_backend_eval_callback)(int node_index, struct ggml_tensor * t1, struct ggml_tensor * t2, void * user_data);
+
+    // Compare the output of two backends
+    GGML_API bool ggml_backend_compare_graph_backend(ggml_backend_t backend1, ggml_backend_t backend2, struct ggml_cgraph * graph, ggml_backend_eval_callback callback, void * user_data);
+
+    // Tensor initialization
+    GGML_API void ggml_backend_tensor_alloc(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, void * addr);
+    GGML_API void ggml_backend_view_init(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
+
+
+#ifdef  __cplusplus
+}
+#endif
--- a/bindings/ruby/ext/ggml-common.h
+++ b/bindings/ruby/ext/ggml-common.h
@ -6,20 +6,7 @@
 typedef uint16_t ggml_half;
 typedef uint32_t ggml_half2;

-#define GGML_COMMON_AGGR_U
-#define GGML_COMMON_AGGR_S
-
-#define GGML_COMMON_DECL
-#elif defined(GGML_COMMON_DECL_CPP)
-#include <cstdint>
-
-typedef uint16_t ggml_half;
-typedef uint32_t ggml_half2;
-
-// std-c++ allow anonymous unions but some compiler warn on it
-#define GGML_COMMON_AGGR_U data
-// std-c++ do not allow it.
-#define GGML_COMMON_AGGR_S data
+#define GGML_COMMON_AGGR

 #define GGML_COMMON_DECL
 #elif defined(GGML_COMMON_DECL_METAL)
@ -28,23 +15,17 @@ typedef uint32_t ggml_half2;
 typedef half  ggml_half;
 typedef half2 ggml_half2;

-#define GGML_COMMON_AGGR_U
-#define GGML_COMMON_AGGR_S
+#define GGML_COMMON_AGGR

 #define GGML_COMMON_DECL
 #elif defined(GGML_COMMON_DECL_CUDA)
-#if defined(GGML_COMMON_DECL_MUSA)
-#include <musa_fp16.h>
-#else
 #include <cuda_fp16.h>
-#endif
 #include <cstdint>

 typedef half  ggml_half;
 typedef half2 ggml_half2;

-#define GGML_COMMON_AGGR_U
-#define GGML_COMMON_AGGR_S data
+#define GGML_COMMON_AGGR data

 #define GGML_COMMON_DECL
 #elif defined(GGML_COMMON_DECL_HIP)
@ -54,8 +35,7 @@ typedef half2 ggml_half2;
 typedef half  ggml_half;
 typedef half2 ggml_half2;

-#define GGML_COMMON_AGGR_U
-#define GGML_COMMON_AGGR_S data
+#define GGML_COMMON_AGGR data

 #define GGML_COMMON_DECL
 #elif defined(GGML_COMMON_DECL_SYCL)
@ -65,8 +45,7 @@ typedef half2 ggml_half2;
 typedef sycl::half  ggml_half;
 typedef sycl::half2 ggml_half2;

-#define GGML_COMMON_AGGR_U
-#define GGML_COMMON_AGGR_S data
+#define GGML_COMMON_AGGR data

 #define GGML_COMMON_DECL
 #endif
@ -86,8 +65,13 @@ typedef sycl::half2 ggml_half2;
 // QK = number of values after dequantization
 // QK_K = super-block size

+#ifdef GGML_QKK_64
+#define QK_K 64
+#define K_SCALE_SIZE 4
+#else
 #define QK_K 256
 #define K_SCALE_SIZE 12
+#endif // GGML_QKK_64

 #if defined(GGML_COMMON_DECL_CUDA) || defined(GGML_COMMON_DECL_HIP) || defined(GGML_COMMON_DECL_SYCL)
 // QR = QK / number of values before dequantization
@ -127,43 +111,36 @@ typedef sycl::half2 ggml_half2;
 #define QR6_K 2

 #define QI2_XXS (QK_K / (4*QR2_XXS))
-#define QR2_XXS 4
+#define QR2_XXS 8

 #define QI2_XS (QK_K / (4*QR2_XS))
-#define QR2_XS 4
+#define QR2_XS 8

 #define QI2_S (QK_K / (4*QR2_S))
-#define QR2_S 4
+#define QR2_S 8

 #define QI3_XXS (QK_K / (4*QR3_XXS))
-#define QR3_XXS 4
+#define QR3_XXS 8

 #define QI3_XS (QK_K / (4*QR3_XS))
-#define QR3_XS 4
+#define QR3_XS 8

 #define QI1_S (QK_K / (4*QR1_S))
 #define QR1_S 8

-#define QI1_M (QK_K / (4*QR1_M))
-#define QR1_M 8
-
 #define QI4_NL (QK4_NL / (4*QR4_NL))
 #define QR4_NL 2

+#if QK_K == 64
+#define QI4_XS QI4_NL
+#define QR4_XS QR4_NL
+#else
 #define QI4_XS (QK_K / (4*QR4_XS))
-#define QR4_XS 2
-
-#define QI3_S (QK_K / (4*QR3_S))
-#define QR3_S 4
+#define QR4_XS 8
+#endif

 #endif // GGML_COMMON_DECL_CUDA || GGML_COMMON_DECL_HIP

-#ifdef _MSC_VER
-#define GGML_EXTENSION
-#else // _MSC_VER
-#define GGML_EXTENSION __extension__
-#endif // _MSC_VER
-
 #define QK4_0 32
 typedef struct {
    ggml_half d;           // delta
@ -173,13 +150,13 @@ static_assert(sizeof(block_q4_0) == sizeof(ggml_half) + QK4_0 / 2, "wrong q4_0 b

 #define QK4_1 32
 typedef struct {
-    GGML_EXTENSION union {
+    union {
        struct {
            ggml_half d; // delta
            ggml_half m; // min
-        } GGML_COMMON_AGGR_S;
+        } GGML_COMMON_AGGR;
        ggml_half2 dm;
-    } GGML_COMMON_AGGR_U;
+    };
    uint8_t qs[QK4_1 / 2]; // nibbles / quants
 } block_q4_1;
 static_assert(sizeof(block_q4_1) == 2 * sizeof(ggml_half) + QK4_1 / 2, "wrong q4_1 block size/padding");
@ -194,13 +171,13 @@ static_assert(sizeof(block_q5_0) == sizeof(ggml_half) + sizeof(uint32_t) + QK5_0

 #define QK5_1 32
 typedef struct {
-    GGML_EXTENSION union {
+    union {
        struct {
            ggml_half d; // delta
            ggml_half m; // min
-        } GGML_COMMON_AGGR_S;
+        } GGML_COMMON_AGGR;
        ggml_half2 dm;
-    } GGML_COMMON_AGGR_U;
+    };
    uint8_t qh[4];         // 5-th bit of quants
    uint8_t qs[QK5_1 / 2]; // nibbles / quants
 } block_q5_1;
@ -215,36 +192,17 @@ static_assert(sizeof(block_q8_0) == sizeof(ggml_half) + QK8_0, "wrong q8_0 block

 #define QK8_1 32
 typedef struct {
-    GGML_EXTENSION union {
+    union {
        struct {
            ggml_half d; // delta
            ggml_half s; // d * sum(qs[i])
-        } GGML_COMMON_AGGR_S;
+        } GGML_COMMON_AGGR;
        ggml_half2 ds;
-    } GGML_COMMON_AGGR_U;
+    };
    int8_t qs[QK8_1]; // quants
 } block_q8_1;
 static_assert(sizeof(block_q8_1) == 2*sizeof(ggml_half) + QK8_1, "wrong q8_1 block size/padding");

-//
-// Ternary quantization
-//
-
-// 1.6875 bpw
-typedef struct {
-    uint8_t qs[(QK_K - 4 * QK_K / 64) / 5]; // 5 elements per byte (3^5 = 243 < 256)
-    uint8_t qh[QK_K/64]; // 4 elements per byte
-    ggml_half d;
-} block_tq1_0;
-static_assert(sizeof(block_tq1_0) == sizeof(ggml_half) + QK_K / 64 + (QK_K - 4 * QK_K / 64) / 5, "wrong tq1_0 block size/padding");
-
-// 2.0625 bpw
-typedef struct {
-    uint8_t qs[QK_K/4]; // 2 bits per element
-    ggml_half d;
-} block_tq2_0;
-static_assert(sizeof(block_tq2_0) == sizeof(ggml_half) + QK_K / 4, "wrong tq2_0 block size/padding");
-
 //
 // Super-block quantization structures
 //
@ -256,13 +214,13 @@ static_assert(sizeof(block_tq2_0) == sizeof(ggml_half) + QK_K / 4, "wrong tq2_0
 typedef struct {
    uint8_t scales[QK_K/16]; // scales and mins, quantized with 4 bits
    uint8_t qs[QK_K/4];      // quants
-    GGML_EXTENSION union {
+    union {
        struct {
            ggml_half d;    // super-block scale for quantized scales
            ggml_half dmin; // super-block scale for quantized mins
-        } GGML_COMMON_AGGR_S;
+        } GGML_COMMON_AGGR;
        ggml_half2 dm;
-    } GGML_COMMON_AGGR_U;
+    };
 } block_q2_K;
 static_assert(sizeof(block_q2_K) == 2*sizeof(ggml_half) + QK_K/16 + QK_K/4, "wrong q2_K block size/padding");

@ -270,6 +228,15 @@ static_assert(sizeof(block_q2_K) == 2*sizeof(ggml_half) + QK_K/16 + QK_K/4, "wro
 // weight is represented as x = a * q
 // 16 blocks of 16 elements each
 // Effectively 3.4375 bits per weight
+#ifdef GGML_QKK_64
+typedef struct {
+    uint8_t hmask[QK_K/8]; // quants - high bit
+    uint8_t qs[QK_K/4];    // quants - low 2 bits
+    uint8_t scales[2];
+    ggml_half d;           // super-block scale
+} block_q3_K;
+static_assert(sizeof(block_q3_K) == sizeof(ggml_half) + QK_K / 4 + QK_K / 8 + 2, "wrong q3_K block size/padding");
+#else
 typedef struct {
    uint8_t hmask[QK_K/8]; // quants - high bit
    uint8_t qs[QK_K/4];    // quants - low 2 bits
@ -277,41 +244,61 @@ typedef struct {
    ggml_half d;           // super-block scale
 } block_q3_K;
 static_assert(sizeof(block_q3_K) == sizeof(ggml_half) + QK_K / 4 + QK_K / 8 + 12, "wrong q3_K block size/padding");
+#endif

 // 4-bit quantization
 // 8 blocks of 32 elements each
 // weight is represented as x = a * q + b
 // Effectively 4.5 bits per weight
+#ifdef GGML_QKK_64
 typedef struct {
-    GGML_EXTENSION union {
+    ggml_half d[2];     // super-block scales/mins
+    uint8_t scales[2];  // 4-bit block scales/mins
+    uint8_t qs[QK_K/2]; // 4--bit quants
+} block_q4_K;
+static_assert(sizeof(block_q4_K) == 2*sizeof(ggml_half) + QK_K/2 + 2, "wrong q4_K block size/padding");
+#else
+typedef struct {
+    union {
        struct {
            ggml_half d;    // super-block scale for quantized scales
            ggml_half dmin; // super-block scale for quantized mins
-        } GGML_COMMON_AGGR_S;
+        } GGML_COMMON_AGGR;
        ggml_half2 dm;
-    } GGML_COMMON_AGGR_U;
+    };
    uint8_t scales[K_SCALE_SIZE]; // scales and mins, quantized with 6 bits
    uint8_t qs[QK_K/2];           // 4--bit quants
 } block_q4_K;
 static_assert(sizeof(block_q4_K) == 2*sizeof(ggml_half) + K_SCALE_SIZE + QK_K/2, "wrong q4_K block size/padding");
+#endif

 // 5-bit quantization
 // 8 blocks of 32 elements each
 // weight is represented as x = a * q + b
 // Effectively 5.5 bits per weight
+#ifdef GGML_QKK_64
 typedef struct {
-    GGML_EXTENSION union {
+    ggml_half d;             // super-block scale
+    int8_t  scales[QK_K/16]; // 8-bit block scales
+    uint8_t qh[QK_K/8];      // quants, high bit
+    uint8_t qs[QK_K/2];      // quants, low 4 bits
+} block_q5_K;
+static_assert(sizeof(block_q5_K) == sizeof(ggml_half) + QK_K/2 + QK_K/8 + QK_K/16, "wrong q5_K block size/padding");
+#else
+typedef struct {
+    union {
        struct {
            ggml_half d;    // super-block scale for quantized scales
            ggml_half dmin; // super-block scale for quantized mins
-        } GGML_COMMON_AGGR_S;
+        } GGML_COMMON_AGGR;
        ggml_half2 dm;
-    } GGML_COMMON_AGGR_U;
+    };
    uint8_t scales[K_SCALE_SIZE]; // scales and mins, quantized with 6 bits
    uint8_t qh[QK_K/8];           // quants, high bit
    uint8_t qs[QK_K/2];           // quants, low 4 bits
 } block_q5_K;
 static_assert(sizeof(block_q5_K) == 2*sizeof(ggml_half) + K_SCALE_SIZE + QK_K/2 + QK_K/8, "wrong q5_K block size/padding");
+#endif

 // 6-bit quantization
 // weight is represented as x = a * q
@ -369,7 +356,11 @@ typedef struct {
 static_assert(sizeof(block_iq3_xxs) == sizeof(ggml_half) + 3*(QK_K/8), "wrong iq3_xxs block size/padding");

 // 3.4375 bpw
+#if QK_K == 64
+#define IQ3S_N_SCALE 2
+#else
 #define IQ3S_N_SCALE QK_K/64
+#endif
 typedef struct {
    ggml_half d;
    uint8_t qs[QK_K/4];
@ -379,7 +370,6 @@ typedef struct {
 } block_iq3_s;
 static_assert(sizeof(block_iq3_s) == sizeof(ggml_half) + 13*(QK_K/32) + IQ3S_N_SCALE, "wrong iq3_s block size/padding");

-// 1.5625 bpw
 typedef struct {
    ggml_half d;
    uint8_t  qs[QK_K/8];
@ -391,9 +381,16 @@ static_assert(sizeof(block_iq1_s) == sizeof(ggml_half) + QK_K/8 + QK_K/16, "wron
 typedef struct {
    uint8_t  qs[QK_K/8];      // grid index, low 8 bits
    uint8_t  qh[QK_K/16];     // grid index, high 3 bits + grid shift bit (for two groups of 8)
+#if QK_K == 64
+    ggml_half d;
+#endif
    uint8_t  scales[QK_K/32]; // 3-bit block scales (4-bit if QK_K == 64)
 } block_iq1_m;
+#if QK_K == 64
+static_assert(sizeof(block_iq1_m) == QK_K/8 + QK_K/16 + QK_K/32 + sizeof(ggml_half), "wrong iq1_m block size/padding");
+#else
 static_assert(sizeof(block_iq1_m) == QK_K/8 + QK_K/16 + QK_K/32, "wrong iq1_m block size/padding");
+#endif

 // Used by IQ1_M quants
 typedef union {
@ -409,6 +406,9 @@ typedef struct {
 } block_iq4_nl;
 static_assert(sizeof(block_iq4_nl) == sizeof(ggml_half) + QK4_NL/2, "wrong iq4_nl block size/padding");

+#if QK_K == 64
+#define block_iq4_xs block_iq4_nl
+#else
 typedef struct {
    ggml_half d;
    uint16_t scales_h;
@ -416,6 +416,7 @@ typedef struct {
    uint8_t  qs[QK_K/2];
 } block_iq4_xs;
 static_assert(sizeof(block_iq4_xs) == sizeof(ggml_half) + sizeof(uint16_t) + QK_K/64 + QK_K/2, "wrong iq4_xs block size/padding");
+#endif

 #endif // GGML_COMMON_DECL
 #endif // GGML_COMMON_DECL
@ -430,13 +431,6 @@ static_assert(sizeof(block_iq4_xs) == sizeof(ggml_half) + sizeof(uint16_t) + QK_
 #define GGML_TABLE_BEGIN(type, name, size) static const type name[size] = {
 #define GGML_TABLE_END() };

-#define GGML_COMMON_IMPL
-#elif defined(GGML_COMMON_IMPL_CPP)
-#include <cstdint>
-
-#define GGML_TABLE_BEGIN(type, name, size) static const type name[size] = {
-#define GGML_TABLE_END() };
-
 #define GGML_COMMON_IMPL
 #elif defined(GGML_COMMON_IMPL_METAL)
 #include <metal_stdlib>
@ -445,7 +439,7 @@ static_assert(sizeof(block_iq4_xs) == sizeof(ggml_half) + sizeof(uint16_t) + QK_
 #define GGML_TABLE_END() };

 #define GGML_COMMON_IMPL
-#elif defined(GGML_COMMON_IMPL_CUDA) || defined(GGML_COMMON_IMPL_HIP) || defined(GGML_COMMON_IMPL_MUSA)
+#elif defined(GGML_COMMON_IMPL_CUDA) || defined(GGML_COMMON_IMPL_HIP)
 #include <cstdint>

 #define GGML_TABLE_BEGIN(type, name, size) static const __device__ type name[size] = {
@ -479,6 +473,7 @@ GGML_TABLE_BEGIN(uint8_t, ksigns_iq2xs, 128)
    240, 113, 114, 243, 116, 245, 246, 119, 120, 249, 250, 123, 252, 125, 126, 255,
 GGML_TABLE_END()

+//#if __CUDA_ARCH__ >= MIN_CC_DP4A // lowest compute capability for integer intrinsics
 GGML_TABLE_BEGIN(uint64_t, ksigns64, 128)
    0x0000000000000000, 0xff000000000000ff, 0xff0000000000ff00, 0x000000000000ffff,
    0xff00000000ff0000, 0x0000000000ff00ff, 0x0000000000ffff00, 0xff00000000ffffff,
@ -513,6 +508,7 @@ GGML_TABLE_BEGIN(uint64_t, ksigns64, 128)
    0x00ffffffff000000, 0xffffffffff0000ff, 0xffffffffff00ff00, 0x00ffffffff00ffff,
    0xffffffffffff0000, 0x00ffffffffff00ff, 0x00ffffffffffff00, 0xffffffffffffffff,
 GGML_TABLE_END()
+//#endif


 GGML_TABLE_BEGIN(uint64_t, iq2xxs_grid, 256)
@ -1074,10 +1070,6 @@ GGML_TABLE_BEGIN(uint32_t, iq3s_grid, 512)
    0x0f090307, 0x0f090501, 0x0f090b01, 0x0f0b0505, 0x0f0b0905, 0x0f0d0105, 0x0f0d0703, 0x0f0f0101,
 GGML_TABLE_END()

-GGML_TABLE_BEGIN(int8_t, kvalues_iq4nl, 16)
-    -127, -104, -83, -65, -49, -35, -22, -10, 1, 13, 25, 38, 53, 69, 89, 113,
-GGML_TABLE_END()
-
 #define NGRID_IQ1S 2048
 #define IQ1S_DELTA 0.125f
 #define IQ1M_DELTA 0.125f
--- a/bindings/ruby/ext/ggml-cuda.h
+++ b/bindings/ruby/ext/ggml-cuda.h
@ -0,0 +1,43 @@
+#pragma once
+
+#include "ggml.h"
+#include "ggml-backend.h"
+
+#ifdef GGML_USE_HIPBLAS
+#define GGML_CUDA_NAME "ROCm"
+#define GGML_CUBLAS_NAME "hipBLAS"
+#else
+#define GGML_CUDA_NAME "CUDA"
+#define GGML_CUBLAS_NAME "cuBLAS"
+#endif
+
+#ifdef  __cplusplus
+extern "C" {
+#endif
+
+#define GGML_CUDA_MAX_DEVICES       16
+
+// backend API
+GGML_API GGML_CALL ggml_backend_t ggml_backend_cuda_init(int device);
+
+GGML_API GGML_CALL bool ggml_backend_is_cuda(ggml_backend_t backend);
+
+// device buffer
+GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_cuda_buffer_type(int device);
+
+// split tensor buffer that splits matrices by rows across multiple devices
+GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_cuda_split_buffer_type(const float * tensor_split);
+
+// pinned host buffer for use with the CPU backend for faster copies between CPU and GPU
+GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_cuda_host_buffer_type(void);
+
+GGML_API GGML_CALL int  ggml_backend_cuda_get_device_count(void);
+GGML_API GGML_CALL void ggml_backend_cuda_get_device_description(int device, char * description, size_t description_size);
+GGML_API GGML_CALL void ggml_backend_cuda_get_device_memory(int device, size_t * free, size_t * total);
+
+GGML_API GGML_CALL bool ggml_backend_cuda_register_host_buffer(void * buffer, size_t size);
+GGML_API GGML_CALL void ggml_backend_cuda_unregister_host_buffer(void * buffer);
+
+#ifdef  __cplusplus
+}
+#endif
--- a/bindings/ruby/ext/ggml-impl.h
+++ b/bindings/ruby/ext/ggml-impl.h
@ -0,0 +1,272 @@
+#pragma once
+
+#include "ggml.h"
+
+// GGML internal header
+
+#include <assert.h>
+#include <stdlib.h> // load `stdlib.h` before other headers to work around MinGW bug: https://sourceforge.net/p/mingw-w64/bugs/192/
+#include <stddef.h>
+#include <stdbool.h>
+#include <string.h> // memcpy
+#include <math.h>   // fabsf
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// static_assert should be a #define, but if it's not,
+// fall back to the _Static_assert C11 keyword.
+// if C99 - static_assert is noop
+// ref: https://stackoverflow.com/a/53923785/4039976
+#ifndef __cplusplus
+#ifndef static_assert
+#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201100L)
+#define static_assert(cond, msg) _Static_assert(cond, msg)
+#else
+#define static_assert(cond, msg) struct global_scope_noop_trick
+#endif
+#endif
+#endif
+
+// __FMA__ and __F16C__ are not defined in MSVC, however they are implied with AVX2/AVX512
+#if defined(_MSC_VER) && (defined(__AVX2__) || defined(__AVX512F__))
+#ifndef __FMA__
+#define __FMA__
+#endif
+#ifndef __F16C__
+#define __F16C__
+#endif
+#endif
+
+// __SSE3__ and __SSSE3__ are not defined in MSVC, but SSE3/SSSE3 are present when AVX/AVX2/AVX512 are available
+#if defined(_MSC_VER) && (defined(__AVX__) || defined(__AVX2__) || defined(__AVX512F__))
+#ifndef __SSE3__
+#define __SSE3__
+#endif
+#ifndef __SSSE3__
+#define __SSSE3__
+#endif
+#endif
+
+// 16-bit float
+// on Arm, we use __fp16
+// on x86, we use uint16_t
+#if defined(__ARM_NEON) && !defined(_MSC_VER)
+
+// if YCM cannot find <arm_neon.h>, make a symbolic link to it, for example:
+//
+//   $ ln -sfn /Library/Developer/CommandLineTools/usr/lib/clang/13.1.6/include/arm_neon.h ./src/
+//
+#include <arm_neon.h>
+
+typedef __fp16 ggml_fp16_internal_t;
+
+#define GGML_COMPUTE_FP16_TO_FP32(x) ggml_compute_fp16_to_fp32(x)
+#define GGML_COMPUTE_FP32_TO_FP16(x) ggml_compute_fp32_to_fp16(x)
+
+#define GGML_FP16_TO_FP32(x) ggml_compute_fp16_to_fp32(x)
+
+static inline float ggml_compute_fp16_to_fp32(ggml_fp16_t h) {
+    ggml_fp16_internal_t tmp;
+    memcpy(&tmp, &h, sizeof(ggml_fp16_t));
+    return (float)tmp;
+}
+
+static inline ggml_fp16_t ggml_compute_fp32_to_fp16(float f) {
+    ggml_fp16_t res;
+    ggml_fp16_internal_t tmp = f;
+    memcpy(&res, &tmp, sizeof(ggml_fp16_t));
+    return res;
+}
+
+#else
+
+typedef uint16_t ggml_fp16_internal_t;
+
+#ifdef __wasm_simd128__
+#include <wasm_simd128.h>
+#else
+#ifdef __POWER9_VECTOR__
+#include <altivec.h>
+#undef bool
+#define bool _Bool
+#else
+#if defined(_MSC_VER) || defined(__MINGW32__)
+#include <intrin.h>
+#else
+#if defined(__AVX__) || defined(__AVX2__) || defined(__AVX512F__) || defined(__SSSE3__) || defined(__SSE3__)
+#if !defined(__riscv)
+#include <immintrin.h>
+#endif
+#endif
+#endif
+#endif
+#endif
+
+#ifdef __riscv_v_intrinsic
+#include <riscv_vector.h>
+#endif
+
+#ifdef __F16C__
+
+#ifdef _MSC_VER
+#define GGML_COMPUTE_FP16_TO_FP32(x) _mm_cvtss_f32(_mm_cvtph_ps(_mm_cvtsi32_si128(x)))
+#define GGML_COMPUTE_FP32_TO_FP16(x) _mm_extract_epi16(_mm_cvtps_ph(_mm_set_ss(x), 0), 0)
+#else
+#define GGML_COMPUTE_FP16_TO_FP32(x) _cvtsh_ss(x)
+#define GGML_COMPUTE_FP32_TO_FP16(x) _cvtss_sh(x, 0)
+#endif
+
+#elif defined(__POWER9_VECTOR__)
+
+#define GGML_COMPUTE_FP16_TO_FP32(x) ggml_compute_fp16_to_fp32(x)
+#define GGML_COMPUTE_FP32_TO_FP16(x) ggml_compute_fp32_to_fp16(x)
+/* the inline asm below is about 12% faster than the lookup method */
+#define GGML_FP16_TO_FP32(x) GGML_COMPUTE_FP16_TO_FP32(x)
+#define GGML_FP32_TO_FP16(x) GGML_COMPUTE_FP32_TO_FP16(x)
+
+static inline float ggml_compute_fp16_to_fp32(ggml_fp16_t h) {
+    register float f;
+    register double d;
+    __asm__(
+        "mtfprd %0,%2\n"
+        "xscvhpdp %0,%0\n"
+        "frsp %1,%0\n" :
+        /* temp */ "=d"(d),
+        /* out */  "=f"(f):
+        /* in */   "r"(h));
+    return f;
+}
+
+static inline ggml_fp16_t ggml_compute_fp32_to_fp16(float f) {
+    register double d;
+    register ggml_fp16_t r;
+    __asm__( /* xscvdphp can work on double or single precision */
+        "xscvdphp %0,%2\n"
+        "mffprd %1,%0\n" :
+        /* temp */ "=d"(d),
+        /* out */  "=r"(r):
+        /* in */   "f"(f));
+    return r;
+}
+
+#else
+
+// FP16 <-> FP32
+// ref: https://github.com/Maratyszcza/FP16
+
+static inline float fp32_from_bits(uint32_t w) {
+    union {
+        uint32_t as_bits;
+        float as_value;
+    } fp32;
+    fp32.as_bits = w;
+    return fp32.as_value;
+}
+
+static inline uint32_t fp32_to_bits(float f) {
+    union {
+        float as_value;
+        uint32_t as_bits;
+    } fp32;
+    fp32.as_value = f;
+    return fp32.as_bits;
+}
+
+static inline float ggml_compute_fp16_to_fp32(ggml_fp16_t h) {
+    const uint32_t w = (uint32_t) h << 16;
+    const uint32_t sign = w & UINT32_C(0x80000000);
+    const uint32_t two_w = w + w;
+
+    const uint32_t exp_offset = UINT32_C(0xE0) << 23;
+#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) || defined(__GNUC__) && !defined(__STRICT_ANSI__)
+    const float exp_scale = 0x1.0p-112f;
+#else
+    const float exp_scale = fp32_from_bits(UINT32_C(0x7800000));
+#endif
+    const float normalized_value = fp32_from_bits((two_w >> 4) + exp_offset) * exp_scale;
+
+    const uint32_t magic_mask = UINT32_C(126) << 23;
+    const float magic_bias = 0.5f;
+    const float denormalized_value = fp32_from_bits((two_w >> 17) | magic_mask) - magic_bias;
+
+    const uint32_t denormalized_cutoff = UINT32_C(1) << 27;
+    const uint32_t result = sign |
+        (two_w < denormalized_cutoff ? fp32_to_bits(denormalized_value) : fp32_to_bits(normalized_value));
+    return fp32_from_bits(result);
+}
+
+static inline ggml_fp16_t ggml_compute_fp32_to_fp16(float f) {
+#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) || defined(__GNUC__) && !defined(__STRICT_ANSI__)
+    const float scale_to_inf = 0x1.0p+112f;
+    const float scale_to_zero = 0x1.0p-110f;
+#else
+    const float scale_to_inf = fp32_from_bits(UINT32_C(0x77800000));
+    const float scale_to_zero = fp32_from_bits(UINT32_C(0x08800000));
+#endif
+    float base = (fabsf(f) * scale_to_inf) * scale_to_zero;
+
+    const uint32_t w = fp32_to_bits(f);
+    const uint32_t shl1_w = w + w;
+    const uint32_t sign = w & UINT32_C(0x80000000);
+    uint32_t bias = shl1_w & UINT32_C(0xFF000000);
+    if (bias < UINT32_C(0x71000000)) {
+        bias = UINT32_C(0x71000000);
+    }
+
+    base = fp32_from_bits((bias >> 1) + UINT32_C(0x07800000)) + base;
+    const uint32_t bits = fp32_to_bits(base);
+    const uint32_t exp_bits = (bits >> 13) & UINT32_C(0x00007C00);
+    const uint32_t mantissa_bits = bits & UINT32_C(0x00000FFF);
+    const uint32_t nonsign = exp_bits + mantissa_bits;
+    return (sign >> 16) | (shl1_w > UINT32_C(0xFF000000) ? UINT16_C(0x7E00) : nonsign);
+}
+
+#define GGML_COMPUTE_FP16_TO_FP32(x) ggml_compute_fp16_to_fp32(x)
+#define GGML_COMPUTE_FP32_TO_FP16(x) ggml_compute_fp32_to_fp16(x)
+
+#endif // __F16C__
+
+#endif // __ARM_NEON
+
+// precomputed f32 table for f16 (256 KB)
+// defined in ggml.c, initialized in ggml_init()
+extern float ggml_table_f32_f16[1 << 16];
+
+// On ARM NEON, it's quicker to directly convert x -> x instead of calling into ggml_lookup_fp16_to_fp32,
+// so we define GGML_FP16_TO_FP32 and GGML_FP32_TO_FP16 elsewhere for NEON.
+// This is also true for POWER9.
+#if !defined(GGML_FP16_TO_FP32)
+inline static float ggml_lookup_fp16_to_fp32(ggml_fp16_t f) {
+    uint16_t s;
+    memcpy(&s, &f, sizeof(uint16_t));
+    return ggml_table_f32_f16[s];
+}
+
+#define GGML_FP16_TO_FP32(x) ggml_lookup_fp16_to_fp32(x)
+#endif
+
+#if !defined(GGML_FP32_TO_FP16)
+#define GGML_FP32_TO_FP16(x) GGML_COMPUTE_FP32_TO_FP16(x)
+#endif
+
+#define GGML_HASHTABLE_FULL ((size_t)-1)
+#define GGML_HASHTABLE_ALREADY_EXISTS ((size_t)-2)
+
+struct ggml_hash_set ggml_hash_set_new(size_t size);
+
+bool   ggml_hash_contains      (const struct ggml_hash_set hash_set, struct ggml_tensor * key);
+
+// returns GGML_HASHTABLE_FULL if table is full, otherwise the current index of the key or where it should be inserted
+size_t ggml_hash_find          (const struct ggml_hash_set hash_set, struct ggml_tensor * key);
+
+// returns GGML_HASHTABLE_ALREADY_EXISTS if key already exists, index otherwise, asserts if table is full
+size_t ggml_hash_insert        (      struct ggml_hash_set hash_set, struct ggml_tensor * key);
+
+// return index, asserts if table is full
+size_t ggml_hash_find_or_insert(      struct ggml_hash_set hash_set, struct ggml_tensor * key);
+
+#ifdef __cplusplus
+}
+#endif
--- a/bindings/ruby/ext/ggml-kompute.h
+++ b/bindings/ruby/ext/ggml-kompute.h
@ -11,8 +11,6 @@
 extern "C" {
 #endif

-#define GGML_KOMPUTE_MAX_DEVICES 16
-
 struct ggml_vk_device {
    int index;
    int type; // same as VkPhysicalDeviceType
@ -37,13 +35,11 @@ struct ggml_vk_device ggml_vk_current_device(void);
 // forward declaration
 typedef struct ggml_backend * ggml_backend_t;

-GGML_BACKEND_API ggml_backend_t ggml_backend_kompute_init(int device);
+GGML_API ggml_backend_t ggml_backend_kompute_init(int device);

-GGML_BACKEND_API bool ggml_backend_is_kompute(ggml_backend_t backend);
+GGML_API bool ggml_backend_is_kompute(ggml_backend_t backend);

-GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_kompute_buffer_type(int device);
-
-GGML_BACKEND_API ggml_backend_reg_t ggml_backend_kompute_reg(void);
+GGML_API ggml_backend_buffer_type_t ggml_backend_kompute_buffer_type(int device);

 #ifdef __cplusplus
 }
--- a/bindings/ruby/ext/ggml-metal.h
+++ b/bindings/ruby/ext/ggml-metal.h
@ -1,9 +1,7 @@
-// Note: this description is outdated
-//
 // An interface allowing to compute ggml_cgraph with Metal
 //
 // This is a fully functional interface that extends ggml with GPU support for Apple devices.
-// A similar interface can be created for other GPU backends (e.g. Vulkan, CUDA, etc.)
+// A similar interface can be created for other GPU backends (e.g. Vulkan, CUDA, OpenCL, etc.)
 //
 // How it works?
 //
@ -27,6 +25,9 @@
 #include <stddef.h>
 #include <stdbool.h>

+// max memory buffers that can be mapped to the device
+#define GGML_METAL_MAX_BUFFERS 64
+
 struct ggml_tensor;
 struct ggml_cgraph;

@ -39,28 +40,27 @@ extern "C" {
 // user-code should use only these functions
 //

-GGML_BACKEND_API ggml_backend_t ggml_backend_metal_init(void);
+GGML_API void ggml_backend_metal_log_set_callback(ggml_log_callback log_callback, void * user_data);

-GGML_BACKEND_API bool ggml_backend_is_metal(ggml_backend_t backend);
+GGML_API ggml_backend_t ggml_backend_metal_init(void);

-GGML_DEPRECATED(
-        GGML_BACKEND_API ggml_backend_buffer_t ggml_backend_metal_buffer_from_ptr(void * data, size_t size, size_t max_size),
-        "obsoleted by the new device interface - https://github.com/ggml-org/llama.cpp/pull/9713");
+GGML_API bool ggml_backend_is_metal(ggml_backend_t backend);

-GGML_BACKEND_API void ggml_backend_metal_set_abort_callback(ggml_backend_t backend, ggml_abort_callback abort_callback, void * user_data);
+GGML_API GGML_CALL ggml_backend_buffer_t ggml_backend_metal_buffer_from_ptr(void * data, size_t size, size_t max_size);

-GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_metal_buffer_type(void);
+GGML_API void ggml_backend_metal_set_n_cb(ggml_backend_t backend, int n_cb);
+
+GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_metal_buffer_type(void);

 // helper to check if the device supports a specific family
 // ideally, the user code should be doing these checks
 // ref: https://developer.apple.com/metal/Metal-Feature-Set-Tables.pdf
-GGML_BACKEND_API bool ggml_backend_metal_supports_family(ggml_backend_t backend, int family);
+GGML_API bool ggml_backend_metal_supports_family(ggml_backend_t backend, int family);

 // capture all command buffers committed the next time `ggml_backend_graph_compute` is called
-GGML_BACKEND_API void ggml_backend_metal_capture_next_compute(ggml_backend_t backend);
-
-GGML_BACKEND_API ggml_backend_reg_t ggml_backend_metal_reg(void);
+GGML_API void ggml_backend_metal_capture_next_compute(ggml_backend_t backend);

 #ifdef __cplusplus
 }
 #endif
+
--- a/bindings/ruby/ext/ggml-opencl.h
+++ b/bindings/ruby/ext/ggml-opencl.h
@ -0,0 +1,36 @@
+#pragma once
+
+#include "ggml.h"
+#include "ggml-backend.h"
+
+#ifdef  __cplusplus
+extern "C" {
+#endif
+
+GGML_API void ggml_cl_init(void);
+
+GGML_API void   ggml_cl_mul(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst);
+GGML_API void   ggml_cl_add(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst);
+GGML_API bool   ggml_cl_can_mul_mat(const struct ggml_tensor * src0, const struct ggml_tensor * src1, const struct ggml_tensor * dst);
+GGML_API size_t ggml_cl_mul_mat_get_wsize(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst);
+GGML_API void   ggml_cl_mul_mat(const struct ggml_tensor * src0, const struct ggml_tensor * src1, struct ggml_tensor * dst, void * wdata, size_t wsize);
+
+// GGML_API void * ggml_cl_host_malloc(size_t size);
+// GGML_API void   ggml_cl_host_free(void * ptr);
+
+GGML_API void ggml_cl_free_data(const struct ggml_tensor* tensor);
+
+GGML_API void ggml_cl_transform_tensor(void * data, struct ggml_tensor * tensor);
+
+// backend API
+
+// GGML_API ggml_backend_t ggml_backend_opencl_init(void);
+
+// GGML_API bool ggml_backend_is_opencl(ggml_backend_t backend);
+
+GGML_API ggml_backend_buffer_type_t ggml_backend_opencl_buffer_type(void);
+// GGML_API ggml_backend_buffer_type_t ggml_backend_opencl_host_buffer_type(void);
+
+#ifdef  __cplusplus
+}
+#endif
--- a/bindings/ruby/ext/ggml-quants.c
+++ b/bindings/ruby/ext/ggml-quants.c
--- a/bindings/ruby/ext/ggml-quants.h
+++ b/bindings/ruby/ext/ggml-quants.h
@ -0,0 +1,133 @@
+#pragma once
+
+#define GGML_COMMON_DECL_C
+#include "ggml-common.h"
+
+#include "ggml.h"
+
+// GGML internal header
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// Quantization
+void quantize_row_q4_0_reference(const float * GGML_RESTRICT x, block_q4_0 * GGML_RESTRICT y, int64_t k);
+void quantize_row_q4_1_reference(const float * GGML_RESTRICT x, block_q4_1 * GGML_RESTRICT y, int64_t k);
+void quantize_row_q5_0_reference(const float * GGML_RESTRICT x, block_q5_0 * GGML_RESTRICT y, int64_t k);
+void quantize_row_q5_1_reference(const float * GGML_RESTRICT x, block_q5_1 * GGML_RESTRICT y, int64_t k);
+void quantize_row_q8_0_reference(const float * GGML_RESTRICT x, block_q8_0 * GGML_RESTRICT y, int64_t k);
+void quantize_row_q8_1_reference(const float * GGML_RESTRICT x, block_q8_1 * GGML_RESTRICT y, int64_t k);
+
+void quantize_row_q2_K_reference(const float * GGML_RESTRICT x, block_q2_K * GGML_RESTRICT y, int64_t k);
+void quantize_row_q3_K_reference(const float * GGML_RESTRICT x, block_q3_K * GGML_RESTRICT y, int64_t k);
+void quantize_row_q4_K_reference(const float * GGML_RESTRICT x, block_q4_K * GGML_RESTRICT y, int64_t k);
+void quantize_row_q5_K_reference(const float * GGML_RESTRICT x, block_q5_K * GGML_RESTRICT y, int64_t k);
+void quantize_row_q6_K_reference(const float * GGML_RESTRICT x, block_q6_K * GGML_RESTRICT y, int64_t k);
+void quantize_row_q8_K_reference(const float * GGML_RESTRICT x, block_q8_K * GGML_RESTRICT y, int64_t k);
+
+void quantize_row_iq3_xxs_reference(const float * GGML_RESTRICT x, block_iq3_xxs * GGML_RESTRICT y, int64_t k);
+void quantize_row_iq4_nl_reference (const float * GGML_RESTRICT x, block_iq4_nl  * GGML_RESTRICT y, int64_t k);
+void quantize_row_iq4_xs_reference (const float * GGML_RESTRICT x, block_iq4_xs  * GGML_RESTRICT y, int64_t k);
+void quantize_row_iq3_s_reference  (const float * GGML_RESTRICT x, block_iq3_s   * GGML_RESTRICT y, int64_t k);
+void quantize_row_iq2_s_reference  (const float * GGML_RESTRICT x, block_iq2_s   * GGML_RESTRICT y, int64_t k);
+
+void quantize_row_q4_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q4_1(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q5_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q5_1(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q8_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q8_1(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+
+void quantize_row_q2_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q3_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q4_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q5_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q6_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_q8_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+
+void quantize_row_iq3_xxs(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_iq4_nl (const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_iq4_xs (const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_iq3_s  (const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+void quantize_row_iq2_s  (const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
+
+// Dequantization
+void dequantize_row_q4_0(const block_q4_0 * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_q4_1(const block_q4_1 * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_q5_0(const block_q5_0 * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_q5_1(const block_q5_1 * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_q8_0(const block_q8_0 * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+//void dequantize_row_q8_1(const block_q8_1 * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+
+void dequantize_row_q2_K(const block_q2_K * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_q3_K(const block_q3_K * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_q4_K(const block_q4_K * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_q5_K(const block_q5_K * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_q6_K(const block_q6_K * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_q8_K(const block_q8_K * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+
+void dequantize_row_iq2_xxs(const block_iq2_xxs * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_iq2_xs (const block_iq2_xs  * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_iq2_s  (const block_iq2_s   * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_iq3_xxs(const block_iq3_xxs * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_iq1_s  (const block_iq1_s   * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_iq1_m  (const block_iq1_m   * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_iq4_nl (const block_iq4_nl  * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_iq4_xs (const block_iq4_xs  * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+void dequantize_row_iq3_s  (const block_iq3_s   * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
+
+// Dot product
+void ggml_vec_dot_q4_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q4_1_q8_1(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q5_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q5_1_q8_1(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q8_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+
+void ggml_vec_dot_q2_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q3_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q4_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q5_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_q6_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+
+void ggml_vec_dot_iq2_xxs_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq2_xs_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq2_s_q8_K  (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq3_xxs_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq1_s_q8_K  (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq1_m_q8_K  (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq4_nl_q8_0 (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq4_xs_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+void ggml_vec_dot_iq3_s_q8_K  (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+
+// Quantization utilizing an importance matrix (a.k.a. "Activation aWare Quantization")
+size_t quantize_iq2_xxs(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_iq2_xs (const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_iq2_s  (const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_iq3_xxs(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_iq1_s  (const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_iq1_m  (const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_iq4_nl (const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_iq4_xs (const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_iq3_s  (const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+
+size_t quantize_q2_K(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_q3_K(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_q4_K(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_q5_K(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_q6_K(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_q4_0(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_q4_1(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_q5_0(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_q5_1(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+size_t quantize_q8_0(const float * GGML_RESTRICT src, void * GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
+
+void iq2xs_init_impl(enum ggml_type type);
+void iq2xs_free_impl(enum ggml_type type);
+void iq3xs_init_impl(int grid_size);
+void iq3xs_free_impl(int grid_size);
+
+#ifdef __cplusplus
+}
+#endif
+
--- a/bindings/ruby/ext/ggml-sycl.h
+++ b/bindings/ruby/ext/ggml-sycl.h
@ -0,0 +1,49 @@
+//
+//  MIT license
+//  Copyright (C) 2024 Intel Corporation
+//  SPDX-License-Identifier: MIT
+//
+
+#pragma once
+
+#include "ggml.h"
+#include "ggml-backend.h"
+
+#ifdef  __cplusplus
+extern "C" {
+#endif
+
+#define GGML_SYCL_MAX_DEVICES       48
+#define GGML_SYCL_NAME "SYCL"
+
+// backend API
+GGML_API ggml_backend_t ggml_backend_sycl_init(int device);
+
+// devide buffer
+GGML_API ggml_backend_buffer_type_t ggml_backend_sycl_buffer_type(int device);
+
+// split tensor buffer that splits matrices by rows across multiple devices
+GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_sycl_split_buffer_type(const float * tensor_split);
+
+// pinned host buffer for use with the CPU backend for faster copies between CPU and GPU
+GGML_API ggml_backend_buffer_type_t ggml_backend_sycl_host_buffer_type(void);
+
+GGML_API void   ggml_backend_sycl_print_sycl_devices(void);
+GGML_API GGML_CALL void   ggml_sycl_get_gpu_list(int *id_list, int max_len);
+GGML_API GGML_CALL void   ggml_sycl_get_device_description(int device, char *description, size_t description_size);
+GGML_API GGML_CALL int   ggml_backend_sycl_get_device_count();
+GGML_API GGML_CALL void ggml_backend_sycl_get_device_memory(int device, size_t *free, size_t *total);
+GGML_API GGML_CALL int ggml_backend_sycl_get_device_index(int device_id);
+
+// TODO: these are temporary
+//       ref: https://github.com/ggerganov/llama.cpp/pull/6022#issuecomment-1992615670
+GGML_API GGML_CALL int ggml_backend_sycl_get_device_id(int device_index);
+GGML_API GGML_CALL void ggml_backend_sycl_set_single_device_mode(int main_gpu_id);
+GGML_API GGML_CALL void ggml_backend_sycl_set_mul_device_mode();
+
+// SYCL doesn't support registering host memory, keep here for reference
+// GGML_API GGML_CALL bool ggml_backend_sycl_register_host_buffer(void * buffer, size_t size);
+// GGML_API GGML_CALL void ggml_backend_sycl_unregister_host_buffer(void * buffer);
+#ifdef  __cplusplus
+}
+#endif
--- a/bindings/ruby/ext/ggml-vulkan.h
+++ b/bindings/ruby/ext/ggml-vulkan.h
@ -0,0 +1,29 @@
+#pragma once
+
+#include "ggml.h"
+#include "ggml-backend.h"
+
+#ifdef  __cplusplus
+extern "C" {
+#endif
+
+#define GGML_VK_NAME "Vulkan"
+#define GGML_VK_MAX_DEVICES 16
+
+GGML_API void ggml_vk_instance_init(void);
+
+// backend API
+GGML_API GGML_CALL ggml_backend_t ggml_backend_vk_init(size_t dev_num);
+
+GGML_API GGML_CALL bool ggml_backend_is_vk(ggml_backend_t backend);
+GGML_API GGML_CALL int  ggml_backend_vk_get_device_count(void);
+GGML_API GGML_CALL void ggml_backend_vk_get_device_description(int device, char * description, size_t description_size);
+GGML_API GGML_CALL void ggml_backend_vk_get_device_memory(int device, size_t * free, size_t * total);
+
+GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_vk_buffer_type(size_t dev_num);
+// pinned host buffer for use with the CPU backend for faster copies between CPU and GPU
+GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_vk_host_buffer_type(void);
+
+#ifdef  __cplusplus
+}
+#endif
--- a/bindings/ruby/ext/options.rb
+++ b/bindings/ruby/ext/options.rb
@ -1,82 +0,0 @@
-class Options
-  def initialize(cmake="cmake")
-    @cmake = cmake
-    @options = {}
-
-    configure
-  end
-
-  def to_s
-    @options
-      .reject {|name, (type, value)| value.nil?}
-      .collect {|name, (type, value)| "-D #{name}=#{value == true ? "ON" : value == false ? "OFF" : value.shellescape}"}
-      .join(" ")
-  end
-
-  def cmake_options
-    return @cmake_options if @cmake_options
-
-    output = nil
-    Dir.chdir __dir__ do
-      output = `#{@cmake.shellescape} -S sources -B build -L`
-    end
-    @cmake_options = output.lines.drop_while {|line| line.chomp != "-- Cache values"}.drop(1)
-                       .filter_map {|line|
-                         option, value = line.chomp.split("=", 2)
-                         name, type = option.split(":", 2)
-                         [
-                           name,
-                           [
-                             type,
-                             type == "BOOL" ? value == "ON" : value
-                           ]
-                         ]
-                       }.to_h
-  end
-
-  private
-
-  def configure
-    cmake_options.each_pair do |name, (type, default_value)|
-      option = option_name(name)
-      value = type == "BOOL" ? enable_config(option) : arg_config("--#{option}")
-      @options[name] = [type, value]
-    end
-
-    configure_accelerate
-    configure_metal
-    configure_coreml
-  end
-
-  # See ggml/src/ggml-cpu/CMakeLists.txt
-  def configure_accelerate
-    if RUBY_PLATFORM.match?(/darwin/) && enabled?("GGML_ACCELERATE")
-      $LDFLAGS << " -framework Accelerate"
-    end
-  end
-
-  # See ggml/src/ggml-metal/CMakeLists.txt
-  def configure_metal
-    $LDFLAGS << " -framework Foundation -framework Metal -framework MetalKit" if enabled?("GGML_METAL")
-  end
-
-  # See src/CmakeLists.txt
-  def configure_coreml
-    if enabled?("WHISPER_COREML")
-      $LDFLAGS << " -framework Foundation -framework CoreML"
-      $CPPFLAGS << " -DRUBY_WHISPER_USE_COREML"
-    end
-  end
-
-  def option_name(name)
-    name.downcase.gsub("_", "-")
-  end
-
-  def enabled?(option)
-    if @options[option][1].nil?
-      cmake_options[option][1]
-    else
-      @options[option][1]
-    end
-  end
-end
--- a/bindings/ruby/ext/ruby_whisper.c
+++ b/bindings/ruby/ext/ruby_whisper.c
@ -1,176 +0,0 @@
-#include <ruby.h>
-#include <ruby/memory_view.h>
-#include "ruby_whisper.h"
-
-VALUE mWhisper;
-VALUE mVAD;
-VALUE cContext;
-VALUE cParams;
-VALUE cVADParams;
-VALUE eError;
-
-VALUE cSegment;
-VALUE cModel;
-
-ID id_to_s;
-ID id_call;
-ID id___method__;
-ID id_to_enum;
-ID id_length;
-ID id_next;
-ID id_new;
-ID id_to_path;
-ID id_URI;
-ID id_pre_converted_models;
-ID id_coreml_compiled_models;
-ID id_cache;
-ID id_n_processors;
-
-static bool is_log_callback_finalized = false;
-
-// High level API
-extern VALUE ruby_whisper_segment_allocate(VALUE klass);
-
-extern void init_ruby_whisper_context(VALUE *mWhisper);
-extern void init_ruby_whisper_params(VALUE *mWhisper);
-extern void init_ruby_whisper_error(VALUE *mWhisper);
-extern void init_ruby_whisper_segment(VALUE *mWhisper, VALUE *cSegment);
-extern void init_ruby_whisper_model(VALUE *mWhisper);
-extern void init_ruby_whisper_vad_params(VALUE *mVAD);
-extern void register_callbacks(ruby_whisper_params *rwp, VALUE *context);
-
-/*
- * call-seq:
- *   lang_max_id -> Integer
- */
-static VALUE ruby_whisper_s_lang_max_id(VALUE self) {
-  return INT2NUM(whisper_lang_max_id());
-}
-
-/*
- * call-seq:
- *   lang_id(lang_name) -> Integer
- */
-static VALUE ruby_whisper_s_lang_id(VALUE self, VALUE lang) {
-  const char * lang_str = StringValueCStr(lang);
-  const int id = whisper_lang_id(lang_str);
-  if (-1 == id) {
-    rb_raise(rb_eArgError, "language not found: %s", lang_str);
-  }
-  return INT2NUM(id);
-}
-
-/*
- * call-seq:
- *   lang_str(lang_id) -> String
- */
-static VALUE ruby_whisper_s_lang_str(VALUE self, VALUE id) {
-  const int lang_id = NUM2INT(id);
-  const char * str = whisper_lang_str(lang_id);
-  if (NULL == str) {
-    rb_raise(rb_eIndexError, "id %d outside of language id", lang_id);
-  }
-  return rb_str_new2(str);
-}
-
-/*
- * call-seq:
- *   lang_str(lang_id) -> String
- */
-static VALUE ruby_whisper_s_lang_str_full(VALUE self, VALUE id) {
-  const int lang_id = NUM2INT(id);
-  const char * str_full = whisper_lang_str_full(lang_id);
-  if (NULL == str_full) {
-    rb_raise(rb_eIndexError, "id %d outside of language id", lang_id);
-  }
-  return rb_str_new2(str_full);
-}
-
-/*
- * call-seq:
- *   system_info_str -> String
- */
-static VALUE ruby_whisper_s_system_info_str(VALUE self) {
-  return rb_str_new2(whisper_print_system_info());
-}
-
-static VALUE ruby_whisper_s_finalize_log_callback(VALUE self, VALUE id) {
-  is_log_callback_finalized = true;
-  return Qnil;
-}
-
-static void
-ruby_whisper_log_callback(enum ggml_log_level level, const char * buffer, void * user_data) {
-  if (is_log_callback_finalized) {
-    return;
-  }
-  VALUE log_callback = rb_iv_get(mWhisper, "log_callback");
-  VALUE udata = rb_iv_get(mWhisper, "user_data");
-  rb_funcall(log_callback, id_call, 3, INT2NUM(level), rb_str_new2(buffer), udata);
-}
-
-/*
- * call-seq:
- *   log_set ->(level, buffer, user_data) { ... }, user_data -> nil
- */
-static VALUE ruby_whisper_s_log_set(VALUE self, VALUE log_callback, VALUE user_data) {
-  VALUE old_callback = rb_iv_get(self, "log_callback");
-  if (!NIL_P(old_callback)) {
-    rb_undefine_finalizer(old_callback);
-  }
-
-  rb_iv_set(self, "log_callback", log_callback);
-  rb_iv_set(self, "user_data", user_data);
-
-  VALUE finalize_log_callback = rb_funcall(mWhisper, rb_intern("method"), 1, rb_str_new2("finalize_log_callback"));
-  rb_define_finalizer(log_callback, finalize_log_callback);
-
-  whisper_log_set(ruby_whisper_log_callback, NULL);
-
-  return Qnil;
-}
-
-void Init_whisper() {
-  id_to_s = rb_intern("to_s");
-  id_call = rb_intern("call");
-  id___method__ = rb_intern("__method__");
-  id_to_enum = rb_intern("to_enum");
-  id_length = rb_intern("length");
-  id_next = rb_intern("next");
-  id_new = rb_intern("new");
-  id_to_path = rb_intern("to_path");
-  id_URI = rb_intern("URI");
-  id_pre_converted_models = rb_intern("pre_converted_models");
-  id_coreml_compiled_models = rb_intern("coreml_compiled_models");
-  id_cache = rb_intern("cache");
-  id_n_processors = rb_intern("n_processors");
-
-  mWhisper = rb_define_module("Whisper");
-  mVAD = rb_define_module_under(mWhisper, "VAD");
-
-  rb_define_const(mWhisper, "LOG_LEVEL_NONE", INT2NUM(GGML_LOG_LEVEL_NONE));
-  rb_define_const(mWhisper, "LOG_LEVEL_INFO", INT2NUM(GGML_LOG_LEVEL_INFO));
-  rb_define_const(mWhisper, "LOG_LEVEL_WARN", INT2NUM(GGML_LOG_LEVEL_WARN));
-  rb_define_const(mWhisper, "LOG_LEVEL_ERROR", INT2NUM(GGML_LOG_LEVEL_ERROR));
-  rb_define_const(mWhisper, "LOG_LEVEL_DEBUG", INT2NUM(GGML_LOG_LEVEL_DEBUG));
-  rb_define_const(mWhisper, "LOG_LEVEL_CONT", INT2NUM(GGML_LOG_LEVEL_CONT));
-
-  rb_define_singleton_method(mWhisper, "lang_max_id", ruby_whisper_s_lang_max_id, 0);
-  rb_define_singleton_method(mWhisper, "lang_id", ruby_whisper_s_lang_id, 1);
-  rb_define_singleton_method(mWhisper, "lang_str", ruby_whisper_s_lang_str, 1);
-  rb_define_singleton_method(mWhisper, "lang_str_full", ruby_whisper_s_lang_str_full, 1);
-  rb_define_singleton_method(mWhisper, "system_info_str", ruby_whisper_s_system_info_str, 0);
-  rb_define_singleton_method(mWhisper, "log_set", ruby_whisper_s_log_set, 2);
-  rb_define_private_method(rb_singleton_class(mWhisper), "finalize_log_callback", ruby_whisper_s_finalize_log_callback, 1);
-
-  init_ruby_whisper_context(&mWhisper);
-  init_ruby_whisper_params(&mWhisper);
-  init_ruby_whisper_error(&mWhisper);
-  init_ruby_whisper_segment(&mWhisper, &cContext);
-  init_ruby_whisper_model(&mWhisper);
-  init_ruby_whisper_vad_params(&mVAD);
-
-  rb_require("whisper/context");
-  rb_require("whisper/segment");
-  rb_require("whisper/model/uri");
-}
--- a/bindings/ruby/ext/ruby_whisper.cpp
+++ b/bindings/ruby/ext/ruby_whisper.cpp
@ -0,0 +1,418 @@
+#include <ruby.h>
+#include "ruby_whisper.h"
+#define DR_WAV_IMPLEMENTATION
+#include "dr_wav.h"
+#include <cmath>
+#include <fstream>
+#include <cstdio>
+#include <string>
+#include <thread>
+#include <vector>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define BOOL_PARAMS_SETTER(self, prop, value) \
+  ruby_whisper_params *rwp; \
+  Data_Get_Struct(self, ruby_whisper_params, rwp); \
+  if (value == Qfalse || value == Qnil) { \
+    rwp->params.prop = false; \
+  } else { \
+    rwp->params.prop = true; \
+  } \
+  return value; \
+
+#define BOOL_PARAMS_GETTER(self,  prop) \
+  ruby_whisper_params *rwp; \
+  Data_Get_Struct(self, ruby_whisper_params, rwp); \
+  if (rwp->params.prop) { \
+    return Qtrue; \
+  } else { \
+    return Qfalse; \
+  }
+
+VALUE mWhisper;
+VALUE cContext;
+VALUE cParams;
+
+static void ruby_whisper_free(ruby_whisper *rw) {
+  if (rw->context) {
+    whisper_free(rw->context);
+    rw->context = NULL;
+  }
+}
+static void ruby_whisper_params_free(ruby_whisper_params *rwp) {
+}
+
+void rb_whisper_mark(ruby_whisper *rw) {
+  // call rb_gc_mark on any ruby references in rw
+}
+
+void rb_whisper_free(ruby_whisper *rw) {
+  ruby_whisper_free(rw);
+  free(rw);
+}
+
+void rb_whisper_params_mark(ruby_whisper_params *rwp) {
+}
+
+void rb_whisper_params_free(ruby_whisper_params *rwp) {
+  ruby_whisper_params_free(rwp);
+  free(rwp);
+}
+
+static VALUE ruby_whisper_allocate(VALUE klass) {
+  ruby_whisper *rw;
+  rw = ALLOC(ruby_whisper);
+  rw->context = NULL;
+  return Data_Wrap_Struct(klass, rb_whisper_mark, rb_whisper_free, rw);
+}
+
+static VALUE ruby_whisper_params_allocate(VALUE klass) {
+  ruby_whisper_params *rwp;
+  rwp = ALLOC(ruby_whisper_params);
+  rwp->params = whisper_full_default_params(WHISPER_SAMPLING_GREEDY);
+  return Data_Wrap_Struct(klass, rb_whisper_params_mark, rb_whisper_params_free, rwp);
+}
+
+static VALUE ruby_whisper_initialize(int argc, VALUE *argv, VALUE self) {
+  ruby_whisper *rw;
+  VALUE whisper_model_file_path;
+
+  // TODO: we can support init from buffer here too maybe another ruby object to expose
+  rb_scan_args(argc, argv, "01", &whisper_model_file_path);
+  Data_Get_Struct(self, ruby_whisper, rw);
+
+  if (!rb_respond_to(whisper_model_file_path, rb_intern("to_s"))) {
+    rb_raise(rb_eRuntimeError, "Expected file path to model to initialize Whisper::Context");
+  }
+  rw->context = whisper_init_from_file_with_params(StringValueCStr(whisper_model_file_path), whisper_context_default_params());
+  if (rw->context == nullptr) {
+    rb_raise(rb_eRuntimeError, "error: failed to initialize whisper context");
+  }
+  return self;
+}
+
+/*
+ * transcribe a single file
+ * can emit to a block results
+ *
+ **/
+static VALUE ruby_whisper_transcribe(int argc, VALUE *argv, VALUE self) {
+  ruby_whisper *rw;
+  ruby_whisper_params *rwp;
+  VALUE wave_file_path, blk, params;
+
+  rb_scan_args(argc, argv, "02&", &wave_file_path, &params, &blk);
+  Data_Get_Struct(self, ruby_whisper, rw);
+  Data_Get_Struct(params, ruby_whisper_params, rwp);
+
+  if (!rb_respond_to(wave_file_path, rb_intern("to_s"))) {
+    rb_raise(rb_eRuntimeError, "Expected file path to wave file");
+  }
+
+  std::string fname_inp = StringValueCStr(wave_file_path);
+
+  std::vector<float> pcmf32; // mono-channel F32 PCM
+  std::vector<std::vector<float>> pcmf32s; // stereo-channel F32 PCM
+
+  // WAV input - this is directly from main.cpp example
+  {
+    drwav wav;
+    std::vector<uint8_t> wav_data; // used for pipe input from stdin
+
+    if (fname_inp == "-") {
+      {
+        uint8_t buf[1024];
+        while (true) {
+          const size_t n = fread(buf, 1, sizeof(buf), stdin);
+          if (n == 0) {
+            break;
+          }
+          wav_data.insert(wav_data.end(), buf, buf + n);
+        }
+      }
+
+      if (drwav_init_memory(&wav, wav_data.data(), wav_data.size(), nullptr) == false) {
+        fprintf(stderr, "error: failed to open WAV file from stdin\n");
+        return self;
+      }
+
+      fprintf(stderr, "%s: read %zu bytes from stdin\n", __func__, wav_data.size());
+    } else if (drwav_init_file(&wav, fname_inp.c_str(), nullptr) == false) {
+      fprintf(stderr, "error: failed to open '%s' as WAV file\n", fname_inp.c_str());
+      return self;
+    }
+
+    if (wav.channels != 1 && wav.channels != 2) {
+      fprintf(stderr, "WAV file '%s' must be mono or stereo\n", fname_inp.c_str());
+      return self;
+    }
+
+    if (rwp->diarize && wav.channels != 2 && rwp->params.print_timestamps == false) {
+      fprintf(stderr, "WAV file '%s' must be stereo for diarization and timestamps have to be enabled\n", fname_inp.c_str());
+      return self;
+    }
+
+    if (wav.sampleRate != WHISPER_SAMPLE_RATE) {
+      fprintf(stderr, "WAV file '%s' must be %i kHz\n", fname_inp.c_str(), WHISPER_SAMPLE_RATE/1000);
+      return self;
+    }
+
+    if (wav.bitsPerSample != 16) {
+      fprintf(stderr, "WAV file '%s' must be 16-bit\n", fname_inp.c_str());
+      return self;
+    }
+
+    const uint64_t n = wav_data.empty() ? wav.totalPCMFrameCount : wav_data.size()/(wav.channels*wav.bitsPerSample/8);
+
+    std::vector<int16_t> pcm16;
+    pcm16.resize(n*wav.channels);
+    drwav_read_pcm_frames_s16(&wav, n, pcm16.data());
+    drwav_uninit(&wav);
+
+    // convert to mono, float
+    pcmf32.resize(n);
+    if (wav.channels == 1) {
+      for (uint64_t i = 0; i < n; i++) {
+        pcmf32[i] = float(pcm16[i])/32768.0f;
+      }
+    } else {
+      for (uint64_t i = 0; i < n; i++) {
+        pcmf32[i] = float(pcm16[2*i] + pcm16[2*i + 1])/65536.0f;
+      }
+    }
+
+    if (rwp->diarize) {
+      // convert to stereo, float
+      pcmf32s.resize(2);
+
+      pcmf32s[0].resize(n);
+      pcmf32s[1].resize(n);
+      for (uint64_t i = 0; i < n; i++) {
+        pcmf32s[0][i] = float(pcm16[2*i])/32768.0f;
+        pcmf32s[1][i] = float(pcm16[2*i + 1])/32768.0f;
+      }
+    }
+  }
+  {
+    static bool is_aborted = false; // NOTE: this should be atomic to avoid data race
+
+    rwp->params.encoder_begin_callback = [](struct whisper_context * /*ctx*/, struct whisper_state * /*state*/, void * user_data) {
+      bool is_aborted = *(bool*)user_data;
+      return !is_aborted;
+    };
+    rwp->params.encoder_begin_callback_user_data = &is_aborted;
+  }
+
+  if (whisper_full_parallel(rw->context, rwp->params, pcmf32.data(), pcmf32.size(), 1) != 0) {
+    fprintf(stderr, "failed to process audio\n");
+    return self;
+  }
+  const int n_segments = whisper_full_n_segments(rw->context);
+  VALUE output = rb_str_new2("");
+  for (int i = 0; i < n_segments; ++i) {
+    const char * text = whisper_full_get_segment_text(rw->context, i);
+    output = rb_str_concat(output, rb_str_new2(text));
+  }
+  VALUE idCall = rb_intern("call");
+  if (blk != Qnil) {
+    rb_funcall(blk, idCall, 1, output);
+  }
+  return self;
+}
+
+/*
+ * params.language = "auto" | "en", etc...
+ */
+static VALUE ruby_whisper_params_set_language(VALUE self, VALUE value) {
+  ruby_whisper_params *rwp;
+  Data_Get_Struct(self, ruby_whisper_params, rwp);
+  if (value == Qfalse || value == Qnil) {
+    rwp->params.language = "auto";
+  } else {
+    rwp->params.language = StringValueCStr(value);
+  }
+  return value;
+}
+static VALUE ruby_whisper_params_get_language(VALUE self) {
+  ruby_whisper_params *rwp;
+  Data_Get_Struct(self, ruby_whisper_params, rwp);
+  if (rwp->params.language) {
+    return rb_str_new2(rwp->params.language);
+  } else {
+    return rb_str_new2("auto");
+  }
+}
+static VALUE ruby_whisper_params_set_translate(VALUE self, VALUE value) {
+  BOOL_PARAMS_SETTER(self, translate, value)
+}
+static VALUE ruby_whisper_params_get_translate(VALUE self) {
+  BOOL_PARAMS_GETTER(self, translate)
+}
+static VALUE ruby_whisper_params_set_no_context(VALUE self, VALUE value) {
+  BOOL_PARAMS_SETTER(self, no_context, value)
+}
+static VALUE ruby_whisper_params_get_no_context(VALUE self) {
+  BOOL_PARAMS_GETTER(self, no_context)
+}
+static VALUE ruby_whisper_params_set_single_segment(VALUE self, VALUE value) {
+  BOOL_PARAMS_SETTER(self, single_segment, value)
+}
+static VALUE ruby_whisper_params_get_single_segment(VALUE self) {
+  BOOL_PARAMS_GETTER(self, single_segment)
+}
+static VALUE ruby_whisper_params_set_print_special(VALUE self, VALUE value) {
+  BOOL_PARAMS_SETTER(self, print_special, value)
+}
+static VALUE ruby_whisper_params_get_print_special(VALUE self) {
+  BOOL_PARAMS_GETTER(self, print_special)
+}
+static VALUE ruby_whisper_params_set_print_progress(VALUE self, VALUE value) {
+  BOOL_PARAMS_SETTER(self, print_progress, value)
+}
+static VALUE ruby_whisper_params_get_print_progress(VALUE self) {
+  BOOL_PARAMS_GETTER(self, print_progress)
+}
+static VALUE ruby_whisper_params_set_print_realtime(VALUE self, VALUE value) {
+  BOOL_PARAMS_SETTER(self, print_realtime, value)
+}
+static VALUE ruby_whisper_params_get_print_realtime(VALUE self) {
+  BOOL_PARAMS_GETTER(self, print_realtime)
+}
+static VALUE ruby_whisper_params_set_print_timestamps(VALUE self, VALUE value) {
+  BOOL_PARAMS_SETTER(self, print_timestamps, value)
+}
+static VALUE ruby_whisper_params_get_print_timestamps(VALUE self) {
+  BOOL_PARAMS_GETTER(self, print_timestamps)
+}
+static VALUE ruby_whisper_params_set_suppress_blank(VALUE self, VALUE value) {
+  BOOL_PARAMS_SETTER(self, suppress_blank, value)
+}
+static VALUE ruby_whisper_params_get_suppress_blank(VALUE self) {
+  BOOL_PARAMS_GETTER(self, suppress_blank)
+}
+static VALUE ruby_whisper_params_set_suppress_non_speech_tokens(VALUE self, VALUE value) {
+  BOOL_PARAMS_SETTER(self, suppress_non_speech_tokens, value)
+}
+static VALUE ruby_whisper_params_get_suppress_non_speech_tokens(VALUE self) {
+  BOOL_PARAMS_GETTER(self, suppress_non_speech_tokens)
+}
+static VALUE ruby_whisper_params_get_token_timestamps(VALUE self) {
+  BOOL_PARAMS_GETTER(self, token_timestamps)
+}
+static VALUE ruby_whisper_params_set_token_timestamps(VALUE self, VALUE value) {
+  BOOL_PARAMS_SETTER(self, token_timestamps, value)
+}
+static VALUE ruby_whisper_params_get_split_on_word(VALUE self) {
+  BOOL_PARAMS_GETTER(self, split_on_word)
+}
+static VALUE ruby_whisper_params_set_split_on_word(VALUE self, VALUE value) {
+  BOOL_PARAMS_SETTER(self, split_on_word, value)
+}
+static VALUE ruby_whisper_params_get_diarize(VALUE self) {
+  ruby_whisper_params *rwp;
+  Data_Get_Struct(self, ruby_whisper_params, rwp);
+  if (rwp->diarize) {
+    return Qtrue;
+  } else {
+    return Qfalse;
+  }
+}
+static VALUE ruby_whisper_params_set_diarize(VALUE self, VALUE value) {
+  ruby_whisper_params *rwp;
+  Data_Get_Struct(self, ruby_whisper_params, rwp);
+  if (value == Qfalse || value == Qnil) {
+    rwp->diarize = false;
+  } else {
+    rwp->diarize = true;
+  } \
+  return value;
+}
+
+static VALUE ruby_whisper_params_get_offset(VALUE self) {
+  ruby_whisper_params *rwp;
+  Data_Get_Struct(self, ruby_whisper_params, rwp);
+  return INT2NUM(rwp->params.offset_ms);
+}
+static VALUE ruby_whisper_params_set_offset(VALUE self, VALUE value) {
+  ruby_whisper_params *rwp;
+  Data_Get_Struct(self, ruby_whisper_params, rwp);
+  rwp->params.offset_ms = NUM2INT(value);
+  return value;
+}
+static VALUE ruby_whisper_params_get_duration(VALUE self) {
+  ruby_whisper_params *rwp;
+  Data_Get_Struct(self, ruby_whisper_params, rwp);
+  return INT2NUM(rwp->params.duration_ms);
+}
+static VALUE ruby_whisper_params_set_duration(VALUE self, VALUE value) {
+  ruby_whisper_params *rwp;
+  Data_Get_Struct(self, ruby_whisper_params, rwp);
+  rwp->params.duration_ms = NUM2INT(value);
+  return value;
+}
+
+static VALUE ruby_whisper_params_get_max_text_tokens(VALUE self) {
+  ruby_whisper_params *rwp;
+  Data_Get_Struct(self, ruby_whisper_params, rwp);
+  return INT2NUM(rwp->params.n_max_text_ctx);
+}
+static VALUE ruby_whisper_params_set_max_text_tokens(VALUE self, VALUE value) {
+  ruby_whisper_params *rwp;
+  Data_Get_Struct(self, ruby_whisper_params, rwp);
+  rwp->params.n_max_text_ctx = NUM2INT(value);
+  return value;
+}
+
+void Init_whisper() {
+  mWhisper = rb_define_module("Whisper");
+  cContext = rb_define_class_under(mWhisper, "Context", rb_cObject);
+  cParams  = rb_define_class_under(mWhisper, "Params", rb_cObject);
+
+  rb_define_alloc_func(cContext, ruby_whisper_allocate);
+  rb_define_method(cContext, "initialize", ruby_whisper_initialize, -1);
+
+  rb_define_method(cContext, "transcribe", ruby_whisper_transcribe, -1);
+
+  rb_define_alloc_func(cParams, ruby_whisper_params_allocate);
+
+  rb_define_method(cParams, "language=", ruby_whisper_params_set_language, 1);
+  rb_define_method(cParams, "language", ruby_whisper_params_get_language, 0);
+  rb_define_method(cParams, "translate=", ruby_whisper_params_set_translate, 1);
+  rb_define_method(cParams, "translate", ruby_whisper_params_get_translate, 0);
+  rb_define_method(cParams, "no_context=", ruby_whisper_params_set_no_context, 1);
+  rb_define_method(cParams, "no_context", ruby_whisper_params_get_no_context, 0);
+  rb_define_method(cParams, "single_segment=", ruby_whisper_params_set_single_segment, 1);
+  rb_define_method(cParams, "single_segment", ruby_whisper_params_get_single_segment, 0);
+  rb_define_method(cParams, "print_special", ruby_whisper_params_get_print_special, 0);
+  rb_define_method(cParams, "print_special=", ruby_whisper_params_set_print_special, 1);
+  rb_define_method(cParams, "print_progress", ruby_whisper_params_get_print_progress, 0);
+  rb_define_method(cParams, "print_progress=", ruby_whisper_params_set_print_progress, 1);
+  rb_define_method(cParams, "print_realtime", ruby_whisper_params_get_print_realtime, 0);
+  rb_define_method(cParams, "print_realtime=", ruby_whisper_params_set_print_realtime, 1);
+  rb_define_method(cParams, "print_timestamps", ruby_whisper_params_get_print_timestamps, 0);
+  rb_define_method(cParams, "print_timestamps=", ruby_whisper_params_set_print_timestamps, 1);
+  rb_define_method(cParams, "suppress_blank", ruby_whisper_params_get_suppress_blank, 0);
+  rb_define_method(cParams, "suppress_blank=", ruby_whisper_params_set_suppress_blank, 1);
+  rb_define_method(cParams, "suppress_non_speech_tokens", ruby_whisper_params_get_suppress_non_speech_tokens, 0);
+  rb_define_method(cParams, "suppress_non_speech_tokens=", ruby_whisper_params_set_suppress_non_speech_tokens, 1);
+  rb_define_method(cParams, "token_timestamps", ruby_whisper_params_get_token_timestamps, 0);
+  rb_define_method(cParams, "token_timestamps=", ruby_whisper_params_set_token_timestamps, 1);
+  rb_define_method(cParams, "split_on_word", ruby_whisper_params_get_split_on_word, 0);
+  rb_define_method(cParams, "split_on_word=", ruby_whisper_params_set_split_on_word, 1);
+  rb_define_method(cParams, "diarize", ruby_whisper_params_get_diarize, 0);
+  rb_define_method(cParams, "diarize=", ruby_whisper_params_set_diarize, 1);
+
+  rb_define_method(cParams, "offset", ruby_whisper_params_get_offset, 0);
+  rb_define_method(cParams, "offset=", ruby_whisper_params_set_offset, 1);
+  rb_define_method(cParams, "duration", ruby_whisper_params_get_duration, 0);
+  rb_define_method(cParams, "duration=", ruby_whisper_params_set_duration, 1);
+
+  rb_define_method(cParams, "max_text_tokens", ruby_whisper_params_get_max_text_tokens, 0);
+  rb_define_method(cParams, "max_text_tokens=", ruby_whisper_params_set_max_text_tokens, 1);
+}
+#ifdef __cplusplus
+}
+#endif
--- a/bindings/ruby/ext/ruby_whisper.h
+++ b/bindings/ruby/ext/ruby_whisper.h
@ -1,15 +1,8 @@
-#ifndef RUBY_WHISPER_H
-#define RUBY_WHISPER_H
+#ifndef __RUBY_WHISPER_H
+#define __RUBY_WHISPER_H

 #include "whisper.h"

-typedef struct {
-  VALUE *context;
-  VALUE user_data;
-  VALUE callback;
-  VALUE callbacks;
-} ruby_whisper_callback_container;
-
 typedef struct {
  struct whisper_context *context;
 } ruby_whisper;
@ -17,24 +10,6 @@ typedef struct {
 typedef struct {
  struct whisper_full_params params;
  bool diarize;
-  ruby_whisper_callback_container *new_segment_callback_container;
-  ruby_whisper_callback_container *progress_callback_container;
-  ruby_whisper_callback_container *encoder_begin_callback_container;
-  ruby_whisper_callback_container *abort_callback_container;
-  VALUE vad_params;
 } ruby_whisper_params;

-typedef struct {
-  struct whisper_vad_params params;
-} ruby_whisper_vad_params;
-
-typedef struct {
-  VALUE context;
-  int index;
-} ruby_whisper_segment;
-
-typedef struct {
-  VALUE context;
-} ruby_whisper_model;
-
 #endif
--- a/bindings/ruby/ext/ruby_whisper_context.c
+++ b/bindings/ruby/ext/ruby_whisper_context.c
@ -1,672 +0,0 @@
-#include <ruby.h>
-#include <ruby/memory_view.h>
-#include "ruby_whisper.h"
-
-extern ID id_to_s;
-extern ID id___method__;
-extern ID id_to_enum;
-extern ID id_length;
-extern ID id_next;
-extern ID id_new;
-extern ID id_to_path;
-extern ID id_URI;
-extern ID id_pre_converted_models;
-extern ID id_coreml_compiled_models;
-extern ID id_cache;
-extern ID id_n_processors;
-
-extern VALUE cContext;
-extern VALUE eError;
-extern VALUE cModel;
-
-extern const rb_data_type_t ruby_whisper_params_type;
-extern VALUE ruby_whisper_transcribe(int argc, VALUE *argv, VALUE self);
-extern VALUE rb_whisper_model_s_new(VALUE context);
-extern VALUE rb_whisper_segment_s_new(VALUE context, int index);
-extern void prepare_transcription(ruby_whisper_params *rwp, VALUE *context);
-
-ID transcribe_option_names[1];
-
-static void
-ruby_whisper_free(ruby_whisper *rw)
-{
-  if (rw->context) {
-    whisper_free(rw->context);
-    rw->context = NULL;
-  }
-}
-
-void
-rb_whisper_mark(ruby_whisper *rw)
-{
-  // call rb_gc_mark on any ruby references in rw
-}
-
-void
-rb_whisper_free(void *p)
-{
-  ruby_whisper *rw = (ruby_whisper *)p;
-  ruby_whisper_free(rw);
-  free(rw);
-}
-
-static size_t
-ruby_whisper_memsize(const void *p)
-{
-  const ruby_whisper *rw = (const ruby_whisper *)p;
-  size_t size = sizeof(rw);
-  if (!rw) {
-    return 0;
-  }
-  if (rw->context) {
-    size += sizeof(rw->context);
-  }
-  return size;
-}
-
-const rb_data_type_t ruby_whisper_type = {
-  "ruby_whisper",
-  {0, rb_whisper_free, ruby_whisper_memsize,},
-  0, 0,
-  0
-};
-
-static VALUE
-ruby_whisper_allocate(VALUE klass)
-{
-  ruby_whisper *rw;
-  VALUE obj = TypedData_Make_Struct(klass, ruby_whisper, &ruby_whisper_type, rw);
-  rw->context = NULL;
-  return obj;
-}
-
-VALUE
-ruby_whisper_normalize_model_path(VALUE model_path)
-{
-  VALUE pre_converted_models = rb_funcall(cModel, id_pre_converted_models, 0);
-  VALUE pre_converted_model = rb_hash_aref(pre_converted_models, model_path);
-  if (!NIL_P(pre_converted_model)) {
-    model_path = pre_converted_model;
-#ifdef RUBY_WHISPER_USE_COREML
-    VALUE coreml_converted_models = rb_funcall(cModel, id_coreml_compiled_models, 0);
-    VALUE coreml_converted_model = rb_hash_aref(coreml_converted_models, pre_converted_model);
-    if (!NIL_P(coreml_converted_model)) {
-      rb_funcall(coreml_converted_model, id_cache, 0);
-    }
-#endif
-  }
-  else if (TYPE(model_path) == T_STRING) {
-    const char * model_path_str = StringValueCStr(model_path);
-    if (strncmp("http://", model_path_str, 7) == 0 || strncmp("https://", model_path_str, 8) == 0) {
-      VALUE uri_class = rb_const_get(cModel, id_URI);
-      model_path = rb_class_new_instance(1, &model_path, uri_class);
-    }
-  }
-  else if (rb_obj_is_kind_of(model_path, rb_path2class("URI::HTTP"))) {
-    VALUE uri_class = rb_const_get(cModel, id_URI);
-    model_path = rb_class_new_instance(1, &model_path, uri_class);
-  }
-  if (rb_respond_to(model_path, id_to_path)) {
-    model_path = rb_funcall(model_path, id_to_path, 0);
-  }
-
-  return model_path;
-}
-
-/*
- * call-seq:
- *   new("base.en") -> Whisper::Context
- *   new("path/to/model.bin") -> Whisper::Context
- *   new(Whisper::Model::URI.new("https://example.net/uri/of/model.bin")) -> Whisper::Context
- */
-static VALUE
-ruby_whisper_initialize(int argc, VALUE *argv, VALUE self)
-{
-  ruby_whisper *rw;
-  VALUE whisper_model_file_path;
-
-  // TODO: we can support init from buffer here too maybe another ruby object to expose
-  rb_scan_args(argc, argv, "01", &whisper_model_file_path);
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-
-  whisper_model_file_path = ruby_whisper_normalize_model_path(whisper_model_file_path);
-  if (!rb_respond_to(whisper_model_file_path, id_to_s)) {
-    rb_raise(rb_eRuntimeError, "Expected file path to model to initialize Whisper::Context");
-  }
-  rw->context = whisper_init_from_file_with_params(StringValueCStr(whisper_model_file_path), whisper_context_default_params());
-  if (rw->context == NULL) {
-    rb_raise(rb_eRuntimeError, "error: failed to initialize whisper context");
-  }
-  return self;
-}
-
-/*
- * call-seq:
- *   model_n_vocab -> Integer
- */
-VALUE ruby_whisper_model_n_vocab(VALUE self)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_vocab(rw->context));
-}
-
-/*
- * call-seq:
- *   model_n_audio_ctx -> Integer
- */
-VALUE ruby_whisper_model_n_audio_ctx(VALUE self)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_audio_ctx(rw->context));
-}
-
-/*
- * call-seq:
- *   model_n_audio_state -> Integer
- */
-VALUE ruby_whisper_model_n_audio_state(VALUE self)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_audio_state(rw->context));
-}
-
-/*
- * call-seq:
- *   model_n_audio_head -> Integer
- */
-VALUE ruby_whisper_model_n_audio_head(VALUE self)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_audio_head(rw->context));
-}
-
-/*
- * call-seq:
- *   model_n_audio_layer -> Integer
- */
-VALUE ruby_whisper_model_n_audio_layer(VALUE self)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_audio_layer(rw->context));
-}
-
-/*
- * call-seq:
- *   model_n_text_ctx -> Integer
- */
-VALUE ruby_whisper_model_n_text_ctx(VALUE self)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_text_ctx(rw->context));
-}
-
-/*
- * call-seq:
- *   model_n_text_state -> Integer
- */
-VALUE ruby_whisper_model_n_text_state(VALUE self)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_text_state(rw->context));
-}
-
-/*
- * call-seq:
- *   model_n_text_head -> Integer
- */
-VALUE ruby_whisper_model_n_text_head(VALUE self)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_text_head(rw->context));
-}
-
-/*
- * call-seq:
- *   model_n_text_layer -> Integer
- */
-VALUE ruby_whisper_model_n_text_layer(VALUE self)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_text_layer(rw->context));
-}
-
-/*
- * call-seq:
- *   model_n_mels -> Integer
- */
-VALUE ruby_whisper_model_n_mels(VALUE self)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_mels(rw->context));
-}
-
-/*
- * call-seq:
- *   model_ftype -> Integer
- */
-VALUE ruby_whisper_model_ftype(VALUE self)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_ftype(rw->context));
-}
-
-/*
- * call-seq:
- *   model_type -> String
- */
-VALUE ruby_whisper_model_type(VALUE self)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  return rb_str_new2(whisper_model_type_readable(rw->context));
-}
-
-/*
- * Run the entire model: PCM -> log mel spectrogram -> encoder -> decoder -> text
- * Not thread safe for same context
- * Uses the specified decoding strategy to obtain the text.
- *
- * call-seq:
- *   full(params, samples, n_samples) -> nil
- *   full(params, samples) -> nil
- *
- * The second argument +samples+ must be an array of samples, respond to :length, or be a MemoryView of an array of float. It must be 32 bit float PCM audio data.
- */
-VALUE ruby_whisper_full(int argc, VALUE *argv, VALUE self)
-{
-  if (argc < 2 || argc > 3) {
-    rb_raise(rb_eArgError, "wrong number of arguments (given %d, expected 2..3)", argc);
-  }
-
-  ruby_whisper *rw;
-  ruby_whisper_params *rwp;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  VALUE params = argv[0];
-  TypedData_Get_Struct(params, ruby_whisper_params, &ruby_whisper_params_type, rwp);
-  VALUE samples = argv[1];
-  int n_samples;
-  rb_memory_view_t view;
-  const bool memory_view_available_p = rb_memory_view_available_p(samples);
-  if (argc == 3) {
-    n_samples = NUM2INT(argv[2]);
-    if (TYPE(samples) == T_ARRAY) {
-      if (RARRAY_LEN(samples) < n_samples) {
-        rb_raise(rb_eArgError, "samples length %ld is less than n_samples %d", RARRAY_LEN(samples), n_samples);
-      }
-    }
-    // Should check when samples.respond_to?(:length)?
-  } else {
-    if (TYPE(samples) == T_ARRAY) {
-      if (RARRAY_LEN(samples) > INT_MAX) {
-        rb_raise(rb_eArgError, "samples are too long");
-      }
-      n_samples = (int)RARRAY_LEN(samples);
-    } else if (memory_view_available_p) {
-      if (!rb_memory_view_get(samples, &view, RUBY_MEMORY_VIEW_SIMPLE)) {
-        view.obj = Qnil;
-        rb_raise(rb_eArgError, "unable to get a memory view");
-      }
-      ssize_t n_samples_size = view.byte_size / view.item_size;
-      if (n_samples_size > INT_MAX) {
-        rb_raise(rb_eArgError, "samples are too long");
-      }
-      n_samples = (int)n_samples_size;
-    } else if (rb_respond_to(samples, id_length)) {
-      n_samples = NUM2INT(rb_funcall(samples, id_length, 0));
-    } else {
-      rb_raise(rb_eArgError, "samples must respond to :length or be a MemoryView of an array of flaot when n_samples is not given");
-    }
-  }
-  float * c_samples = (float *)malloc(n_samples * sizeof(float));
-  if (memory_view_available_p)  {
-    c_samples = (float *)view.data;
-  } else {
-    if (TYPE(samples) == T_ARRAY) {
-      for (int i = 0; i < n_samples; i++) {
-        c_samples[i] = RFLOAT_VALUE(rb_ary_entry(samples, i));
-      }
-    } else {
-      // TODO: use rb_block_call
-      VALUE iter = rb_funcall(samples, id_to_enum, 1, rb_str_new2("each"));
-      for (int i = 0; i < n_samples; i++) {
-        // TODO: check if iter is exhausted and raise ArgumentError appropriately
-        VALUE sample = rb_funcall(iter, id_next, 0);
-        c_samples[i] = RFLOAT_VALUE(sample);
-      }
-    }
-  }
-  prepare_transcription(rwp, &self);
-  const int result = whisper_full(rw->context, rwp->params, c_samples, n_samples);
-  if (0 == result) {
-    return self;
-  } else {
-    rb_exc_raise(rb_funcall(eError, id_new, 1, result));
-  }
-}
-
-/*
- * Split the input audio in chunks and process each chunk separately using whisper_full_with_state()
- * Result is stored in the default state of the context
- * Not thread safe if executed in parallel on the same context.
- * It seems this approach can offer some speedup in some cases.
- * However, the transcription accuracy can be worse at the beginning and end of each chunk.
- *
- * call-seq:
- *   full_parallel(params, samples) -> nil
- *   full_parallel(params, samples, n_samples) -> nil
- *   full_parallel(params, samples, n_samples, n_processors) -> nil
- *   full_parallel(params, samples, nil, n_processors) -> nil
- */
-static VALUE
-ruby_whisper_full_parallel(int argc, VALUE *argv,VALUE self)
-{
-  if (argc < 2 || argc > 4) {
-    rb_raise(rb_eArgError, "wrong number of arguments (given %d, expected 2..3)", argc);
-  }
-
-  ruby_whisper *rw;
-  ruby_whisper_params *rwp;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  VALUE params = argv[0];
-  TypedData_Get_Struct(params, ruby_whisper_params, &ruby_whisper_params_type, rwp);
-  VALUE samples = argv[1];
-  int n_samples;
-  int n_processors;
-  rb_memory_view_t view;
-  const bool memory_view_available_p = rb_memory_view_available_p(samples);
-  switch (argc) {
-  case 2:
-    n_processors = 1;
-    break;
-  case 3:
-    n_processors = 1;
-    break;
-  case 4:
-    n_processors = NUM2INT(argv[3]);
-    break;
-  }
-  if (argc >= 3 && !NIL_P(argv[2])) {
-    n_samples = NUM2INT(argv[2]);
-    if (TYPE(samples) == T_ARRAY) {
-      if (RARRAY_LEN(samples) < n_samples) {
-        rb_raise(rb_eArgError, "samples length %ld is less than n_samples %d", RARRAY_LEN(samples), n_samples);
-      }
-    }
-    // Should check when samples.respond_to?(:length)?
-  } else if (memory_view_available_p) {
-    if (!rb_memory_view_get(samples, &view, RUBY_MEMORY_VIEW_SIMPLE)) {
-      view.obj = Qnil;
-      rb_raise(rb_eArgError, "unable to get a memory view");
-    }
-    ssize_t n_samples_size = view.byte_size / view.item_size;
-    if (n_samples_size > INT_MAX) {
-      rb_raise(rb_eArgError, "samples are too long");
-    }
-    n_samples = (int)n_samples_size;
-  } else {
-    if (TYPE(samples) == T_ARRAY) {
-      if (RARRAY_LEN(samples) > INT_MAX) {
-        rb_raise(rb_eArgError, "samples are too long");
-      }
-      n_samples = (int)RARRAY_LEN(samples);
-    } else if (rb_respond_to(samples, id_length)) {
-      n_samples = NUM2INT(rb_funcall(samples, id_length, 0));
-    } else {
-      rb_raise(rb_eArgError, "samples must respond to :length or be a MemoryView of an array of flaot when n_samples is not given");
-    }
-  }
-  float * c_samples = (float *)malloc(n_samples * sizeof(float));
-  if (memory_view_available_p) {
-    c_samples = (float *)view.data;
-  } else {
-    if (TYPE(samples) == T_ARRAY) {
-      for (int i = 0; i < n_samples; i++) {
-        c_samples[i] = RFLOAT_VALUE(rb_ary_entry(samples, i));
-      }
-    } else {
-      // FIXME: use rb_block_call
-      VALUE iter = rb_funcall(samples, id_to_enum, 1, rb_str_new2("each"));
-      for (int i = 0; i < n_samples; i++) {
-        // TODO: check if iter is exhausted and raise ArgumentError
-        VALUE sample = rb_funcall(iter, id_next, 0);
-        c_samples[i] = RFLOAT_VALUE(sample);
-      }
-    }
-  }
-  prepare_transcription(rwp, &self);
-  const int result = whisper_full_parallel(rw->context, rwp->params, c_samples, n_samples, n_processors);
-  if (0 == result) {
-    return self;
-  } else {
-    rb_exc_raise(rb_funcall(eError, id_new, 1, result));
-  }
-}
-
-/*
- * Number of segments.
- *
- * call-seq:
- *   full_n_segments -> Integer
- */
-static VALUE
-ruby_whisper_full_n_segments(VALUE self)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_full_n_segments(rw->context));
-}
-
-/*
- * Language ID, which can be converted to string by Whisper.lang_str and Whisper.lang_str_full.
- *
- * call-seq:
- *   full_lang_id -> Integer
- */
-static VALUE
-ruby_whisper_full_lang_id(VALUE self)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_full_lang_id(rw->context));
-}
-
-static int ruby_whisper_full_check_segment_index(const ruby_whisper * rw, const VALUE i_segment)
-{
-  const int c_i_segment = NUM2INT(i_segment);
-  if (c_i_segment < 0 || c_i_segment >= whisper_full_n_segments(rw->context)) {
-    rb_raise(rb_eIndexError, "segment index %d out of range", c_i_segment);
-  }
-  return c_i_segment;
-}
-
-/*
- * Start time of a segment indexed by +segment_index+ in centiseconds (10 times milliseconds).
- *
- *   full_get_segment_t0(3) # => 1668 (16680 ms)
- *
- * call-seq:
- *   full_get_segment_t0(segment_index) -> Integer
- */
-static VALUE
-ruby_whisper_full_get_segment_t0(VALUE self, VALUE i_segment)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  const int c_i_segment = ruby_whisper_full_check_segment_index(rw, i_segment);
-  const int64_t t0 = whisper_full_get_segment_t0(rw->context, c_i_segment);
-  return LONG2NUM(t0);
-}
-
-/*
- * End time of a segment indexed by +segment_index+ in centiseconds (10 times milliseconds).
- *
- *   full_get_segment_t1(3) # => 1668 (16680 ms)
- *
- * call-seq:
- *   full_get_segment_t1(segment_index) -> Integer
- */
-static VALUE
-ruby_whisper_full_get_segment_t1(VALUE self, VALUE i_segment)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  const int c_i_segment = ruby_whisper_full_check_segment_index(rw, i_segment);
-  const int64_t t1 = whisper_full_get_segment_t1(rw->context, c_i_segment);
-  return LONG2NUM(t1);
-}
-
-/*
- * Whether the next segment indexed by +segment_index+ is predicated as a speaker turn.
- *
- *   full_get_segment_speacker_turn_next(3) # => true
- *
- * call-seq:
- *   full_get_segment_speacker_turn_next(segment_index) -> bool
- */
-static VALUE
-ruby_whisper_full_get_segment_speaker_turn_next(VALUE self, VALUE i_segment)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  const int c_i_segment = ruby_whisper_full_check_segment_index(rw, i_segment);
-  const bool speaker_turn_next = whisper_full_get_segment_speaker_turn_next(rw->context, c_i_segment);
-  return speaker_turn_next ? Qtrue : Qfalse;
-}
-
-/*
- * Text of a segment indexed by +segment_index+.
- *
- *   full_get_segment_text(3) # => "ask not what your country can do for you, ..."
- *
- * call-seq:
- *   full_get_segment_text(segment_index) -> String
- */
-static VALUE
-ruby_whisper_full_get_segment_text(VALUE self, VALUE i_segment)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  const int c_i_segment = ruby_whisper_full_check_segment_index(rw, i_segment);
-  const char * text = whisper_full_get_segment_text(rw->context, c_i_segment);
-  return rb_str_new2(text);
-}
-
-/*
- * call-seq:
- *   full_get_segment_no_speech_prob(segment_index) -> Float
- */
-static VALUE
-ruby_whisper_full_get_segment_no_speech_prob(VALUE self, VALUE i_segment)
-{
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  const int c_i_segment = ruby_whisper_full_check_segment_index(rw, i_segment);
-  const float no_speech_prob = whisper_full_get_segment_no_speech_prob(rw->context, c_i_segment);
-  return DBL2NUM(no_speech_prob);
-}
-
-// High level API
-
-static VALUE
-ruby_whisper_full_get_segment(VALUE self, VALUE i_segment)
-{
-  return rb_whisper_segment_s_new(self, NUM2INT(i_segment));
-}
-
-/*
- * Yields each Whisper::Segment:
- *
- *   whisper.transcribe("path/to/audio.wav", params)
- *   whisper.each_segment do |segment|
- *     puts segment.text
- *   end
- *
- * Returns an Enumerator if no block given:
- *
- *   whisper.transcribe("path/to/audio.wav", params)
- *   enum = whisper.each_segment
- *   enum.to_a # => [#<Whisper::Segment>, ...]
- *
- * call-seq:
- *   each_segment {|segment| ... }
- *   each_segment -> Enumerator
- */
-static VALUE
-ruby_whisper_each_segment(VALUE self)
-{
-  if (!rb_block_given_p()) {
-    const VALUE method_name = rb_funcall(self, id___method__, 0);
-    return rb_funcall(self, id_to_enum, 1, method_name);
-  }
-
-  ruby_whisper *rw;
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-
-  const int n_segments = whisper_full_n_segments(rw->context);
-  for (int i = 0; i < n_segments; ++i) {
-    rb_yield(rb_whisper_segment_s_new(self, i));
-  }
-
-  return self;
-}
-
-/*
- * call-seq:
- *   model -> Whisper::Model
- */
-static VALUE
-ruby_whisper_get_model(VALUE self)
-{
-  return rb_whisper_model_s_new(self);
-}
-
-void
-init_ruby_whisper_context(VALUE *mWhisper)
-{
-  cContext = rb_define_class_under(*mWhisper, "Context", rb_cObject);
-
-  transcribe_option_names[0] = id_n_processors;
-
-  rb_define_alloc_func(cContext, ruby_whisper_allocate);
-  rb_define_method(cContext, "initialize", ruby_whisper_initialize, -1);
-
-  rb_define_method(cContext, "transcribe", ruby_whisper_transcribe, -1);
-  rb_define_method(cContext, "model_n_vocab", ruby_whisper_model_n_vocab, 0);
-  rb_define_method(cContext, "model_n_audio_ctx", ruby_whisper_model_n_audio_ctx, 0);
-  rb_define_method(cContext, "model_n_audio_state", ruby_whisper_model_n_audio_state, 0);
-  rb_define_method(cContext, "model_n_audio_head", ruby_whisper_model_n_audio_head, 0);
-  rb_define_method(cContext, "model_n_audio_layer", ruby_whisper_model_n_audio_layer, 0);
-  rb_define_method(cContext, "model_n_text_ctx", ruby_whisper_model_n_text_ctx, 0);
-  rb_define_method(cContext, "model_n_text_state", ruby_whisper_model_n_text_state, 0);
-  rb_define_method(cContext, "model_n_text_head", ruby_whisper_model_n_text_head, 0);
-  rb_define_method(cContext, "model_n_text_layer", ruby_whisper_model_n_text_layer, 0);
-  rb_define_method(cContext, "model_n_mels", ruby_whisper_model_n_mels, 0);
-  rb_define_method(cContext, "model_ftype", ruby_whisper_model_ftype, 0);
-  rb_define_method(cContext, "model_type", ruby_whisper_model_type, 0);
-  rb_define_method(cContext, "full_n_segments", ruby_whisper_full_n_segments, 0);
-  rb_define_method(cContext, "full_lang_id", ruby_whisper_full_lang_id, 0);
-  rb_define_method(cContext, "full_get_segment_t0", ruby_whisper_full_get_segment_t0, 1);
-  rb_define_method(cContext, "full_get_segment_t1", ruby_whisper_full_get_segment_t1, 1);
-  rb_define_method(cContext, "full_get_segment_speaker_turn_next", ruby_whisper_full_get_segment_speaker_turn_next, 1);
-  rb_define_method(cContext, "full_get_segment_text", ruby_whisper_full_get_segment_text, 1);
-  rb_define_method(cContext, "full_get_segment_no_speech_prob", ruby_whisper_full_get_segment_no_speech_prob, 1);
-  rb_define_method(cContext, "full", ruby_whisper_full, -1);
-  rb_define_method(cContext, "full_parallel", ruby_whisper_full_parallel, -1);
-
-  // High level
-  rb_define_method(cContext, "full_get_segment", ruby_whisper_full_get_segment, 1);
-  rb_define_method(cContext, "each_segment", ruby_whisper_each_segment, 0);
-
-  rb_define_method(cContext, "model", ruby_whisper_get_model, 0);
-}
--- a/bindings/ruby/ext/ruby_whisper_error.c
+++ b/bindings/ruby/ext/ruby_whisper_error.c
@ -1,52 +0,0 @@
-#include <ruby.h>
-
-extern VALUE eError;
-
-VALUE ruby_whisper_error_initialize(VALUE self, VALUE code)
-{
-  const int c_code = NUM2INT(code);
-  const char *raw_message;
-  switch (c_code) {
-  case -2:
-    raw_message = "failed to compute log mel spectrogram";
-    break;
-  case -3:
-    raw_message = "failed to auto-detect language";
-    break;
-  case -4:
-    raw_message = "too many decoders requested";
-    break;
-  case -5:
-    raw_message = "audio_ctx is larger than the maximum allowed";
-    break;
-  case -6:
-    raw_message = "failed to encode";
-    break;
-  case -7:
-    raw_message = "whisper_kv_cache_init() failed for self-attention cache";
-    break;
-  case -8:
-    raw_message = "failed to decode";
-    break;
-  case -9:
-    raw_message = "failed to decode";
-    break;
-  default:
-    raw_message = "unknown error";
-    break;
-  }
-  const VALUE message = rb_str_new2(raw_message);
-  rb_call_super(1, &message);
-  rb_iv_set(self, "@code", code);
-
-  return self;
-}
-
-void
-init_ruby_whisper_error(VALUE *mWhisper)
-{
-  eError = rb_define_class_under(*mWhisper, "Error", rb_eStandardError);
-
-  rb_define_attr(eError, "code", true, false);
-  rb_define_method(eError, "initialize", ruby_whisper_error_initialize, 1);
-}
--- a/bindings/ruby/ext/ruby_whisper_model.c
+++ b/bindings/ruby/ext/ruby_whisper_model.c
@ -1,232 +0,0 @@
-#include <ruby.h>
-#include "ruby_whisper.h"
-
-extern const rb_data_type_t ruby_whisper_type;
-
-extern VALUE cModel;
-
-static void rb_whisper_model_mark(void *p) {
-  ruby_whisper_model *rwm = (ruby_whisper_model *)p;
-  if (rwm->context) {
-    rb_gc_mark(rwm->context);
-  }
-}
-
-static size_t
-ruby_whisper_model_memsize(const void *p)
-{
-  const ruby_whisper_model *rwm = (const ruby_whisper_model *)p;
-  size_t size = sizeof(rwm);
-  if (!rwm) {
-    return 0;
-  }
-  return size;
-}
-
-static const rb_data_type_t rb_whisper_model_type = {
-  "ruby_whisper_model",
-  {rb_whisper_model_mark, RUBY_DEFAULT_FREE, ruby_whisper_model_memsize,},
-  0, 0,
-  0
-};
-
-static VALUE ruby_whisper_model_allocate(VALUE klass) {
-  ruby_whisper_model *rwm;
-  return TypedData_Make_Struct(klass, ruby_whisper_model, &rb_whisper_model_type, rwm);
-}
-
-VALUE rb_whisper_model_s_new(VALUE context) {
-  ruby_whisper_model *rwm;
-  const VALUE model = ruby_whisper_model_allocate(cModel);
-  TypedData_Get_Struct(model, ruby_whisper_model, &rb_whisper_model_type, rwm);
-  rwm->context = context;
-  return model;
-};
-
-/*
- * call-seq:
- *   n_vocab -> Integer
- */
-static VALUE
-ruby_whisper_model_n_vocab(VALUE self)
-{
-  ruby_whisper_model *rwm;
-  TypedData_Get_Struct(self, ruby_whisper_model, &rb_whisper_model_type, rwm);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rwm->context, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_vocab(rw->context));
-}
-
-/*
- * call-seq:
- *   n_audio_ctx -> Integer
- */
-static VALUE
-ruby_whisper_model_n_audio_ctx(VALUE self)
-{
-  ruby_whisper_model *rwm;
-  TypedData_Get_Struct(self, ruby_whisper_model, &rb_whisper_model_type, rwm);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rwm->context, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_audio_ctx(rw->context));
-}
-
-/*
- * call-seq:
- *   n_audio_state -> Integer
- */
-static VALUE
-ruby_whisper_model_n_audio_state(VALUE self)
-{
-  ruby_whisper_model *rwm;
-  TypedData_Get_Struct(self, ruby_whisper_model, &rb_whisper_model_type, rwm);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rwm->context, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_audio_state(rw->context));
-}
-
-/*
- * call-seq:
- *   n_audio_head -> Integer
- */
-static VALUE
-ruby_whisper_model_n_audio_head(VALUE self)
-{
-  ruby_whisper_model *rwm;
-  TypedData_Get_Struct(self, ruby_whisper_model, &rb_whisper_model_type, rwm);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rwm->context, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_audio_head(rw->context));
-}
-
-/*
- * call-seq:
- *   n_audio_layer -> Integer
- */
-static VALUE
-ruby_whisper_model_n_audio_layer(VALUE self)
-{
-  ruby_whisper_model *rwm;
-  TypedData_Get_Struct(self, ruby_whisper_model, &rb_whisper_model_type, rwm);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rwm->context, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_audio_layer(rw->context));
-}
-
-/*
- * call-seq:
- *   n_text_ctx -> Integer
- */
-static VALUE
-ruby_whisper_model_n_text_ctx(VALUE self)
-{
-  ruby_whisper_model *rwm;
-  TypedData_Get_Struct(self, ruby_whisper_model, &rb_whisper_model_type, rwm);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rwm->context, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_text_ctx(rw->context));
-}
-
-/*
- * call-seq:
- *   n_text_state -> Integer
- */
-static VALUE
-ruby_whisper_model_n_text_state(VALUE self)
-{
-  ruby_whisper_model *rwm;
-  TypedData_Get_Struct(self, ruby_whisper_model, &rb_whisper_model_type, rwm);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rwm->context, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_text_state(rw->context));
-}
-
-/*
- * call-seq:
- *   n_text_head -> Integer
- */
-static VALUE
-ruby_whisper_model_n_text_head(VALUE self)
-{
-  ruby_whisper_model *rwm;
-  TypedData_Get_Struct(self, ruby_whisper_model, &rb_whisper_model_type, rwm);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rwm->context, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_text_head(rw->context));
-}
-
-/*
- * call-seq:
- *   n_text_layer -> Integer
- */
-static VALUE
-ruby_whisper_model_n_text_layer(VALUE self)
-{
-  ruby_whisper_model *rwm;
-  TypedData_Get_Struct(self, ruby_whisper_model, &rb_whisper_model_type, rwm);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rwm->context, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_text_layer(rw->context));
-}
-
-/*
- * call-seq:
- *   n_mels -> Integer
- */
-static VALUE
-ruby_whisper_model_n_mels(VALUE self)
-{
-  ruby_whisper_model *rwm;
-  TypedData_Get_Struct(self, ruby_whisper_model, &rb_whisper_model_type, rwm);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rwm->context, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_n_mels(rw->context));
-}
-
-/*
- * call-seq:
- *   ftype -> Integer
- */
-static VALUE
-ruby_whisper_model_ftype(VALUE self)
-{
-  ruby_whisper_model *rwm;
-  TypedData_Get_Struct(self, ruby_whisper_model, &rb_whisper_model_type, rwm);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rwm->context, ruby_whisper, &ruby_whisper_type, rw);
-  return INT2NUM(whisper_model_ftype(rw->context));
-}
-
-/*
- * call-seq:
- *   type -> String
- */
-static VALUE
-ruby_whisper_model_type(VALUE self)
-{
-  ruby_whisper_model *rwm;
-  TypedData_Get_Struct(self, ruby_whisper_model, &rb_whisper_model_type, rwm);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rwm->context, ruby_whisper, &ruby_whisper_type, rw);
-  return rb_str_new2(whisper_model_type_readable(rw->context));
-}
-
-void
-init_ruby_whisper_model(VALUE *mWhisper)
-{
-  cModel = rb_define_class_under(*mWhisper, "Model", rb_cObject);
-
-  rb_define_alloc_func(cModel, ruby_whisper_model_allocate);
-  rb_define_method(cModel, "n_vocab", ruby_whisper_model_n_vocab, 0);
-  rb_define_method(cModel, "n_audio_ctx", ruby_whisper_model_n_audio_ctx, 0);
-  rb_define_method(cModel, "n_audio_state", ruby_whisper_model_n_audio_state, 0);
-  rb_define_method(cModel, "n_audio_head", ruby_whisper_model_n_audio_head, 0);
-  rb_define_method(cModel, "n_audio_layer", ruby_whisper_model_n_audio_layer, 0);
-  rb_define_method(cModel, "n_text_ctx", ruby_whisper_model_n_text_ctx, 0);
-  rb_define_method(cModel, "n_text_state", ruby_whisper_model_n_text_state, 0);
-  rb_define_method(cModel, "n_text_head", ruby_whisper_model_n_text_head, 0);
-  rb_define_method(cModel, "n_text_layer", ruby_whisper_model_n_text_layer, 0);
-  rb_define_method(cModel, "n_mels", ruby_whisper_model_n_mels, 0);
-  rb_define_method(cModel, "ftype", ruby_whisper_model_ftype, 0);
-  rb_define_method(cModel, "type", ruby_whisper_model_type, 0);
-}
--- a/bindings/ruby/ext/ruby_whisper_params.c
+++ b/bindings/ruby/ext/ruby_whisper_params.c
--- a/bindings/ruby/ext/ruby_whisper_segment.c
+++ b/bindings/ruby/ext/ruby_whisper_segment.c
@ -1,220 +0,0 @@
-#include <ruby.h>
-#include "ruby_whisper.h"
-
-#define N_KEY_NAMES 5
-
-static VALUE sym_start_time;
-static VALUE sym_end_time;
-static VALUE sym_text;
-static VALUE sym_no_speech_prob;
-static VALUE sym_speaker_turn_next;
-static VALUE key_names;
-
-extern const rb_data_type_t ruby_whisper_type;
-
-extern VALUE cSegment;
-
-static void
-rb_whisper_segment_mark(void *p)
-{
-  ruby_whisper_segment *rws = (ruby_whisper_segment *)p;
-  rb_gc_mark(rws->context);
-}
-
-static size_t
-ruby_whisper_segment_memsize(const void *p)
-{
-  const ruby_whisper_segment *rws = (const ruby_whisper_segment *)p;
-  size_t size = sizeof(rws);
-  if (!rws) {
-    return 0;
-  }
-  return size;
-}
-
-static const rb_data_type_t ruby_whisper_segment_type = {
-  "ruby_whisper_segment",
-  {rb_whisper_segment_mark, RUBY_DEFAULT_FREE, ruby_whisper_segment_memsize,},
-  0, 0,
-  0
-};
-
-VALUE
-ruby_whisper_segment_allocate(VALUE klass)
-{
-  ruby_whisper_segment *rws;
-  return TypedData_Make_Struct(klass, ruby_whisper_segment, &ruby_whisper_segment_type, rws);
-}
-
-VALUE
-rb_whisper_segment_s_new(VALUE context, int index)
-{
-  ruby_whisper_segment *rws;
-  const VALUE segment = ruby_whisper_segment_allocate(cSegment);
-  TypedData_Get_Struct(segment, ruby_whisper_segment, &ruby_whisper_segment_type, rws);
-  rws->context = context;
-  rws->index = index;
-  return segment;
-};
-
-/*
- * Start time in milliseconds.
- *
- * call-seq:
- *   start_time -> Integer
- */
-static VALUE
-ruby_whisper_segment_get_start_time(VALUE self)
-{
-  ruby_whisper_segment *rws;
-  TypedData_Get_Struct(self, ruby_whisper_segment, &ruby_whisper_segment_type, rws);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rws->context, ruby_whisper, &ruby_whisper_type, rw);
-  const int64_t t0 = whisper_full_get_segment_t0(rw->context, rws->index);
-  // able to multiply 10 without overflow because to_timestamp() in whisper.cpp does it
-  return LONG2NUM(t0 * 10);
-}
-
-/*
- * End time in milliseconds.
- *
- * call-seq:
- *   end_time -> Integer
- */
-static VALUE
-ruby_whisper_segment_get_end_time(VALUE self)
-{
-  ruby_whisper_segment *rws;
-  TypedData_Get_Struct(self, ruby_whisper_segment, &ruby_whisper_segment_type, rws);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rws->context, ruby_whisper, &ruby_whisper_type, rw);
-  const int64_t t1 = whisper_full_get_segment_t1(rw->context, rws->index);
-  // able to multiply 10 without overflow because to_timestamp() in whisper.cpp does it
-  return LONG2NUM(t1 * 10);
-}
-
-/*
- * Whether the next segment is predicted as a speaker turn.
- *
- * call-seq:
- *   speaker_turn_next? -> bool
- */
-static VALUE
-ruby_whisper_segment_get_speaker_turn_next(VALUE self)
-{
-  ruby_whisper_segment *rws;
-  TypedData_Get_Struct(self, ruby_whisper_segment, &ruby_whisper_segment_type, rws);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rws->context, ruby_whisper, &ruby_whisper_type, rw);
-  return whisper_full_get_segment_speaker_turn_next(rw->context, rws->index) ? Qtrue : Qfalse;
-}
-
-/*
- * call-seq:
- *   text -> String
- */
-static VALUE
-ruby_whisper_segment_get_text(VALUE self)
-{
-  ruby_whisper_segment *rws;
-  TypedData_Get_Struct(self, ruby_whisper_segment, &ruby_whisper_segment_type, rws);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rws->context, ruby_whisper, &ruby_whisper_type, rw);
-  const char * text = whisper_full_get_segment_text(rw->context, rws->index);
-  return rb_str_new2(text);
-}
-
-/*
- * call-seq:
- *   no_speech_prob -> Float
- */
-static VALUE
-ruby_whisper_segment_get_no_speech_prob(VALUE self)
-{
-  ruby_whisper_segment *rws;
-  TypedData_Get_Struct(self, ruby_whisper_segment, &ruby_whisper_segment_type, rws);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rws->context, ruby_whisper, &ruby_whisper_type, rw);
-  return DBL2NUM(whisper_full_get_segment_no_speech_prob(rw->context, rws->index));
-}
-
-/*
- * call-seq:
- *   deconstruct_keys(keys) -> hash
- *
- *  Possible keys: :start_time, :end_time, :text, :no_speech_prob, :speaker_turn_next
- *
- *   whisper.each_segment do |segment|
- *     segment => {start_time:, end_time:, text:, no_speech_prob:, speaker_turn_next:}
- *
- *     puts "[#{start_time} --> #{end_time}] #{text} (no speech prob: #{no_speech_prob}#{speaker_turn_next ? ', speaker turns next' : ''})"
- *   end
- */
-static VALUE
-ruby_whisper_segment_deconstruct_keys(VALUE self, VALUE keys)
-{
-  ruby_whisper_segment *rws;
-  TypedData_Get_Struct(self, ruby_whisper_segment, &ruby_whisper_segment_type, rws);
-  ruby_whisper *rw;
-  TypedData_Get_Struct(rws->context, ruby_whisper, &ruby_whisper_type, rw);
-
-  VALUE hash = rb_hash_new();
-  long n_keys;
-  if (NIL_P(keys)) {
-    keys = key_names;
-    n_keys = N_KEY_NAMES;
-  } else {
-    n_keys = RARRAY_LEN(keys);
-    if (n_keys > N_KEY_NAMES) {
-      return hash;
-    }
-  }
-  for (int i = 0; i < n_keys; i++) {
-    VALUE key = rb_ary_entry(keys, i);
-    if (key == sym_start_time) {
-      rb_hash_aset(hash, key, ruby_whisper_segment_get_start_time(self));
-    }
-    if (key == sym_end_time) {
-      rb_hash_aset(hash, key, ruby_whisper_segment_get_end_time(self));
-    }
-    if (key == sym_text) {
-      rb_hash_aset(hash, key, ruby_whisper_segment_get_text(self));
-    }
-    if (key == sym_no_speech_prob) {
-      rb_hash_aset(hash, key, ruby_whisper_segment_get_no_speech_prob(self));
-    }
-    if (key == sym_speaker_turn_next) {
-      rb_hash_aset(hash, key, ruby_whisper_segment_get_speaker_turn_next(self));
-    }
-  }
-
-  return hash;
-}
-
-void
-init_ruby_whisper_segment(VALUE *mWhisper, VALUE *cContext)
-{
-  cSegment  = rb_define_class_under(*mWhisper, "Segment", rb_cObject);
-
-  sym_start_time = ID2SYM(rb_intern("start_time"));
-  sym_end_time = ID2SYM(rb_intern("end_time"));
-  sym_text = ID2SYM(rb_intern("text"));
-  sym_no_speech_prob = ID2SYM(rb_intern("no_speech_prob"));
-  sym_speaker_turn_next = ID2SYM(rb_intern("speaker_turn_next"));
-  key_names = rb_ary_new3(
-    N_KEY_NAMES,
-    sym_start_time,
-    sym_end_time,
-    sym_text,
-    sym_no_speech_prob,
-    sym_speaker_turn_next
-  );
-
-  rb_define_alloc_func(cSegment, ruby_whisper_segment_allocate);
-  rb_define_method(cSegment, "start_time", ruby_whisper_segment_get_start_time, 0);
-  rb_define_method(cSegment, "end_time", ruby_whisper_segment_get_end_time, 0);
-  rb_define_method(cSegment, "speaker_turn_next?", ruby_whisper_segment_get_speaker_turn_next, 0);
-  rb_define_method(cSegment, "text", ruby_whisper_segment_get_text, 0);
-  rb_define_method(cSegment, "no_speech_prob", ruby_whisper_segment_get_no_speech_prob, 0);
-  rb_define_method(cSegment, "deconstruct_keys", ruby_whisper_segment_deconstruct_keys, 1);
-}
--- a/bindings/ruby/ext/ruby_whisper_transcribe.cpp
+++ b/bindings/ruby/ext/ruby_whisper_transcribe.cpp
@ -1,93 +0,0 @@
-#include <ruby.h>
-#include "ruby_whisper.h"
-#include "common-whisper.h"
-#include <string>
-#include <vector>
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-extern const rb_data_type_t ruby_whisper_type;
-extern const rb_data_type_t ruby_whisper_params_type;
-
-extern ID id_to_s;
-extern ID id_call;
-extern ID transcribe_option_names[1];
-
-extern void
-prepare_transcription(ruby_whisper_params * rwp, VALUE * self);
-
-/*
- * transcribe a single file
- * can emit to a block results
- *
- *   params = Whisper::Params.new
- *   params.duration = 60_000
- *   whisper.transcribe "path/to/audio.wav", params do |text|
- *     puts text
- *   end
- *
- * call-seq:
- *   transcribe(path_to_audio, params) {|text| ...}
- **/
-VALUE
-ruby_whisper_transcribe(int argc, VALUE *argv, VALUE self) {
-  ruby_whisper *rw;
-  ruby_whisper_params *rwp;
-  VALUE wave_file_path, blk, params, kws;
-  VALUE opts[1];
-
-  rb_scan_args_kw(RB_SCAN_ARGS_LAST_HASH_KEYWORDS, argc, argv, "2:&", &wave_file_path, &params, &kws, &blk);
-  rb_get_kwargs(kws, transcribe_option_names, 0, 1, opts);
-
-  int n_processors = opts[0] == Qundef ? 1 : NUM2INT(opts[0]);
-
-  TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
-  TypedData_Get_Struct(params, ruby_whisper_params, &ruby_whisper_params_type, rwp);
-
-  if (!rb_respond_to(wave_file_path, id_to_s)) {
-    rb_raise(rb_eRuntimeError, "Expected file path to wave file");
-  }
-
-  std::string fname_inp = StringValueCStr(wave_file_path);
-
-  std::vector<float> pcmf32; // mono-channel F32 PCM
-  std::vector<std::vector<float>> pcmf32s; // stereo-channel F32 PCM
-
-  if (!read_audio_data(fname_inp, pcmf32, pcmf32s, rwp->diarize)) {
-    fprintf(stderr, "error: failed to open '%s' as WAV file\n", fname_inp.c_str());
-    return self;
-  }
-  // Commented out because it is work in progress
-  // {
-  //   static bool is_aborted = false; // NOTE: this should be atomic to avoid data race
-
-  //   rwp->params.encoder_begin_callback = [](struct whisper_context * /*ctx*/, struct whisper_state * /*state*/, void * user_data) {
-  //     bool is_aborted = *(bool*)user_data;
-  //     return !is_aborted;
-  //   };
-  //   rwp->params.encoder_begin_callback_user_data = &is_aborted;
-  // }
-
-  prepare_transcription(rwp, &self);
-
-  if (whisper_full_parallel(rw->context, rwp->params, pcmf32.data(), pcmf32.size(), n_processors) != 0) {
-    fprintf(stderr, "failed to process audio\n");
-    return self;
-  }
-  if (NIL_P(blk)) {
-    return self;
-  }
-  const int n_segments = whisper_full_n_segments(rw->context);
-  VALUE output = rb_str_new2("");
-  for (int i = 0; i < n_segments; ++i) {
-    const char * text = whisper_full_get_segment_text(rw->context, i);
-    output = rb_str_concat(output, rb_str_new2(text));
-  }
-  rb_funcall(blk, id_call, 1, output);
-  return self;
-}
-#ifdef __cplusplus
-}
-#endif
--- a/bindings/ruby/ext/ruby_whisper_vad_params.c
+++ b/bindings/ruby/ext/ruby_whisper_vad_params.c
@ -1,288 +0,0 @@
-#include <ruby.h>
-#include "ruby_whisper.h"
-
-#define DEFINE_PARAM(param_name, nth) \
-  id_ ## param_name = rb_intern(#param_name); \
-  param_names[nth] = id_ ## param_name; \
-  rb_define_method(cVADParams, #param_name, ruby_whisper_vad_params_get_ ## param_name, 0); \
-  rb_define_method(cVADParams, #param_name "=", ruby_whisper_vad_params_set_ ## param_name, 1);
-
-#define NUM_PARAMS 6
-
-extern VALUE cVADParams;
-
-static size_t
-ruby_whisper_vad_params_memsize(const void *p)
-{
-  const struct ruby_whisper_vad_params *params = p;
-  size_t size = sizeof(params);
-  if (!params) {
-    return 0;
-  }
-  return size;
-}
-
-static ID param_names[NUM_PARAMS];
-static ID id_threshold;
-static ID id_min_speech_duration_ms;
-static ID id_min_silence_duration_ms;
-static ID id_max_speech_duration_s;
-static ID id_speech_pad_ms;
-static ID id_samples_overlap;
-
-const rb_data_type_t ruby_whisper_vad_params_type = {
-  "ruby_whisper_vad_params",
-  {0, 0, ruby_whisper_vad_params_memsize,},
-  0, 0,
-  0
-};
-
-static VALUE
-ruby_whisper_vad_params_s_allocate(VALUE klass)
-{
-  ruby_whisper_vad_params *rwvp;
-  VALUE obj = TypedData_Make_Struct(klass, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-  rwvp->params = whisper_vad_default_params();
-  return obj;
-}
-
-/*
- * Probability threshold to consider as speech.
- *
- * call-seq:
- *   threshold = th -> th
- */
-static VALUE
-ruby_whisper_vad_params_set_threshold(VALUE self, VALUE value)
-{
-  ruby_whisper_vad_params *rwvp;
-  TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-  rwvp->params.threshold = RFLOAT_VALUE(value);
-  return value;
-}
-
-static VALUE
-ruby_whisper_vad_params_get_threshold(VALUE self)
-{
-  ruby_whisper_vad_params *rwvp;
-  TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-  return DBL2NUM(rwvp->params.threshold);
-}
-
-/*
- * Min duration for a valid speech segment.
- *
- * call-seq:
- *   min_speech_duration_ms = duration_ms -> duration_ms
- */
-static VALUE
-ruby_whisper_vad_params_set_min_speech_duration_ms(VALUE self, VALUE value)
-{
-  ruby_whisper_vad_params *rwvp;
-  TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-  rwvp->params.min_speech_duration_ms = NUM2INT(value);
-  return value;
-}
-
-static VALUE
-ruby_whisper_vad_params_get_min_speech_duration_ms(VALUE self)
-{
-  ruby_whisper_vad_params *rwvp;
-  TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-  return INT2NUM(rwvp->params.min_speech_duration_ms);
-}
-
-/*
- * Min silence duration to consider speech as ended.
- *
- * call-seq:
- *   min_silence_duration_ms = duration_ms -> duration_ms
- */
-static VALUE
-ruby_whisper_vad_params_set_min_silence_duration_ms(VALUE self, VALUE value)
-{
-  ruby_whisper_vad_params *rwvp;
-  TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-  rwvp->params.min_silence_duration_ms = NUM2INT(value);
-  return value;
-}
-
-static VALUE
-ruby_whisper_vad_params_get_min_silence_duration_ms(VALUE self)
-{
-  ruby_whisper_vad_params *rwvp;
-  TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-  return INT2NUM(rwvp->params.min_silence_duration_ms);
-}
-
-/*
- * Max duration of a speech segment before forcing a new segment.
- *
- * call-seq:
- *   max_speech_duration_s = duration_s -> duration_s
- */
-static VALUE
-ruby_whisper_vad_params_set_max_speech_duration_s(VALUE self, VALUE value)
-{
-  ruby_whisper_vad_params *rwvp;
-  TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-  rwvp->params.max_speech_duration_s = RFLOAT_VALUE(value);
-  return value;
-}
-
-static VALUE
-ruby_whisper_vad_params_get_max_speech_duration_s(VALUE self)
-{
-  ruby_whisper_vad_params *rwvp;
-  TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-  return DBL2NUM(rwvp->params.max_speech_duration_s);
-}
-
-/*
- * Padding added before and after speech segments.
- *
- * call-seq:
- *   speech_pad_ms = pad_ms -> pad_ms
- */
-static VALUE
-ruby_whisper_vad_params_set_speech_pad_ms(VALUE self, VALUE value)
-{
-  ruby_whisper_vad_params *rwvp;
-  TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-  rwvp->params.speech_pad_ms = NUM2INT(value);
-  return value;
-}
-
-static VALUE
-ruby_whisper_vad_params_get_speech_pad_ms(VALUE self)
-{
-  ruby_whisper_vad_params *rwvp;
-  TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-  return INT2NUM(rwvp->params.speech_pad_ms);
-}
-
-/*
- * Overlap in seconds when copying audio samples from speech segment.
- *
- * call-seq:
- *   samples_overlap = overlap -> overlap
- */
-static VALUE
-ruby_whisper_vad_params_set_samples_overlap(VALUE self, VALUE value)
-{
-  ruby_whisper_vad_params *rwvp;
-  TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-  rwvp->params.samples_overlap = RFLOAT_VALUE(value);
-  return value;
-}
-
-static VALUE
-ruby_whisper_vad_params_get_samples_overlap(VALUE self)
-{
-  ruby_whisper_vad_params *rwvp;
-  TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-  return DBL2NUM(rwvp->params.samples_overlap);
-}
-
-static VALUE
-ruby_whisper_vad_params_equal(VALUE self, VALUE other)
-{
-  ruby_whisper_vad_params *rwvp1;
-  ruby_whisper_vad_params *rwvp2;
-
-  if (self == other) {
-    return Qtrue;
-  }
-
-  if (!rb_obj_is_kind_of(other, cVADParams)) {
-    return Qfalse;
-  }
-
-  TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp1);
-  TypedData_Get_Struct(other, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp2);
-
-  if (rwvp1->params.threshold != rwvp2->params.threshold) {
-    return Qfalse;
-  }
-  if (rwvp1->params.min_speech_duration_ms != rwvp2->params.min_speech_duration_ms) {
-    return Qfalse;
-  }
-  if (rwvp1->params.min_silence_duration_ms != rwvp2->params.min_silence_duration_ms) {
-    return Qfalse;
-  }
-  if (rwvp1->params.max_speech_duration_s != rwvp2->params.max_speech_duration_s) {
-    return Qfalse;
-  }
-  if (rwvp1->params.speech_pad_ms != rwvp2->params.speech_pad_ms) {
-    return Qfalse;
-  }
-  if (rwvp1->params.samples_overlap != rwvp2->params.samples_overlap) {
-    return Qfalse;
-  }
-
-  return Qtrue;
-}
-
-#define SET_PARAM_IF_SAME(param_name) \
-  if (id == id_ ## param_name) { \
-    ruby_whisper_vad_params_set_ ## param_name(self, value); \
-    continue; \
-  }
-
-VALUE
-ruby_whisper_vad_params_initialize(int argc, VALUE *argv, VALUE self)
-{
-  VALUE kw_hash;
-  VALUE values[NUM_PARAMS] = {Qundef};
-  VALUE value;
-  ruby_whisper_vad_params *rwvp;
-  ID id;
-  int i;
-
-  TypedData_Get_Struct(self, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
-
-  rb_scan_args_kw(RB_SCAN_ARGS_KEYWORDS, argc, argv, ":", &kw_hash);
-  if (NIL_P(kw_hash)) {
-    return self;
-  }
-
-  rb_get_kwargs(kw_hash, param_names, 0, NUM_PARAMS, values);
-
-  for (i = 0; i < NUM_PARAMS; i++) {
-    id = param_names[i];
-    value = values[i];
-    if (value == Qundef) {
-      continue;
-    }
-    SET_PARAM_IF_SAME(threshold)
-    SET_PARAM_IF_SAME(min_speech_duration_ms)
-    SET_PARAM_IF_SAME(min_silence_duration_ms)
-    SET_PARAM_IF_SAME(max_speech_duration_s)
-    SET_PARAM_IF_SAME(speech_pad_ms)
-    SET_PARAM_IF_SAME(samples_overlap)
-  }
-
-  return self;
-}
-
-#undef SET_PARAM_IF_SAME
-
-void
-init_ruby_whisper_vad_params(VALUE *mVAD)
-{
-  cVADParams = rb_define_class_under(*mVAD, "Params", rb_cObject);
-  rb_define_alloc_func(cVADParams, ruby_whisper_vad_params_s_allocate);
-  rb_define_method(cVADParams, "initialize", ruby_whisper_vad_params_initialize, -1);
-
-  DEFINE_PARAM(threshold, 0)
-  DEFINE_PARAM(min_speech_duration_ms, 1)
-  DEFINE_PARAM(min_silence_duration_ms, 2)
-  DEFINE_PARAM(max_speech_duration_s, 3)
-  DEFINE_PARAM(speech_pad_ms, 4)
-  DEFINE_PARAM(samples_overlap, 5)
-
-  rb_define_method(cVADParams, "==", ruby_whisper_vad_params_equal, 1);
-}
-
-#undef DEFINE_PARAM
-#undef NUM_PARAMS
--- a/bindings/ruby/ext/sources/CMakeGraphVizOptions.cmake
+++ b/bindings/ruby/ext/sources/CMakeGraphVizOptions.cmake
@ -1,8 +0,0 @@
-set(GRAPHVIZ_EXECUTABLES FALSE)
-set(GRAPHVIZ_STATIC_LIBS TRUE)
-set(GRAPHVIZ_SHARED_LIBS FALSE)
-set(GRAPHVIZ_MODULE_LIBS FALSE)
-set(GRAPHVIZ_INTERFACE_LIBS FALSE)
-set(GRAPHVIZ_OBJECT_LIBS FALSE)
-set(GRAPHVIZ_UNKNOWN_LIBS FALSE)
-set(GRAPHVIZ_GENERATE_DEPENDERS FALSE)
--- a/bindings/ruby/extsources.rb
+++ b/bindings/ruby/extsources.rb
@ -1,40 +0,0 @@
-require "pathname"
-
-root = Pathname("..")/".."
-ignored_dirs = %w[
-  .devops
-  .github
-  ci
-  examples/wchess/wchess.wasm
-  examples/whisper.android
-  examples/whisper.android.java
-  examples/whisper.objc
-  examples/whisper.swiftui
-  grammars
-  models
-  samples
-  scripts
-].collect {|dir| root/dir}
-ignored_files = %w[
-  AUTHORS
-  Makefile
-  README.md
-  README_sycl.md
-  .gitignore
-  .gitmodules
-  .dockerignore
-  whisper.nvim
-  twitch.sh
-  yt-wsp.sh
-  close-issue.yml
-]
-
-EXTSOURCES =
-  `git ls-files -z #{root}`.split("\x0")
-    .collect {|file| Pathname(file)}
-    .reject {|file|
-      ignored_dirs.any? {|dir| file.descend.any? {|desc| desc == dir}} ||
-        ignored_files.include?(file.basename.to_path) ||
-        (file.descend.to_a[1] != root && file.descend.to_a[1] != Pathname("..")/"javascript")
-    }
-    .collect(&:to_path)
--- a/bindings/ruby/lib/whisper/context.rb
+++ b/bindings/ruby/lib/whisper/context.rb
@ -1,15 +0,0 @@
-module Whisper
-  class Context
-    def to_srt
-      each_segment.with_index.reduce("") {|srt, (segment, index)|
-        srt << "#{index + 1}\n#{segment.to_srt_cue}\n"
-      }
-    end
-
-    def to_webvtt
-      each_segment.with_index.reduce("WEBVTT\n\n") {|webvtt, (segment, index)|
-        webvtt << "#{index + 1}\n#{segment.to_webvtt_cue}\n"
-      }
-    end
-  end
-end
--- a/bindings/ruby/lib/whisper/model/uri.rb
+++ b/bindings/ruby/lib/whisper/model/uri.rb
@ -1,233 +0,0 @@
-require "uri"
-require "net/http"
-require "time"
-require "pathname"
-require "io/console/size"
-
-module Whisper
-  class Model
-    class URI
-      def initialize(uri)
-        @uri = URI(uri)
-      end
-
-      def to_path
-        cache
-        cache_path.to_path
-      end
-
-      def clear_cache
-        path = cache_path
-        path.delete if path.exist?
-      end
-
-      private
-
-      def cache_path
-        base_cache_dir/@uri.host/@uri.path[1..]
-      end
-
-      def base_cache_dir
-        base = case RUBY_PLATFORM
-               when /mswin|mingw/
-                 ENV.key?("LOCALAPPDATA") ? Pathname(ENV["LOCALAPPDATA"]) : Pathname(Dir.home)/"AppData/Local"
-               when /darwin/
-                 Pathname(Dir.home)/"Library/Caches"
-               else
-                 ENV.key?("XDG_CACHE_HOME") ? Pathname(ENV["XDG_CACHE_HOME"]) : Pathname(Dir.home)/".cache"
-               end
-        base/"whisper.cpp"
-      end
-
-      def cache
-        path = cache_path
-        headers = {}
-        headers["if-modified-since"] = path.mtime.httpdate if path.exist?
-        request @uri, headers
-        path
-      end
-
-      def request(uri, headers)
-        Net::HTTP.start uri.host, uri.port, use_ssl: uri.scheme == "https" do |http|
-          request = Net::HTTP::Get.new(uri, headers)
-          http.request request do |response|
-            case response
-            when Net::HTTPNotModified
-              # noop
-            when Net::HTTPOK
-              return if !response.key?("last-modified") && cache_path.exist?
-
-              download response
-            when Net::HTTPRedirection
-              request URI(response["location"]), headers
-            else
-              return if headers.key?("if-modified-since") # Use cache file
-
-              raise "#{response.code} #{response.message}\n#{response.body}"
-            end
-          end
-        end
-      rescue => err
-        if cache_path.exist?
-          warn err
-          # Use cache file
-        else
-          raise
-        end
-      end
-
-      def download(response)
-        path = cache_path
-        path.dirname.mkpath unless path.dirname.exist?
-        downloading_path = Pathname("#{path}.downloading")
-        size = response.content_length
-        downloading_path.open "wb" do |file|
-          downloaded = 0
-          response.read_body do |chunk|
-            file << chunk
-            downloaded += chunk.bytesize
-            show_progress downloaded, size
-          end
-          $stderr.puts
-        end
-        downloading_path.rename path
-      end
-
-      def show_progress(current, size)
-        progress_rate_available = size && $stderr.tty?
-
-        unless @prev
-          @prev = Time.now
-          $stderr.puts "Downloading #{@uri} to #{cache_path}"
-        end
-
-        now = Time.now
-
-        if progress_rate_available
-          return if now - @prev < 1 && current < size
-
-          progress_width = 20
-          progress = current.to_f / size
-          arrow_length = progress * progress_width
-          arrow = "=" * (arrow_length - 1) + ">" + " " * (progress_width - arrow_length)
-          line = "[#{arrow}] (#{format_bytesize(current)} / #{format_bytesize(size)})"
-          padding = ' ' * ($stderr.winsize[1] - line.size)
-          $stderr.print "\r#{line}#{padding}"
-        else
-          return if now - @prev < 1
-
-          $stderr.print "."
-        end
-        @prev = now
-      end
-
-      def format_bytesize(bytesize)
-        return "0.0 B" if bytesize.zero?
-
-        units = %w[B KiB MiB GiB TiB]
-        exp = (Math.log(bytesize) / Math.log(1024)).to_i
-        format("%.1f %s", bytesize.to_f / 1024 ** exp, units[exp])
-      end
-    end
-
-    class ZipURI < URI
-      def cache
-        zip_path = super
-        dest = unzipped_path
-        return if dest.exist? && dest.mtime >= zip_path.mtime
-        escaping dest do
-          system "unzip", "-q", "-d", zip_path.dirname.to_path, zip_path.to_path, exception: true
-        end
-        zip_path
-      end
-
-      def clear_cache
-        super
-        unzipped_path.rmtree if unzipped_path.exist?
-      end
-
-      private
-
-      def unzipped_path
-        cache_path.sub_ext("")
-      end
-
-      def escaping(path)
-        escaped = Pathname("#{path}.removing")
-        if path.exist?
-          escaped.rmtree if escaped.exist?
-          path.rename escaped
-        end
-        yield
-      ensure
-        if path.exist?
-          escaped.rmtree if escaped.exist?
-        else
-          escaped.rename path if escaped.exist?
-        end
-      end
-    end
-
-    @pre_converted_models = %w[
-      tiny
-      tiny.en
-      tiny-q5_1
-      tiny.en-q5_1
-      tiny-q8_0
-      base
-      base.en
-      base-q5_1
-      base.en-q5_1
-      base-q8_0
-      small
-      small.en
-      small.en-tdrz
-      small-q5_1
-      small.en-q5_1
-      small-q8_0
-      medium
-      medium.en
-      medium-q5_0
-      medium.en-q5_0
-      medium-q8_0
-      large-v1
-      large-v2
-      large-v2-q5_0
-      large-v2-q8_0
-      large-v3
-      large-v3-q5_0
-      large-v3-turbo
-      large-v3-turbo-q5_0
-      large-v3-turbo-q8_0
-    ].each_with_object({}) {|name, models|
-      models[name] = URI.new("https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-#{name}.bin")
-    }
-
-    %w[
-      silero-v5.1.2
-    ].each do |name|
-      @pre_converted_models[name] = URI.new("https://huggingface.co/ggml-org/whisper-vad/resolve/main/ggml-#{name}.bin")
-    end
-
-    @coreml_compiled_models = %w[
-      tiny
-      tiny.en
-      base
-      base.en
-      small
-      small.en
-      medium
-      medium.en
-      large-v1
-      large-v2
-      large-v3
-      large-v3-turbo
-    ].each_with_object({}) do |name, models|
-      models[@pre_converted_models[name]] = ZipURI.new("https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-#{name}-encoder.mlmodelc.zip")
-    end
-
-    class << self
-      attr_reader :pre_converted_models, :coreml_compiled_models
-    end
-  end
-end
--- a/bindings/ruby/lib/whisper/segment.rb
+++ b/bindings/ruby/lib/whisper/segment.rb
@ -1,58 +0,0 @@
-module Whisper
-  class Segment
-    SRT_ESCAPES = {
-      "&" => "&amp;",
-      "<" => "&lt;",
-      ">" => "&gt;",
-    }
-    SRT_ESCAPES_RE = Regexp.union(SRT_ESCAPES.keys)
-    private_constant :SRT_ESCAPES, :SRT_ESCAPES_RE
-
-    def to_srt_cue
-      "#{srt_start_time} --> #{srt_end_time}\n#{srt_text}\n"
-    end
-
-    def to_webvtt_cue
-      "#{webvtt_start_time} --> #{webvtt_end_time}\n#{webvtt_text}\n"
-    end
-
-    private
-
-    def time_to_a(time)
-      sec, decimal_part = time.divmod(1000)
-      min, sec = sec.divmod(60)
-      hour, min = min.divmod(60)
-      [hour, min, sec, decimal_part]
-    end
-
-    def srt_time(time)
-      "%02d:%02d:%02d,%03d" % time_to_a(time)
-    end
-
-    def srt_start_time
-      srt_time(start_time)
-    end
-
-    def srt_end_time
-      srt_time(end_time)
-    end
-
-    def srt_text
-      text.gsub(SRT_ESCAPES_RE, SRT_ESCAPES)
-    end
-
-    def webvtt_time(time)
-      "%02d:%02d:%02d.%03d" % time_to_a(time)
-    end
-
-    def webvtt_start_time
-      webvtt_time(start_time)
-    end
-
-    def webvtt_end_time
-      webvtt_time(end_time)
-    end
-
-    alias webvtt_text srt_text
-  end
-end
--- a/bindings/ruby/sig/whisper.rbs
+++ b/bindings/ruby/sig/whisper.rbs
@ -1,509 +0,0 @@
-module Whisper
-  interface _Samples
-    def length: () -> Integer
-    def each: { (Float) -> void } -> void
-  end
-
-  type log_callback = ^(Integer level, String message, Object user_data) -> void
-  type new_segment_callback = ^(Whisper::Context, void, Integer n_new, Object user_data) -> void
-  type progress_callback = ^(Whisper::Context, void, Integer progress, Object user_data) -> void
-  type encoder_begin_callback = ^(Whisper::Context, void, Object user_data) -> void
-  type abort_callback = ^(Whisper::Context, void, Object user_data) -> boolish
-
-  LOG_LEVEL_NONE: Integer
-  LOG_LEVEL_INFO: Integer
-  LOG_LEVEL_WARN: Integer
-  LOG_LEVEL_ERROR: Integer
-  LOG_LEVEL_DEBUG: Integer
-  LOG_LEVEL_CONT: Integer
-
-  def self.lang_max_id: () -> Integer
-  def self.lang_id: (string name) -> Integer
-  def self.lang_str: (Integer id) -> String
-  def self.lang_str_full: (Integer id) -> String
-  def self.log_set: (log_callback, Object? user_data) -> log_callback
-  def self.system_info_str: () -> String
-
-  class Context
-    def self.new: (String | path | ::URI::HTTP) -> instance
-
-    # transcribe a single file
-    # can emit to a block results
-    #
-    #     params = Whisper::Params.new
-    #     params.duration = 60_000
-    #     whisper.transcribe "path/to/audio.wav", params do |text|
-    #       puts text
-    #     end
-    #
-    def transcribe: (string, Params, ?n_processors: Integer) -> self
-                  | (string, Params, ?n_processors: Integer) { (String) -> void } -> self
-
-    def model_n_vocab: () -> Integer
-    def model_n_audio_ctx: () -> Integer
-    def model_n_audio_state: () -> Integer
-    def model_n_text_head: () -> Integer
-    def model_n_text_layer: () -> Integer
-    def model_n_mels: () -> Integer
-    def model_ftype: () -> Integer
-    def model_type: () -> String
-
-    # Yields each Whisper::Segment:
-    #
-    #     whisper.transcribe("path/to/audio.wav", params)
-    #     whisper.each_segment do |segment|
-    #       puts segment.text
-    #     end
-    #
-    # Returns an Enumerator if no block given:
-    #
-    #     whisper.transcribe("path/to/audio.wav", params)
-    #     enum = whisper.each_segment
-    #     enum.to_a # => [#<Whisper::Segment>, ...]
-    #
-    def each_segment: { (Segment) -> void } -> void
-                    | () -> Enumerator[Segment]
-
-    def model: () -> Model
-    def full_get_segment: (Integer nth) -> Segment
-    def full_n_segments: () -> Integer
-
-    # Language ID, which can be converted to string by Whisper.lang_str and Whisper.lang_str_full.
-    #
-    def full_lang_id: () -> Integer
-
-    # Start time of a segment indexed by +segment_index+ in centiseconds (10 times milliseconds).
-    #
-    #     full_get_segment_t0(3) # => 1668 (16680 ms)
-    #
-    def full_get_segment_t0: (Integer) -> Integer
-
-    # End time of a segment indexed by +segment_index+ in centiseconds (10 times milliseconds).
-    #
-    #     full_get_segment_t1(3) # => 1668 (16680 ms)
-    #
-    def full_get_segment_t1: (Integer) -> Integer
-
-    # Whether the next segment indexed by +segment_index+ is predicated as a speaker turn.
-    #
-    #     full_get_segment_speacker_turn_next(3) # => true
-    #
-    def full_get_segment_speaker_turn_next: (Integer) -> (true | false)
-
-    # Text of a segment indexed by +segment_index+.
-    #
-    #     full_get_segment_text(3) # => "ask not what your country can do for you, ..."
-    #
-    def full_get_segment_text: (Integer) -> String
-
-    def full_get_segment_no_speech_prob: (Integer) -> Float
-
-    # Run the entire model: PCM -> log mel spectrogram -> encoder -> decoder -> text
-    # Not thread safe for same context
-    # Uses the specified decoding strategy to obtain the text.
-    #
-    # The second argument +samples+ must be an array of samples, respond to :length, or be a MemoryView of an array of float. It must be 32 bit float PCM audio data.
-    #
-    def full: (Params, Array[Float] samples, ?Integer n_samples) -> self
-            | (Params, _Samples, ?Integer n_samples) -> self
-
-    # Split the input audio in chunks and process each chunk separately using whisper_full_with_state()
-    # Result is stored in the default state of the context
-    # Not thread safe if executed in parallel on the same context.
-    # It seems this approach can offer some speedup in some cases.
-    # However, the transcription accuracy can be worse at the beginning and end of each chunk.
-    #
-    def full_parallel: (Params, Array[Float], ?Integer n_samples) -> self
-                     | (Params, _Samples, ?Integer n_samples) -> self
-                     | (Params, _Samples, ?Integer? n_samples, Integer n_processors) -> self
-
-    def to_srt: () -> String
-    def to_webvtt: () -> String
-  end
-
-  class Params
-    def self.new: (
-      ?language: string,
-      ?translate: boolish,
-      ?no_context: boolish,
-      ?single_segment: boolish,
-      ?print_special: boolish,
-      ?print_progress: boolish,
-      ?print_realtime: boolish,
-      ?print_timestamps: boolish,
-      ?suppress_blank: boolish,
-      ?suppress_nst: boolish,
-      ?token_timestamps: boolish,
-      ?split_on_word: boolish,
-      ?initial_prompt: string | nil,
-      ?diarize: boolish,
-      ?offset: Integer,
-      ?duration: Integer,
-      ?max_text_tokens: Integer,
-      ?temperature: Float,
-      ?max_initial_ts: Float,
-      ?length_penalty: Float,
-      ?temperature_inc: Float,
-      ?entropy_thold: Float,
-      ?logprob_thold: Float,
-      ?no_speech_thold: Float,
-      ?new_segment_callback: new_segment_callback,
-      ?new_segment_callback_user_data: Object,
-      ?progress_callback: progress_callback,
-      ?progress_callback_user_data: Object,
-      ?encoder_begin_callback: encoder_begin_callback,
-      ?encoder_begin_callback_user_data: Object,
-      ?abort_callback: abort_callback,
-      ?abort_callback_user_data: Object,
-      ?vad: boolish,
-      ?vad_model_path: path | URI,
-      ?vad_params: Whisper::VAD::Params
-    ) -> instance
-
-    # params.language = "auto" | "en", etc...
-    #
-    def language=: (String) -> String # TODO: Enumerate lang names
-
-    def language: () -> String
-    def translate=: (boolish) -> boolish
-    def translate: () -> (true | false)
-    def no_context=: (boolish) -> boolish
-
-    # If true, does not use past transcription (if any) as initial prompt for the decoder.
-    #
-    def no_context: () -> (true | false)
-
-    def single_segment=: (boolish) -> boolish
-
-    # If true, forces single segment output (useful for streaming).
-    #
-    def single_segment: () -> (true | false)
-
-    def print_special=: (boolish) -> boolish
-
-    # If true, prints special tokens (e.g. <SOT>, <EOT>, <BEG>, etc.).
-    #
-    def print_special: () -> (true | false)
-
-    def print_progress=: (boolish) -> boolish
-
-    # If true, prints progress information.
-    #
-    def print_progress: () -> (true | false)
-
-    def print_realtime=: (boolish) -> boolish
-
-    # If true, prints results from within whisper.cpp. (avoid it, use callback instead)
-    #
-    def print_realtime: () -> (true | false)
-
-    # If true, prints timestamps for each text segment when printing realtime.
-    #
-    def print_timestamps=: (boolish) -> boolish
-
-    def print_timestamps: () -> (true | false)
-
-    def suppress_blank=: (boolish) -> boolish
-
-    # If true, suppresses blank outputs.
-    #
-    def suppress_blank: () -> (true | false)
-
-    def suppress_nst=: (boolish) -> boolish
-
-    # If true, suppresses non-speech-tokens.
-    #
-    def suppress_nst: () -> (true | false)
-
-    def token_timestamps=: (boolish) -> boolish
-
-    # If true, enables token-level timestamps.
-    #
-    def token_timestamps: () -> (true | false)
-
-    def split_on_word=: (boolish) -> boolish
-
-    # If true, split on word rather than on token (when used with max_len).
-    #
-    def split_on_word: () -> (true | false)
-
-    def initial_prompt=: (_ToS) -> _ToS
-
-    # Tokens to provide to the whisper decoder as initial prompt
-    # these are prepended to any existing text context from a previous call
-    # use whisper_tokenize() to convert text to tokens.
-    # Maximum of whisper_n_text_ctx()/2 tokens are used (typically 224).
-    #
-    def initial_prompt: () -> (String | nil)
-
-    def diarize=: (boolish) -> boolish
-
-    # If true, enables diarization.
-    #
-    def diarize: () -> (true | false)
-
-    def offset=: (Integer) -> Integer
-
-    # Start offset in ms.
-    #
-    def offset: () -> Integer
-
-    def duration=: (Integer) -> Integer
-
-    # Audio duration to process in ms.
-    #
-    def duration: () -> Integer
-
-    def max_text_tokens=: (Integer) -> Integer
-
-    # Max tokens to use from past text as prompt for the decoder.
-    #
-    def max_text_tokens: () -> Integer
-
-    def temperature=: (Float) -> Float
-    def temperature: () -> Float
-    def max_initial_ts=: (Float) -> Float
-
-    # See https://github.com/openai/whisper/blob/f82bc59f5ea234d4b97fb2860842ed38519f7e65/whisper/decoding.py#L97
-    #
-    def max_initial_ts: () -> Float
-
-    def length_penalty=: (Float) -> Float
-    def length_penalty: () -> Float
-    def temperature_inc=: (Float) -> Float
-    def temperature_inc: () -> Float
-    def entropy_thold=: (Float) -> Float
-
-    # Similar to OpenAI's "compression_ratio_threshold"
-    #
-    def entropy_thold: () -> Float
-
-    def logprob_thold=: (Float) -> Float
-    def logprob_thold: () -> Float
-    def no_speech_thold=: (Float) -> Float
-    def no_speech_thold: () -> Float
-
-    # Sets new segment callback, called for every newly generated text segment.
-    #
-    #     params.new_segment_callback = ->(context, _, n_new, user_data) {
-    #       # ...
-    #     }
-    #
-    def new_segment_callback=: (new_segment_callback) -> new_segment_callback
-    def new_segment_callback: () -> (new_segment_callback | nil)
-
-    # Sets user data passed to the last argument of new segment callback.
-    #
-    def new_segment_callback_user_data=: (Object) -> Object
-
-    def new_segment_callback_user_data: () -> Object
-
-    # Sets progress callback, called on each progress update.
-    #
-    #     params.new_segment_callback = ->(context, _, progress, user_data) {
-    #       # ...
-    #     }
-    #
-    # +progress+ is an Integer between 0 and 100.
-    #
-    def progress_callback=: (progress_callback) -> progress_callback
-
-    def progress_callback: () -> (progress_callback | nil)
-
-    # Sets user data passed to the last argument of progress callback.
-    #
-    def progress_callback_user_data=: (Object) -> Object
-
-    def progress_callback_user_data: () -> Object
-
-    # Sets encoder begin callback, called when the encoder starts.
-    #
-    def encoder_begin_callback=: (encoder_begin_callback) -> encoder_begin_callback
-
-    def encoder_begin_callback: () -> (encoder_begin_callback | nil)
-
-    # Sets user data passed to the last argument of encoder begin callback.
-    #
-    def encoder_begin_callback_user_data=: (Object) -> Object
-
-    def encoder_begin_callback_user_data: () -> Object
-
-    # Sets abort callback, called to check if the process should be aborted.
-    #
-    #     params.abort_callback = ->(user_data) {
-    #       # ...
-    #     }
-    #
-    #
-    def abort_callback=: (abort_callback) -> abort_callback
-
-    def abort_callback: () -> (abort_callback | nil)
-
-    # Sets user data passed to the last argument of abort callback.
-    #
-    def abort_callback_user_data=: (Object) -> Object
-
-    def abort_callback_user_data: () -> Object
-
-    # Enable VAD
-    #
-    def vad=: (boolish) -> boolish
-
-    def vad: () -> (true | false)
-
-    # Path to the VAD model
-    def vad_model_path=: (path | URI | nil) -> (path | URI | nil)
-
-    def vad_model_path: () -> (String | nil)
-
-    def vad_params=: (Whisper::VAD::Params) -> Whisper::VAD::Params
-    def vad_params: () -> (Whisper::VAD::Params)
-
-    # Hook called on new segment. Yields each Whisper::Segment.
-    #
-    #     whisper.on_new_segment do |segment|
-    #       # ...
-    #     end
-    #
-    def on_new_segment: { (Segment) -> void } -> void
-
-    # Hook called on progress update. Yields each progress Integer between 0 and 100.
-    #
-    def on_progress: { (Integer progress) -> void } -> void
-
-    # Hook called on encoder starts.
-    #
-    def on_encoder_begin: { () -> void } -> void
-
-    # Call block to determine whether abort or not. Return +true+ when you want to abort.
-    #
-    #     params.abort_on do
-    #       if some_condition
-    #         true # abort
-    #       else
-    #         false # continue
-    #       end
-    #     end
-    #
-    def abort_on: { (Object user_data) -> boolish } -> void
-  end
-
-  class Model
-    def self.pre_converted_models: () -> Hash[String, Model::URI]
-    def self.coreml_compiled_models: () -> Hash[Model::URI, Model::ZipURI]
-    def self.new: () -> instance
-    def n_vocab: () -> Integer
-    def n_audio_ctx: () -> Integer
-    def n_audio_state: () -> Integer
-    def n_audio_head: () -> Integer
-    def n_audio_layer: () -> Integer
-    def n_text_ctx: () -> Integer
-    def n_text_state: () -> Integer
-    def n_text_head: () -> Integer
-    def n_text_layer: () -> Integer
-    def n_mels: () -> Integer
-    def ftype: () -> Integer
-    def type: () -> String
-
-    class URI
-      def self.new: (string | ::URI::HTTP) -> instance
-      def to_path: -> String
-      def clear_cache: -> void
-    end
-
-    class ZipURI < URI
-      def cache: () -> Pathname
-      def clear_cache: () -> void
-    end
-  end
-
-  class Segment
-    type deconstructed_keys = {
-      start_time: (Integer | nil),
-      end_time: (Integer | nil),
-      text: (String | nil),
-      no_speech_prob: (Float | nil),
-      speaker_turn_next: (true | false | nil)
-    }
-
-    # Start time in milliseconds.
-    #
-    def start_time: () -> Integer
-
-    # End time in milliseconds.
-    #
-    def end_time: () -> Integer
-
-    # Whether the next segment is predicted as a speaker turn.
-    def speaker_turn_next?: () -> (true | false)
-
-    def text: () -> String
-    def no_speech_prob: () -> Float
-    def to_srt_cue: () -> String
-    def to_webvtt_cue: () -> String
-
-    #  Possible keys: :start_time, :end_time, :text, :no_speech_prob, :speaker_turn_next
-    #
-    #      whisper.each_segment do |segment|
-    #        segment => {start_time:, end_time:, text:, no_speech_prob:, speaker_turn_next:}
-    #
-    #        puts "[#{start_time} --> #{end_time}] #{text} (no speech prob: #{no_speech_prob}#{speaker_turn_next ? ', speaker turns next' : ''})"
-    #      end
-    def deconstruct_keys: (Array[:start_time | :end_time | :text | :no_speech_prob | :speaker_turn_next] | nil) -> deconstructed_keys
-  end
-
-  module VAD
-    class Params
-      def self.new: (
-        ?threshold: Float,
-        ?min_speech_duration_ms: Integer,
-        ?min_silence_duration_ms: Integer,
-        ?max_speech_duration_s: Float,
-        ?speech_pad_ms: Integer,
-        ?samples_overlap: Float
-      ) -> instance
-
-      # Probability threshold to consider as speech.
-      #
-      def threshold=: (Float) -> Float
-
-      def threshold: () -> Float
-
-      # Min duration for a valid speech segment.
-      #
-      def min_speech_duration_ms=: (Integer) -> Integer
-
-      def min_speech_duration_ms: () -> Integer
-
-      # Min silence duration to consider speech as ended.
-      #
-      def min_silence_duration_ms=: (Integer) -> Integer
-
-      def min_silence_duration_ms: () -> Integer
-
-      # Max duration of a speech segment before forcing a new segment.
-      def max_speech_duration_s=: (Float) -> Float
-
-      def max_speech_duration_s: () -> Float
-
-      # Padding added before and after speech segments.
-      #
-      def speech_pad_ms=: (Integer) -> Integer
-
-      def speech_pad_ms: () -> Integer
-
-      # Overlap in seconds when copying audio samples from speech segment.
-      #
-      def samples_overlap=: (Float) -> Float
-
-      def samples_overlap: () -> Float
-      def ==: (Params) -> (true | false)
-    end
-  end
-
-  class Error < StandardError
-    attr_reader code: Integer
-
-    def self.new: (Integer code) -> instance
-  end
-end
--- a/bindings/ruby/test/helper.rb
+++ b/bindings/ruby/test/helper.rb
@ -1,24 +0,0 @@
-require "test/unit"
-require "whisper"
-require_relative "jfk_reader/jfk_reader"
-
-class TestBase < Test::Unit::TestCase
-  AUDIO = File.join(__dir__, "fixtures", "jfk.wav")
-
-  class << self
-    def whisper
-      return @whisper if @whisper
-
-      @whisper = Whisper::Context.new("base.en")
-      params = Whisper::Params.new
-      params.print_timestamps = false
-      @whisper.transcribe(TestBase::AUDIO, params)
-    end
-  end
-
-  private
-
-  def whisper
-    self.class.whisper
-  end
-end
--- a/bindings/ruby/test/jfk_reader/.gitignore
+++ b/bindings/ruby/test/jfk_reader/.gitignore
@ -1,5 +0,0 @@
-Makefile
-jfk_reader.o
-jfk_reader.so
-jfk_reader.bundle
-jfk_reader.dll
--- a/bindings/ruby/test/jfk_reader/extconf.rb
+++ b/bindings/ruby/test/jfk_reader/extconf.rb
@ -1,3 +0,0 @@
-require "mkmf"
-
-create_makefile("jfk_reader")
--- a/bindings/ruby/test/jfk_reader/jfk_reader.c
+++ b/bindings/ruby/test/jfk_reader/jfk_reader.c
@ -1,68 +0,0 @@
-#include <ruby.h>
-#include <ruby/memory_view.h>
-#include <ruby/encoding.h>
-
-static VALUE
-jfk_reader_initialize(VALUE self, VALUE audio_path)
-{
-  rb_iv_set(self, "audio_path", audio_path);
-  return Qnil;
-}
-
-static bool
-jfk_reader_get_memory_view(const VALUE obj, rb_memory_view_t *view, int flags)
-{
-  VALUE audio_path = rb_iv_get(obj, "audio_path");
-  const char *audio_path_str = StringValueCStr(audio_path);
-  const int n_samples = 176000;
-  float *data = (float *)malloc(n_samples * sizeof(float));
-  short *samples = (short *)malloc(n_samples * sizeof(short));
-  FILE *file = fopen(audio_path_str, "rb");
-
-  fseek(file, 78, SEEK_SET);
-  fread(samples, sizeof(short), n_samples, file);
-  fclose(file);
-  for (int i = 0; i < n_samples; i++) {
-    data[i] = samples[i]/32768.0;
-  }
-
-  view->obj = obj;
-  view->data = (void *)data;
-  view->byte_size = sizeof(float) * n_samples;
-  view->readonly = true;
-  view->format = "f";
-  view->item_size = sizeof(float);
-  view->item_desc.components = NULL;
-  view->item_desc.length = 0;
-  view->ndim = 1;
-  view->shape = NULL;
-  view->sub_offsets = NULL;
-  view->private_data = NULL;
-
-  return true;
-}
-
-static bool
-jfk_reader_release_memory_view(const VALUE obj, rb_memory_view_t *view)
-{
-  return true;
-}
-
-static bool
-jfk_reader_memory_view_available_p(const VALUE obj)
-{
-  return true;
-}
-
-static const rb_memory_view_entry_t jfk_reader_view_entry = {
-  jfk_reader_get_memory_view,
-  jfk_reader_release_memory_view,
-  jfk_reader_memory_view_available_p
-};
-
-void Init_jfk_reader(void)
-{
-  VALUE cJFKReader = rb_define_class("JFKReader", rb_cObject);
-  rb_memory_view_register(cJFKReader, &jfk_reader_view_entry);
-  rb_define_method(cJFKReader, "initialize", jfk_reader_initialize, 1);
-}
--- a/bindings/ruby/test/test_callback.rb
+++ b/bindings/ruby/test/test_callback.rb
@ -1,202 +0,0 @@
-require_relative "helper"
-
-class TestCallback < TestBase
-  def setup
-    GC.start
-    @params = Whisper::Params.new
-    @whisper = Whisper::Context.new("base.en")
-    @audio = File.join(AUDIO)
-  end
-
-  def test_new_segment_callback
-    @params.new_segment_callback = ->(context, state, n_new, user_data) {
-      assert_kind_of Integer, n_new
-      assert n_new > 0
-      assert_same @whisper, context
-
-      n_segments = context.full_n_segments
-      n_new.times do |i|
-        i_segment = n_segments - 1 + i
-        start_time = context.full_get_segment_t0(i_segment) * 10
-        end_time = context.full_get_segment_t1(i_segment) * 10
-        text = context.full_get_segment_text(i_segment)
-
-        assert_kind_of Integer, start_time
-        assert start_time >= 0
-        assert_kind_of Integer, end_time
-        assert end_time > 0
-        assert_match(/ask not what your country can do for you, ask what you can do for your country/, text) if i_segment == 0
-      end
-    }
-
-    @whisper.transcribe(@audio, @params)
-  end
-
-  def test_new_segment_callback_closure
-    search_word = "what"
-    @params.new_segment_callback = ->(context, state, n_new, user_data) {
-      n_segments = context.full_n_segments
-      n_new.times do |i|
-        i_segment = n_segments - 1 + i
-        text = context.full_get_segment_text(i_segment)
-        if text.include?(search_word)
-          t0 = context.full_get_segment_t0(i_segment)
-          t1 = context.full_get_segment_t1(i_segment)
-          raise "search word '#{search_word}' found at between #{t0} and #{t1}"
-        end
-      end
-    }
-
-    assert_raise RuntimeError do
-      @whisper.transcribe(@audio, @params)
-    end
-  end
-
-  def test_new_segment_callback_user_data
-    udata = Object.new
-    @params.new_segment_callback_user_data = udata
-    @params.new_segment_callback = ->(context, state, n_new, user_data) {
-      assert_same udata, user_data
-    }
-
-    @whisper.transcribe(@audio, @params)
-  end
-
-  def test_new_segment_callback_user_data_gc
-    @params.new_segment_callback_user_data = "My user data"
-    @params.new_segment_callback = ->(context, state, n_new, user_data) {
-      assert_equal "My user data", user_data
-    }
-    GC.start
-
-    assert_same @whisper, @whisper.transcribe(@audio, @params)
-  end
-
-  def test_progress_callback
-    first = nil
-    last = nil
-    @params.progress_callback = ->(context, state, progress, user_data) {
-      assert_kind_of Integer, progress
-      assert 0 <= progress && progress <= 100
-      assert_same @whisper, context
-      first = progress if first.nil?
-      last = progress
-    }
-    @whisper.transcribe(@audio, @params)
-    assert_equal 0, first
-    assert_equal 100, last
-  end
-
-  def test_progress_callback_user_data
-    udata = Object.new
-    @params.progress_callback_user_data = udata
-    @params.progress_callback = ->(context, state, n_new, user_data) {
-      assert_same udata, user_data
-    }
-
-    @whisper.transcribe(@audio, @params)
-  end
-
-  def test_on_progress
-    first = nil
-    last = nil
-    @params.on_progress do |progress|
-      assert_kind_of Integer, progress
-      assert 0 <= progress && progress <= 100
-      first = progress if first.nil?
-      last = progress
-    end
-    @whisper.transcribe(@audio, @params)
-    assert_equal 0, first
-    assert_equal 100, last
-  end
-
-  def test_encoder_begin_callback
-    i = 0
-    @params.encoder_begin_callback = ->(context, state, user_data) {
-      i += 1
-    }
-    @whisper.transcribe(@audio, @params)
-    assert i > 0
-  end
-
-  def test_encoder_begin_callback_abort
-    logs = []
-    Whisper.log_set -> (level, buffer, user_data) {
-      logs << buffer if level == Whisper::LOG_LEVEL_ERROR
-    }, logs
-    @params.encoder_begin_callback = ->(context, state, user_data) {
-      return false
-    }
-    @whisper.transcribe(@audio, @params)
-    assert_match(/encoder_begin_callback returned false - aborting/, logs.join)
-    Whisper.log_set ->(level, buffer, user_data) {}, nil
-  end
-
-  def test_encoder_begin_callback_user_data
-    udata = Object.new
-    @params.encoder_begin_callback_user_data = udata
-    yielded = nil
-    @params.encoder_begin_callback = ->(context, state, user_data) {
-      yielded = user_data
-    }
-    @whisper.transcribe(@audio, @params)
-    assert_same udata, yielded
-  end
-
-  def test_on_encoder_begin
-    i = 0
-    @params.on_encoder_begin do
-      i += 1
-    end
-    @whisper.transcribe(@audio, @params)
-    assert i > 0
-  end
-
-  def test_abort_callback
-    i = 0
-    @params.abort_callback = ->(user_data) {
-      assert_nil user_data
-      i += 1
-      return false
-    }
-    @whisper.transcribe(@audio, @params)
-    assert i > 0
-  end
-
-  def test_abort_callback_abort
-    i = 0
-    @params.abort_callback = ->(user_data) {
-      i += 1
-      return i == 3
-    }
-    @whisper.transcribe(@audio, @params)
-    assert_equal 3, i
-  end
-
-  def test_abort_callback_user_data
-    udata = Object.new
-    @params.abort_callback_user_data = udata
-    yielded = nil
-    @params.abort_callback = ->(user_data) {
-      yielded = user_data
-    }
-    @whisper.transcribe(@audio, @params)
-    assert_same udata, yielded
-  end
-
-  def test_abort_on
-    do_abort = false
-    _aborted_from_callback = false
-    @params.on_new_segment do |segment|
-      do_abort = true if segment.text.match?(/ask/)
-    end
-    i = 0
-    @params.abort_on do
-      i += 1
-      do_abort
-    end
-    @whisper.transcribe(@audio, @params)
-    assert i > 0
-  end
-end
--- a/bindings/ruby/test/test_error.rb
+++ b/bindings/ruby/test/test_error.rb
@ -1,20 +0,0 @@
-require_relative "helper"
-
-class TestError < TestBase
-  def test_error
-    error = Whisper::Error.new(-2)
-    assert_equal "failed to compute log mel spectrogram", error.message
-    assert_equal(-2, error.code)
-  end
-
-  def test_unknown_error
-    error = Whisper::Error.new(-20)
-    assert_equal "unknown error", error.message
-  end
-
-  def test_non_int_code
-    assert_raise TypeError do
-      _error = Whisper::Error.new("non int")
-    end
-  end
-end
--- a/bindings/ruby/test/test_model.rb
+++ b/bindings/ruby/test/test_model.rb
@ -1,118 +0,0 @@
-require_relative "helper"
-require "pathname"
-
-class TestModel < TestBase
-  def test_model
-    whisper = Whisper::Context.new("base.en")
-    assert_instance_of Whisper::Model, whisper.model
-  end
-
-  def test_attributes
-    whisper = Whisper::Context.new("base.en")
-    model = whisper.model
-
-    assert_equal 51864, model.n_vocab
-    assert_equal 1500, model.n_audio_ctx
-    assert_equal 512, model.n_audio_state
-    assert_equal 8, model.n_audio_head
-    assert_equal 6, model.n_audio_layer
-    assert_equal 448, model.n_text_ctx
-    assert_equal 512, model.n_text_state
-    assert_equal 8, model.n_text_head
-    assert_equal 6, model.n_text_layer
-    assert_equal 80, model.n_mels
-    assert_equal 1, model.ftype
-    assert_equal "base", model.type
-  end
-
-  def test_gc
-    model = Whisper::Context.new("base.en").model
-    GC.start
-
-    assert_equal 51864, model.n_vocab
-    assert_equal 1500, model.n_audio_ctx
-    assert_equal 512, model.n_audio_state
-    assert_equal 8, model.n_audio_head
-    assert_equal 6, model.n_audio_layer
-    assert_equal 448, model.n_text_ctx
-    assert_equal 512, model.n_text_state
-    assert_equal 8, model.n_text_head
-    assert_equal 6, model.n_text_layer
-    assert_equal 80, model.n_mels
-    assert_equal 1, model.ftype
-    assert_equal "base", model.type
-  end
-
-  def test_pathname
-    path = Pathname(Whisper::Model.pre_converted_models["base.en"].to_path)
-    whisper = Whisper::Context.new(path)
-    model = whisper.model
-
-    assert_equal 51864, model.n_vocab
-    assert_equal 1500, model.n_audio_ctx
-    assert_equal 512, model.n_audio_state
-    assert_equal 8, model.n_audio_head
-    assert_equal 6, model.n_audio_layer
-    assert_equal 448, model.n_text_ctx
-    assert_equal 512, model.n_text_state
-    assert_equal 8, model.n_text_head
-    assert_equal 6, model.n_text_layer
-    assert_equal 80, model.n_mels
-    assert_equal 1, model.ftype
-    assert_equal "base", model.type
-  end
-
-  def test_auto_download
-    path = Whisper::Model.pre_converted_models["base.en"].to_path
-
-    assert_path_exist path
-    assert_equal 147964211, File.size(path)
-  end
-
-  def test_uri_string
-    path = "https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin"
-    whisper = Whisper::Context.new(path)
-    model = whisper.model
-
-    assert_equal 51864, model.n_vocab
-    assert_equal 1500, model.n_audio_ctx
-    assert_equal 512, model.n_audio_state
-    assert_equal 8, model.n_audio_head
-    assert_equal 6, model.n_audio_layer
-    assert_equal 448, model.n_text_ctx
-    assert_equal 512, model.n_text_state
-    assert_equal 8, model.n_text_head
-    assert_equal 6, model.n_text_layer
-    assert_equal 80, model.n_mels
-    assert_equal 1, model.ftype
-    assert_equal "base", model.type
-  end
-
-  def test_uri
-    path = URI("https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin")
-    whisper = Whisper::Context.new(path)
-    model = whisper.model
-
-    assert_equal 51864, model.n_vocab
-    assert_equal 1500, model.n_audio_ctx
-    assert_equal 512, model.n_audio_state
-    assert_equal 8, model.n_audio_head
-    assert_equal 6, model.n_audio_layer
-    assert_equal 448, model.n_text_ctx
-    assert_equal 512, model.n_text_state
-    assert_equal 8, model.n_text_head
-    assert_equal 6, model.n_text_layer
-    assert_equal 80, model.n_mels
-    assert_equal 1, model.ftype
-    assert_equal "base", model.type
-  end
-
-  def test_coreml_model_auto_download
-    uri = Whisper::Model.coreml_compiled_models[Whisper::Model.pre_converted_models["tiny"]]
-    model_path = Pathname(uri.to_path).sub_ext("")
-    model_path.rmtree if model_path.exist?
-
-    uri.cache
-    assert_path_exist model_path
-  end
-end
--- a/bindings/ruby/test/test_package.rb
+++ b/bindings/ruby/test/test_package.rb
@ -1,50 +0,0 @@
-require_relative "helper"
-require 'tempfile'
-require 'tmpdir'
-require 'shellwords'
-
-class TestPackage < TestBase
-  def test_build
-    Tempfile.create do |file|
-      assert system("gem", "build", "whispercpp.gemspec", "--output", file.to_path.shellescape, exception: true)
-      assert file.size > 0
-      assert_path_exist file.to_path
-    end
-  end
-
-  sub_test_case "Building binary on installation" do
-    def setup
-      system "rake", "build", exception: true
-    end
-
-    def test_install
-      gemspec = Gem::Specification.load("whispercpp.gemspec")
-      Dir.mktmpdir do |dir|
-        system "gem", "install", "--install-dir", dir.shellescape, "--no-document", "pkg/#{gemspec.file_name.shellescape}", exception: true
-        assert_installed dir, gemspec.version
-      end
-    end
-
-    def test_install_with_coreml
-      omit_unless RUBY_PLATFORM.match?(/darwin/) do
-        gemspec = Gem::Specification.load("whispercpp.gemspec")
-        Dir.mktmpdir do |dir|
-          system "gem", "install", "--install-dir", dir.shellescape, "--no-document", "pkg/#{gemspec.file_name.shellescape}", "--", "--enable-whisper-coreml", exception: true
-          assert_installed dir, gemspec.version
-          assert_nothing_raised do
-            libdir = File.join(dir, "gems", "#{gemspec.name}-#{gemspec.version}", "lib")
-            system "ruby", "-I", libdir, "-r", "whisper", "-e", "Whisper::Context.new('tiny')", exception: true
-          end
-        end
-      end
-    end
-
-    private
-
-    def assert_installed(dir, version)
-      assert_path_exist File.join(dir, "gems/whispercpp-#{version}/lib", "whisper.#{RbConfig::CONFIG["DLEXT"]}")
-      assert_path_exist File.join(dir, "gems/whispercpp-#{version}/LICENSE")
-      assert_path_not_exist File.join(dir, "gems/whispercpp-#{version}/ext/build")
-    end
-  end
-end
--- a/bindings/ruby/test/test_params.rb
+++ b/bindings/ruby/test/test_params.rb
@ -1,297 +0,0 @@
-require_relative "helper"
-
-class TestParams < TestBase
-  PARAM_NAMES = [
-    :language,
-    :translate,
-    :no_context,
-    :single_segment,
-    :print_special,
-    :print_progress,
-    :print_realtime,
-    :print_timestamps,
-    :suppress_blank,
-    :suppress_nst,
-    :token_timestamps,
-    :split_on_word,
-    :initial_prompt,
-    :diarize,
-    :offset,
-    :duration,
-    :max_text_tokens,
-    :temperature,
-    :max_initial_ts,
-    :length_penalty,
-    :temperature_inc,
-    :entropy_thold,
-    :logprob_thold,
-    :no_speech_thold,
-    :new_segment_callback,
-    :new_segment_callback_user_data,
-    :progress_callback,
-    :progress_callback_user_data,
-    :abort_callback,
-    :abort_callback_user_data,
-    :vad,
-    :vad_model_path,
-    :vad_params,
-  ]
-
-  def setup
-    @params  = Whisper::Params.new
-  end
-
-  def test_language
-    @params.language = "en"
-    assert_equal @params.language, "en"
-    @params.language = "auto"
-    assert_equal @params.language, "auto"
-  end
-
-  def test_offset
-    @params.offset = 10_000
-    assert_equal @params.offset, 10_000
-    @params.offset = 0
-    assert_equal @params.offset, 0
-  end
-
-  def test_duration
-    @params.duration = 60_000
-    assert_equal @params.duration, 60_000
-    @params.duration = 0
-    assert_equal @params.duration, 0
-  end
-
-  def test_max_text_tokens
-    @params.max_text_tokens = 300
-    assert_equal @params.max_text_tokens, 300
-    @params.max_text_tokens = 0
-    assert_equal @params.max_text_tokens, 0
-  end
-
-  def test_translate
-    @params.translate = true
-    assert @params.translate
-    @params.translate = false
-    assert !@params.translate
-  end
-
-  def test_no_context
-    @params.no_context = true
-    assert @params.no_context
-    @params.no_context = false
-    assert !@params.no_context
-  end
-
-  def test_single_segment
-    @params.single_segment = true
-    assert @params.single_segment
-    @params.single_segment = false
-    assert !@params.single_segment
-  end
-
-  def test_print_special
-    @params.print_special = true
-    assert @params.print_special
-    @params.print_special = false
-    assert !@params.print_special
-  end
-
-  def test_print_progress
-    @params.print_progress = true
-    assert @params.print_progress
-    @params.print_progress = false
-    assert !@params.print_progress
-  end
-
-  def test_print_realtime
-    @params.print_realtime = true
-    assert @params.print_realtime
-    @params.print_realtime = false
-    assert !@params.print_realtime
-  end
-
-  def test_print_timestamps
-    @params.print_timestamps = true
-    assert @params.print_timestamps
-    @params.print_timestamps = false
-    assert !@params.print_timestamps
-  end
-
-  def test_suppress_blank
-    @params.suppress_blank = true
-    assert @params.suppress_blank
-    @params.suppress_blank = false
-    assert !@params.suppress_blank
-  end
-
-  def test_suppress_nst
-    @params.suppress_nst = true
-    assert @params.suppress_nst
-    @params.suppress_nst = false
-    assert !@params.suppress_nst
-  end
-
-  def test_token_timestamps
-    @params.token_timestamps = true
-    assert @params.token_timestamps
-    @params.token_timestamps = false
-    assert !@params.token_timestamps
-  end
-
-  def test_split_on_word
-    @params.split_on_word = true
-    assert @params.split_on_word
-    @params.split_on_word = false
-    assert !@params.split_on_word
-  end
-
-  def test_initial_prompt
-    assert_nil @params.initial_prompt
-    @params.initial_prompt = "You are a polite person."
-    assert_equal "You are a polite person.", @params.initial_prompt
-  end
-
-  def test_temperature
-    assert_equal 0.0, @params.temperature
-    @params.temperature = 0.5
-    assert_equal 0.5, @params.temperature
-  end
-
-  def test_max_initial_ts
-    assert_equal 1.0, @params.max_initial_ts
-    @params.max_initial_ts = 600.0
-    assert_equal 600.0, @params.max_initial_ts
-  end
-
-  def test_length_penalty
-    assert_equal(-1.0, @params.length_penalty)
-    @params.length_penalty = 0.5
-    assert_equal 0.5, @params.length_penalty
-  end
-
-  def test_temperature_inc
-    assert_in_delta 0.2, @params.temperature_inc
-    @params.temperature_inc = 0.5
-    assert_in_delta 0.5, @params.temperature_inc
-  end
-
-  def test_entropy_thold
-    assert_in_delta 2.4, @params.entropy_thold
-    @params.entropy_thold = 3.0
-    assert_in_delta 3.0, @params.entropy_thold
-  end
-
-  def test_logprob_thold
-    assert_in_delta(-1.0, @params.logprob_thold)
-    @params.logprob_thold = -0.5
-    assert_in_delta(-0.5, @params.logprob_thold)
-  end
-
-  def test_no_speech_thold
-    assert_in_delta 0.6, @params.no_speech_thold
-    @params.no_speech_thold = 0.2
-    assert_in_delta 0.2, @params.no_speech_thold
-  end
-
-  def test_vad
-    assert_false @params.vad
-    @params.vad = true
-    assert_true @params.vad
-  end
-
-  def test_vad_model_path
-    assert_nil @params.vad_model_path
-    @params.vad_model_path = "silero-v5.1.2"
-    assert_equal Whisper::Model.pre_converted_models["silero-v5.1.2"].to_path, @params.vad_model_path
-  end
-
-  def test_vad_model_path_with_nil
-    @params.vad_model_path = "silero-v5.1.2"
-    @params.vad_model_path = nil
-    assert_nil @params.vad_model_path
-  end
-
-  def test_vad_model_path_with_invalid
-    assert_raise TypeError do
-      @params.vad_model_path = Object.new
-    end
-  end
-
-  def test_vad_model_path_with_URI_string
-    @params.vad_model_path = "https://huggingface.co/ggml-org/whisper-vad/resolve/main/ggml-silero-v5.1.2.bin"
-    assert_equal @params.vad_model_path, Whisper::Model.pre_converted_models["silero-v5.1.2"].to_path
-  end
-
-  def test_vad_model_path_with_URI
-    @params.vad_model_path = URI("https://huggingface.co/ggml-org/whisper-vad/resolve/main/ggml-silero-v5.1.2.bin")
-    assert_equal @params.vad_model_path, Whisper::Model.pre_converted_models["silero-v5.1.2"].to_path
-  end
-
-  def test_vad_params
-    assert_kind_of Whisper::VAD::Params, @params.vad_params
-    default_params = @params.vad_params
-    assert_same default_params, @params.vad_params
-    assert_equal 0.5, default_params.threshold
-    new_params = Whisper::VAD::Params.new
-    @params.vad_params = new_params
-    assert_same new_params, @params.vad_params
-  end
-
-  def test_new_with_kw_args
-    params = Whisper::Params.new(language: "es")
-    assert_equal "es", params.language
-    assert_equal 1.0, params.max_initial_ts
-  end
-
-  def test_new_with_kw_args_non_existent
-    assert_raise ArgumentError do
-      Whisper::Params.new(non_existent: "value")
-    end
-  end
-
-  def test_new_with_kw_args_wrong_type
-    assert_raise TypeError do
-      Whisper::Params.new(language: 3)
-    end
-  end
-
-  data(PARAM_NAMES.collect {|param| [param, param]}.to_h)
-  def test_new_with_kw_args_default_values(param)
-    default_value = @params.send(param)
-    value = case [param, default_value]
-            in [*, true | false]
-              !default_value
-            in [*, Integer | Float]
-              default_value + 1
-            in [:language, *]
-              "es"
-            in [:initial_prompt, *]
-              "Initial prompt"
-            in [/_callback\Z/, *]
-              proc {}
-            in [/_user_data\Z/, *]
-              Object.new
-            in [:vad_model_path, *]
-              Whisper::Model.pre_converted_models["silero-v5.1.2"].to_path
-            in [:vad_params, *]
-              Whisper::VAD::Params.new
-            end
-    params = Whisper::Params.new(param => value)
-    if Float === value
-      assert_in_delta value, params.send(param)
-    else
-      assert_equal value, params.send(param)
-    end
-
-    PARAM_NAMES.reject {|name| name == param}.each do |name|
-      expected = @params.send(name)
-      actual = params.send(name)
-      if Float === expected
-        assert_in_delta expected, actual
-      else
-        assert_equal expected, actual
-      end
-    end
-  end
-end
--- a/bindings/ruby/test/test_segment.rb
+++ b/bindings/ruby/test/test_segment.rb
@ -1,136 +0,0 @@
-require_relative "helper"
-
-class TestSegment < TestBase
-  def test_iteration
-    whisper.each_segment do |segment|
-      assert_instance_of Whisper::Segment, segment
-    end
-  end
-
-  def test_enumerator
-    enum = whisper.each_segment
-    assert_instance_of Enumerator, enum
-    enum.to_a.each_with_index do |segment, index|
-      assert_instance_of Whisper::Segment, segment
-      assert_kind_of Integer, index
-    end
-  end
-
-  def test_start_time
-    i = 0
-    whisper.each_segment do |segment|
-      assert_equal 0, segment.start_time if i == 0
-      i += 1
-    end
-  end
-
-  def test_end_time
-    i = 0
-    whisper.each_segment do |segment|
-      assert_equal whisper.full_get_segment_t1(i) * 10, segment.end_time
-      i += 1
-    end
-  end
-
-  def test_no_speech_prob
-    no_speech_prob = nil
-    whisper.each_segment do |segment|
-      no_speech_prob = segment.no_speech_prob
-    end
-    assert no_speech_prob > 0.0
-  end
-
-  def test_on_new_segment
-    params = Whisper::Params.new
-    seg = nil
-    index = 0
-    params.on_new_segment do |segment|
-      assert_instance_of Whisper::Segment, segment
-      if index == 0
-        seg = segment
-        assert_equal 0, segment.start_time
-        assert_match(/ask not what your country can do for you, ask what you can do for your country/, segment.text)
-      end
-      index += 1
-    end
-    whisper.transcribe(AUDIO, params)
-    assert_equal 0, seg.start_time
-    assert_match(/ask not what your country can do for you, ask what you can do for your country/, seg.text)
-  end
-
-  def test_on_new_segment_twice
-    params = Whisper::Params.new
-    seg = nil
-    params.on_new_segment do |segment|
-      seg = segment
-      return
-    end
-    params.on_new_segment do |segment|
-      assert_same seg, segment
-      return
-    end
-    whisper.transcribe(AUDIO, params)
-  end
-
-  def test_pattern_matching
-    segment = whisper.each_segment.first
-    segment => {start_time:, end_time:, text:, no_speech_prob:, speaker_turn_next:}
-
-    assert_equal segment.start_time, start_time
-    assert_equal segment.end_time, end_time
-    assert_equal segment.text, text
-    assert_equal segment.no_speech_prob, no_speech_prob
-    assert_equal segment.speaker_turn_next?, speaker_turn_next
-  end
-
-  def test_pattern_matching_partial
-    segment = whisper.each_segment.first
-    segment => {start_time:, end_time:, text:}
-
-    assert_equal segment.start_time, start_time
-    assert_equal segment.end_time, end_time
-    assert_equal segment.text, text
-  end
-
-  def test_deconstruct_keys
-    segment = whisper.each_segment.first
-    expected = {
-      start_time: segment.start_time,
-      end_time: segment.end_time,
-      text: segment.text,
-      no_speech_prob: segment.no_speech_prob,
-      speaker_turn_next: segment.speaker_turn_next?
-    }
-    assert_equal expected, segment.deconstruct_keys([:start_time, :end_time, :text, :no_speech_prob, :speaker_turn_next])
-  end
-
-  def test_deconstruct_keys_non_existent
-    omit "Undefined behavior"
-
-    segment = whisper.each_segment.first
-
-    assert_equal({}, segment.deconstruct_keys([:non_existent]))
-  end
-
-  def test_deconstruct_keys_too_many_keys
-    omit "Undefined behavior"
-
-    segment = whisper.each_segment.first
-
-    assert_equal({}, segment.deconstruct_keys([:start_time, :end_time, :text, :no_speech_prob, :speaker_turn_next, :extra_key]))
-  end
-
-  def test_deconstruct_keys_includes_non_existent_keys_not_too_many
-    omit "Undefined behavior"
-
-    segment = whisper.each_segment.first
-
-    expected = {
-      start_time: segment.start_time,
-      end_time: segment.end_time,
-      text: segment.text,
-      no_speech_prob: segment.no_speech_prob
-    }
-    assert_equal(expected, segment.deconstruct_keys([:start_time, :end_time, :text, :no_speech_prob, :non_existent]))
-  end
-end
--- a/bindings/ruby/test/test_vad.rb
+++ b/bindings/ruby/test/test_vad.rb
@ -1,19 +0,0 @@
-require_relative "helper"
-
-class TestVAD < TestBase
-  def setup
-    @whisper = Whisper::Context.new("base.en")
-    vad_params = Whisper::VAD::Params.new
-    @params = Whisper::Params.new(
-      vad: true,
-      vad_model_path: "silero-v5.1.2",
-      vad_params:
-    )
-  end
-
-  def test_transcribe
-    @whisper.transcribe(TestBase::AUDIO, @params) do |text|
-      assert_match(/ask not what your country can do for you[,.] ask what you can do for your country/i, text)
-    end
-  end
-end
--- a/bindings/ruby/test/test_vad_params.rb
+++ b/bindings/ruby/test/test_vad_params.rb
@ -1,103 +0,0 @@
-require_relative "helper"
-
-class TestVADParams < TestBase
-  PARAM_NAMES = [
-    :threshold,
-    :min_speech_duration_ms,
-    :min_silence_duration_ms,
-    :max_speech_duration_s,
-    :speech_pad_ms,
-    :samples_overlap
-  ]
-
-  def setup
-    @params = Whisper::VAD::Params.new
-  end
-
-  def test_new
-    params = Whisper::VAD::Params.new
-    assert_kind_of Whisper::VAD::Params, params
-  end
-
-  def test_threshold
-    assert_in_delta @params.threshold, 0.5
-    @params.threshold = 0.7
-    assert_in_delta @params.threshold, 0.7
-  end
-
-  def test_min_speech_duration
-    pend
-  end
-
-  def test_min_speech_duration_ms
-    assert_equal 250, @params.min_speech_duration_ms
-    @params.min_speech_duration_ms = 500
-    assert_equal 500, @params.min_speech_duration_ms
-  end
-
-  def test_min_silence_duration_ms
-    assert_equal 100, @params.min_silence_duration_ms
-    @params.min_silence_duration_ms = 200
-    assert_equal 200, @params.min_silence_duration_ms
-  end
-
-  def test_max_speech_duration
-    pend
-  end
-
-  def test_max_speech_duration_s
-    assert @params.max_speech_duration_s >= 10e37 # Defaults to FLT_MAX
-    @params.max_speech_duration_s = 60.0
-    assert_equal 60.0, @params.max_speech_duration_s
-  end
-
-  def test_speech_pad_ms
-    assert_equal 30, @params.speech_pad_ms
-    @params.speech_pad_ms = 50
-    assert_equal 50, @params.speech_pad_ms
-  end
-
-  def test_samples_overlap
-    assert_in_delta @params.samples_overlap, 0.1
-    @params.samples_overlap = 0.5
-    assert_in_delta @params.samples_overlap, 0.5
-  end
-
-  def test_equal
-    assert_equal @params, Whisper::VAD::Params.new
-  end
-
-  def test_new_with_kw_args
-    params = Whisper::VAD::Params.new(threshold: 0.7)
-    assert_in_delta params.threshold, 0.7
-    assert_equal 250, params.min_speech_duration_ms
-  end
-
-  def test_new_with_kw_args_non_existent
-    assert_raise ArgumentError do
-      Whisper::VAD::Params.new(non_existent: "value")
-    end
-  end
-
-  data(PARAM_NAMES.collect {|param| [param, param]}.to_h)
-  def test_new_with_kw_args_default_values(param)
-    default_value = @params.send(param)
-    value = default_value + 1
-    params = Whisper::VAD::Params.new(param => value)
-    if Float === value
-      assert_in_delta value, params.send(param)
-    else
-      assert_equal value, params.send(param)
-    end
-
-    PARAM_NAMES.reject {|name| name == param}.each do |name|
-      expected = @params.send(name)
-      actual = params.send(name)
-      if Float === expected
-        assert_in_delta expected, actual
-      else
-        assert_equal expected, actual
-      end
-    end
-  end
-end
--- a/bindings/ruby/test/test_whisper.rb
+++ b/bindings/ruby/test/test_whisper.rb
@ -1,292 +0,0 @@
-require_relative "helper"
-require "stringio"
-require "etc"
-
-# Exists to detect memory-related bug
-Whisper.log_set ->(level, buffer, user_data) {}, nil
-
-class TestWhisper < TestBase
-  def setup
-    @params  = Whisper::Params.new
-  end
-
-  def test_whisper
-    @whisper = Whisper::Context.new("base.en")
-    params  = Whisper::Params.new
-    params.print_timestamps = false
-
-    @whisper.transcribe(AUDIO, params) {|text|
-      assert_match(/ask not what your country can do for you, ask what you can do for your country/, text)
-    }
-  end
-
-  def test_transcribe_non_parallel
-    @whisper = Whisper::Context.new("base.en")
-    params  = Whisper::Params.new
-
-    @whisper.transcribe(AUDIO, params, n_processors: 1) {|text|
-      assert_match(/ask not what your country can do for you, ask what you can do for your country/, text)
-    }
-  end
-
-  def test_transcribe_n_processors
-    @whisper = Whisper::Context.new("base.en")
-    params  = Whisper::Params.new
-
-    @whisper.transcribe(AUDIO, params, n_processors: 4) {|text|
-      assert_match(/ask not what your country can do for you[,.] ask what you can do for your country/i, text)
-    }
-  end
-
-  sub_test_case "After transcription" do
-    def test_full_n_segments
-      assert_equal 1, whisper.full_n_segments
-    end
-
-    def test_full_lang_id
-      assert_equal 0, whisper.full_lang_id
-    end
-
-    def test_full_get_segment
-      segment = whisper.full_get_segment(0)
-      assert_equal 0, segment.start_time
-      assert_match(/ask not what your country can do for you, ask what you can do for your country/, segment.text)
-    end
-
-    def test_full_get_segment_t0
-      assert_equal 0, whisper.full_get_segment_t0(0)
-      assert_raise IndexError do
-        whisper.full_get_segment_t0(whisper.full_n_segments)
-      end
-      assert_raise IndexError do
-        whisper.full_get_segment_t0(-1)
-      end
-    end
-
-    def test_full_get_segment_t1
-      t1 = whisper.full_get_segment_t1(0)
-      assert_kind_of Integer, t1
-      assert t1 > 0
-      assert_raise IndexError do
-        whisper.full_get_segment_t1(whisper.full_n_segments)
-      end
-    end
-
-    def test_full_get_segment_speaker_turn_next
-      assert_false whisper.full_get_segment_speaker_turn_next(0)
-    end
-
-    def test_full_get_segment_text
-      assert_match(/ask not what your country can do for you, ask what you can do for your country/, whisper.full_get_segment_text(0))
-    end
-
-    def test_full_get_segment_no_speech_prob
-      prob = whisper.full_get_segment_no_speech_prob(0)
-      assert prob > 0.0
-      assert prob < 1.0
-    end
-  end
-
-  def test_lang_max_id
-    assert_kind_of Integer, Whisper.lang_max_id
-  end
-
-  def test_lang_id
-    assert_equal 0, Whisper.lang_id("en")
-    assert_raise ArgumentError do
-      Whisper.lang_id("non existing language")
-    end
-  end
-
-  def test_lang_str
-    assert_equal "en", Whisper.lang_str(0)
-    assert_raise IndexError do
-      Whisper.lang_str(Whisper.lang_max_id + 1)
-    end
-  end
-
-  def test_lang_str_full
-    assert_equal "english", Whisper.lang_str_full(0)
-    assert_raise IndexError do
-      Whisper.lang_str_full(Whisper.lang_max_id + 1)
-    end
-  end
-
-  def test_system_info_str
-    assert_match(/\AWHISPER : COREML = \d | OPENVINO = \d |/, Whisper.system_info_str)
-  end
-
-  def test_log_set
-    user_data = Object.new
-    logs = []
-    log_callback = ->(level, buffer, udata) {
-      logs << [level, buffer, udata]
-    }
-    Whisper.log_set log_callback, user_data
-    Whisper::Context.new("base.en")
-
-    assert logs.length > 30
-    logs.each do |log|
-      assert_include [Whisper::LOG_LEVEL_DEBUG, Whisper::LOG_LEVEL_INFO, Whisper::LOG_LEVEL_WARN], log[0]
-      assert_same user_data, log[2]
-    end
-  end
-
-  def test_log_suppress
-    stderr = $stderr
-    Whisper.log_set ->(level, buffer, user_data) {
-      # do nothing
-    }, nil
-    dev = StringIO.new("")
-    $stderr = dev
-    Whisper::Context.new("base.en")
-    assert_empty dev.string
-  ensure
-    $stderr = stderr
-  end
-
-  sub_test_case "full" do
-    def setup
-      super
-      @whisper = Whisper::Context.new("base.en")
-      @samples = File.read(AUDIO, nil, 78).unpack("s<*").collect {|i| i.to_f / 2**15}
-    end
-
-    def test_full
-      @whisper.full(@params, @samples, @samples.length)
-
-      assert_equal 1, @whisper.full_n_segments
-      assert_match(/ask not what your country can do for you, ask what you can do for your country/, @whisper.each_segment.first.text)
-    end
-
-    def test_full_without_length
-      @whisper.full(@params, @samples)
-
-      assert_equal 1, @whisper.full_n_segments
-      assert_match(/ask not what your country can do for you, ask what you can do for your country/, @whisper.each_segment.first.text)
-    end
-
-    def test_full_enumerator
-      samples = @samples.each
-      @whisper.full(@params, samples, @samples.length)
-
-      assert_equal 1, @whisper.full_n_segments
-      assert_match(/ask not what your country can do for you, ask what you can do for your country/, @whisper.each_segment.first.text)
-    end
-
-    def test_full_enumerator_without_length
-      samples = @samples.each
-      assert_raise ArgumentError do
-        @whisper.full(@params, samples)
-      end
-    end
-
-    def test_full_enumerator_with_too_large_length
-      samples = @samples.each.take(10).to_enum
-      assert_raise StopIteration do
-        @whisper.full(@params, samples, 11)
-      end
-    end
-
-    def test_full_with_memory_view
-      samples = JFKReader.new(AUDIO)
-      @whisper.full(@params, samples)
-
-      assert_equal 1, @whisper.full_n_segments
-      assert_match(/ask not what your country can do for you, ask what you can do for your country/, @whisper.each_segment.first.text)
-    end
-
-    def test_full_parallel
-      nprocessors = 2
-      @whisper.full_parallel(@params, @samples, @samples.length, nprocessors)
-
-      assert_equal nprocessors, @whisper.full_n_segments
-      text = @whisper.each_segment.collect(&:text).join
-      assert_match(/ask what you can do/i, text)
-      assert_match(/for your country/i, text)
-    end
-
-    def test_full_parallel_with_memory_view
-      nprocessors = 2
-      samples = JFKReader.new(AUDIO)
-      @whisper.full_parallel(@params, samples, nil, nprocessors)
-
-      assert_equal nprocessors, @whisper.full_n_segments
-      text = @whisper.each_segment.collect(&:text).join
-      assert_match(/ask what you can do/i, text)
-      assert_match(/for your country/i, text)
-    end
-
-    def test_full_parallel_without_length_and_n_processors
-      @whisper.full_parallel(@params, @samples)
-
-      assert_equal 1, @whisper.full_n_segments
-      text = @whisper.each_segment.collect(&:text).join
-      assert_match(/ask what you can do/i, text)
-      assert_match(/for your country/i, text)
-    end
-
-    def test_full_parallel_without_length
-      nprocessors = 2
-      @whisper.full_parallel(@params, @samples, nil, nprocessors)
-
-      assert_equal nprocessors, @whisper.full_n_segments
-      text = @whisper.each_segment.collect(&:text).join
-      assert_match(/ask what you can do/i, text)
-      assert_match(/for your country/i, text)
-    end
-
-    def test_full_parallel_without_n_processors
-      @whisper.full_parallel(@params, @samples, @samples.length)
-
-      assert_equal 1, @whisper.full_n_segments
-      text = @whisper.each_segment.collect(&:text).join
-      assert_match(/ask what you can do/i, text)
-      assert_match(/for your country/i, text)
-    end
-  end
-
-  def test_to_srt
-    whisper = Whisper::Context.new("base.en")
-    whisper.transcribe AUDIO, @params
-
-    lines = whisper.to_srt.lines
-    assert_match(/\A\d+\n/, lines[0])
-    assert_match(/\d{2}:\d{2}:\d{2},\d{3} --> \d{2}:\d{2}:\d{2},\d{3}\n/, lines[1])
-    assert_match(/ask not what your country can do for you, ask what you can do for your country/, lines[2])
-  end
-
-  def test_to_webvtt
-    whisper = Whisper::Context.new("base.en")
-    whisper.transcribe AUDIO, @params
-
-    lines = whisper.to_webvtt.lines
-    assert_equal "WEBVTT\n", lines[0]
-    assert_equal "\n", lines[1]
-    assert_match(/\A\d+\n/, lines[2])
-    assert_match(/\d{2}:\d{2}:\d{2}\.\d{3} --> \d{2}:\d{2}:\d{2}\.\d{3}\n/, lines[3])
-    assert_match(/ask not what your country can do for you, ask what you can do for your country/, lines[4])
-  end
-
-  sub_test_case "Format needs escape" do
-    def setup
-      @whisper = Whisper::Context.new("base.en")
-      @whisper.transcribe AUDIO, Whisper::Params.new
-      segment = @whisper.each_segment.first
-      segment.define_singleton_method :text do
-        "& so my fellow Americans --> ask not what your country can do for you <-- ask what you can do for your country."
-      end
-      @whisper.define_singleton_method :each_segment do
-        Enumerator.new(3) {|yielder| 3.times {yielder << segment}}
-      end
-    end
-
-    def test_to_srt_escape
-      assert_equal "&amp; so my fellow Americans --&gt; ask not what your country can do for you &lt;-- ask what you can do for your country.\n", @whisper.to_srt.lines[2]
-    end
-
-    def test_to_webvtt_escape
-      assert_equal "&amp; so my fellow Americans --&gt; ask not what your country can do for you &lt;-- ask what you can do for your country.\n", @whisper.to_webvtt.lines[4]
-    end
-  end
-end
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Georgi Gerganov	13c5446759	Update ggml-cuda/mmvq.cu Co-authored-by: Johannes Gäßler <johannesg@5d6.de>	2024-06-11 17:37:32 +03:00
Georgi Gerganov	9df6298a91	cuda : fix bounds check for src0 rows in MMVQ kernel	2024-06-11 11:30:12 +03:00