release : v1.7.4

ggml-ci

.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
+    apt-get install -y build-essential git cmake libsdl2-dev wget
 
 WORKDIR /app
 
@@ -23,6 +23,6 @@ ENV CUDA_DOCKER_ARCH=${CUDA_DOCKER_ARCH}
 # Enable cuBLAS
 ENV GGML_CUDA=1
 
-RUN make
+RUN make base.en
 
 ENTRYPOINT ["/app/main"]

.devops/main-cuda.Dockerfile

@@ -17,7 +17,7 @@ ENV CUDA_DOCKER_ARCH=${CUDA_DOCKER_ARCH}
 ENV GGML_CUDA=1
 
 RUN apt-get update && \
-    apt-get install -y build-essential libsdl2-dev \
+    apt-get install -y build-essential libsdl2-dev wget cmake \
     && rm -rf /var/lib/apt/lists/* /var/cache/apt/archives/*
 
 # Ref: https://stackoverflow.com/a/53464012
@@ -25,7 +25,7 @@ ENV CUDA_MAIN_VERSION=12.3
 ENV LD_LIBRARY_PATH /usr/local/cuda-${CUDA_MAIN_VERSION}/compat:$LD_LIBRARY_PATH
 
 COPY .. .
-RUN make
+RUN make base.en
 
 FROM ${BASE_CUDA_RUN_CONTAINER} AS runtime
 ENV CUDA_MAIN_VERSION=12.3
@@ -33,7 +33,7 @@ ENV LD_LIBRARY_PATH /usr/local/cuda-${CUDA_MAIN_VERSION}/compat:$LD_LIBRARY_PATH
 WORKDIR /app
 
 RUN apt-get update && \
-  apt-get install -y curl ffmpeg \
+  apt-get install -y curl ffmpeg wget cmake \
   && rm -rf /var/lib/apt/lists/* /var/cache/apt/archives/*
 
 COPY --from=build /app /app

.devops/main.Dockerfile

@@ -2,17 +2,17 @@ FROM ubuntu:22.04 AS build
 WORKDIR /app
 
 RUN apt-get update && \
-  apt-get install -y build-essential \
+  apt-get install -y build-essential wget cmake \
   && rm -rf /var/lib/apt/lists/* /var/cache/apt/archives/*
 
 COPY .. .
-RUN make
+RUN make base.en
 
 FROM ubuntu:22.04 AS runtime
 WORKDIR /app
 
 RUN apt-get update && \
-  apt-get install -y curl ffmpeg libsdl2-dev \
+  apt-get install -y curl ffmpeg libsdl2-dev wget cmake \
   && rm -rf /var/lib/apt/lists/* /var/cache/apt/archives/*
 
 COPY --from=build /app /app

.github/workflows/bindings-ruby.yml

@@ -0,0 +1,55 @@
+name: Bindings Tests (Ruby)
+on:
+  push:
+    paths:
+      - bindings/ruby/**
+      - src/**/*.c
+      - src/**/*.cpp
+      - src/**/*.h
+      - src/**/*.m
+      - src/**/*.metal
+      - include/**/*.c
+      - include/**/*.cpp
+      - include/**/*.h
+      - include/**/*.m
+      - include/**/*.metal
+      - ggml/**/*.c
+      - ggml/**/*.cpp
+      - ggml/**/*.h
+      - ggml/**/*.m
+      - ggml/**/*.metal
+      - scripts/get-flags.mk
+      - examples/dr_wav.h
+  pull_request:
+    paths:
+      - bindings/ruby/**
+      - src/**/*.c
+      - src/**/*.cpp
+      - src/**/*.h
+      - src/**/*.m
+      - src/**/*.metal
+      - include/**/*.c
+      - include/**/*.cpp
+      - include/**/*.h
+      - include/**/*.m
+      - include/**/*.metal
+      - ggml/**/*.c
+      - ggml/**/*.cpp
+      - ggml/**/*.h
+      - ggml/**/*.m
+      - ggml/**/*.metal
+      - scripts/get-flags.mk
+      - examples/dr_wav.h
+
+jobs:
+  ubuntu-latest:
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: bindings/ruby
+    steps:
+      - uses: ruby/setup-ruby@v1
+        with:
+          ruby-version: '3.1'
+      - uses: actions/checkout@v4
+      - run: rake test

.github/workflows/bindings-ruby.yml.disabled

@@ -1,23 +0,0 @@
-# TODO: fix this workflow file, disabled for now
-name: Bindings Tests (Ruby)
-on:
-  push:
-    paths:
-      - bindings/ruby/**
-      - whisper.h
-  pull_request:
-    paths:
-      - bindings/ruby/**
-      - whisper.h
-
-jobs:
-  ubuntu-latest:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: ruby/setup-ruby@v1
-        with:
-          ruby-version: '3.0'
-      - uses: actions/checkout@v1
-      - run: |
-          cd bindings/ruby/ext
-          ruby extconf.rb && make

.github/workflows/build.yml

@@ -1,8 +1,19 @@
 name: CI
-on: [push, pull_request]
+
+on:
+  push:
+    branches:
+      - master
+  pull_request:
+    types: [opened, synchronize, reopened]
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
+  cancel-in-progress: true
 
 env:
   ubuntu_image: "ubuntu:22.04"
+  VCPKG_BINARY_SOURCES: "clear;x-gha,readwrite"
 
 jobs:
   ubuntu-latest:
@@ -11,7 +22,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        arch: [linux/amd64, linux/arm64, linux/arm/v7, linux/ppc64le]
+        arch: [linux/amd64, linux/ppc64le]
 
     steps:
       - name: Clone
@@ -27,9 +38,61 @@ jobs:
             -w /workspace ${{ env.ubuntu_image }} /bin/sh -c '
             set -e
             apt update
-            apt install -y build-essential libsdl2-dev
-            make
-            make stream'
+            apt install -y build-essential libsdl2-dev cmake
+            cmake -B build
+            cmake --build build --config Release -j $(nproc)'
+
+  ubuntu-latest-arm64:
+    runs-on: ubuntu-latest
+
+    strategy:
+      fail-fast: false
+      matrix:
+        arch: [linux/arm64]
+
+    steps:
+      - name: Clone
+        uses: actions/checkout@v4
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
+      - name: Build ${{ matrix.arch }}
+        run: |
+          docker run --platform ${{ matrix.arch }} --rm \
+            -v ${{ github.workspace }}:/workspace \
+            -w /workspace ${{ env.ubuntu_image }} /bin/sh -c '
+            set -e
+            apt update
+            apt install -y build-essential libsdl2-dev cmake
+            cmake -B build -DGGML_NATIVE=OFF -DGGML_CPU_ARM_ARCH=armv8-a
+            cmake --build build --config Release -j $(nproc)'
+
+  ubuntu-latest-arm-v7:
+    runs-on: ubuntu-latest
+
+    strategy:
+      fail-fast: false
+      matrix:
+        arch: [linux/arm/v7]
+
+    steps:
+      - name: Clone
+        uses: actions/checkout@v4
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
+      - name: Build ${{ matrix.arch }}
+        run: |
+          docker run --platform ${{ matrix.arch }} --rm \
+            -v ${{ github.workspace }}:/workspace \
+            -w /workspace ${{ env.ubuntu_image }} /bin/sh -c '
+            set -e
+            apt update
+            apt install -y build-essential libsdl2-dev cmake
+            cmake -B build -DGGML_NATIVE=OFF -DGGML_CPU_ARM_ARCH=armv7-a+fp
+            cmake --build build --config Release -j $(nproc)'
 
   macOS-latest:
     runs-on: macOS-latest
@@ -41,30 +104,30 @@ jobs:
       - name: Dependencies
         run: |
           brew update
-          brew install sdl2
+          brew install sdl2 cmake
 
       - name: Build
         run: |
-          make
-          make stream
+          cmake -B build
+          cmake --build build --config Release
 
-  freeBSD-latest:
-    runs-on: macos-12
-
-    steps:
-      - name: Clone
-        uses: actions/checkout@v4
-
-      - name: Build
-        uses: cross-platform-actions/action@v0.24.0
-        with:
-          operating_system: freebsd
-          version: '13.3'
-          run: |
-            sudo pkg update
-            sudo pkg install -y gmake sdl2
-            gmake
-            gmake stream
+#  freeBSD-latest:
+#    runs-on: macos-12
+#
+#    steps:
+#      - name: Clone
+#        uses: actions/checkout@v4
+#
+#      - name: Build
+#        uses: cross-platform-actions/action@v0.24.0
+#        with:
+#          operating_system: freebsd
+#          version: '13.3'
+#          run: |
+#            sudo pkg update
+#            sudo pkg install -y gmake sdl2 cmake
+#            cmake -B build
+#            cmake --build build --config Release
 
   ubuntu-latest-gcc:
     runs-on: ubuntu-latest
@@ -73,7 +136,7 @@ jobs:
       fail-fast: false
       matrix:
         build: [Debug, Release]
-        arch: [linux/amd64, linux/arm64, linux/arm/v7, linux/ppc64le]
+        arch: [linux/amd64, linux/ppc64le]
 
     steps:
       - name: Clone
@@ -94,6 +157,62 @@ jobs:
             make
             ctest -L gh --output-on-failure'
 
+  ubuntu-latest-gcc-arm64:
+    runs-on: ubuntu-latest
+
+    strategy:
+      fail-fast: false
+      matrix:
+        build: [Debug, Release]
+        arch: [linux/arm64]
+
+    steps:
+      - name: Clone
+        uses: actions/checkout@v4
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
+      - name: Build ${{ matrix.arch }}
+        run: |
+          docker run --platform ${{ matrix.arch }} --rm \
+            -v ${{ github.workspace }}:/workspace \
+            -w /workspace ${{ env.ubuntu_image }} /bin/sh -c '
+            set -e
+            apt update
+            apt install -y build-essential cmake libsdl2-dev
+            cmake . -DWHISPER_SDL2=ON -DCMAKE_BUILD_TYPE=${{ matrix.build }} -DGGML_NATIVE=OFF -DGGML_CPU_ARM_ARCH=armv8-a
+            make
+            ctest -L gh --output-on-failure'
+
+  ubuntu-latest-gcc-arm-v7:
+    runs-on: ubuntu-latest
+
+    strategy:
+      fail-fast: false
+      matrix:
+        build: [Debug, Release]
+        arch: [linux/arm/v7]
+
+    steps:
+      - name: Clone
+        uses: actions/checkout@v4
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
+      - name: Build ${{ matrix.arch }}
+        run: |
+          docker run --platform ${{ matrix.arch }} --rm \
+            -v ${{ github.workspace }}:/workspace \
+            -w /workspace ${{ env.ubuntu_image }} /bin/sh -c '
+            set -e
+            apt update
+            apt install -y build-essential cmake libsdl2-dev
+            cmake . -DWHISPER_SDL2=ON -DCMAKE_BUILD_TYPE=${{ matrix.build }} -DGGML_NATIVE=OFF -DGGML_CPU_ARM_ARCH=armv7-a+fp
+            make
+            ctest -L gh --output-on-failure'
+
   ubuntu-latest-clang:
     runs-on: ubuntu-latest
 
@@ -279,25 +398,10 @@ jobs:
             mingw-w64-${{matrix.env}}-SDL2
             mingw-w64-${{matrix.env}}-openblas
 
-      - name: Build using make
-        shell: msys2 {0}
-        run: |
-            make -j $(nproc)
-
-      - name: Clean after building using make
-        shell: msys2 {0}
-        run: |
-            make clean
-
-      - name: Build using make w/ OpenBLAS
-        shell: msys2 {0}
-        run: |
-            make GGML_OPENBLAS=1 -j $(nproc)
-
       - name: Build using CMake
         shell: msys2 {0}
         run: |
-            cmake -B build
+            cmake -B build -DWHISPER_SDL2=ON
             cmake --build build --config ${{ matrix.build }} -j $(nproc)
 
       - name: Clean after building using CMake
@@ -308,7 +412,7 @@ jobs:
       - name: Build using CMake w/ OpenBLAS
         shell: msys2 {0}
         run: |
-            cmake -B build -DGGML_OPENBLAS=ON
+            cmake -B build -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS
             cmake --build build --config ${{ matrix.build }} -j $(nproc)
 
   windows:
@@ -382,10 +486,8 @@ jobs:
         sdl2: [ON]
         include:
           - arch: Win32
-            obzip: https://github.com/OpenMathLib/OpenBLAS/releases/download/v0.3.25/OpenBLAS-0.3.25-x86.zip
             s2arc: x86
           - arch: x64
-            obzip: https://github.com/OpenMathLib/OpenBLAS/releases/download/v0.3.25/OpenBLAS-0.3.25-x64.zip
             s2arc: x64
           - sdl2: ON
             s2ver: 2.28.5
@@ -394,17 +496,21 @@ jobs:
       - name: Clone
         uses: actions/checkout@v4
 
+      - name: Export GitHub Actions cache environment variables
+        uses: actions/github-script@v7
+        with:
+          script: |
+            core.exportVariable('ACTIONS_CACHE_URL', process.env.ACTIONS_CACHE_URL || '');
+            core.exportVariable('ACTIONS_RUNTIME_TOKEN', process.env.ACTIONS_RUNTIME_TOKEN || '');
+
       - name: Add msbuild to PATH
         uses: microsoft/setup-msbuild@v2
 
-      - name: Fetch OpenBLAS
+      - name: Install OpenBLAS and pkgconfiglite
         if: matrix.blas == 'ON'
         run: |
-          C:/msys64/usr/bin/wget.exe -qO blas.zip ${{ matrix.obzip }}
-          7z x blas.zip -oblas -y
-          copy blas/include/cblas.h .
-          copy blas/include/openblas_config.h .
-          echo "OPENBLAS_PATH=$env:GITHUB_WORKSPACE/blas" >> $env:GITHUB_ENV
+          vcpkg install --triplet=${{ matrix.s2arc }}-windows openblas
+          choco install pkgconfiglite
 
       - name: Fetch SDL2 and set SDL2_DIR
         if: matrix.sdl2 == 'ON'
@@ -416,9 +522,10 @@ jobs:
       - name: Configure
         run: >
           cmake -S . -B ./build -A ${{ matrix.arch }}
+          -DCMAKE_TOOLCHAIN_FILE="$env:VCPKG_INSTALLATION_ROOT/scripts/buildsystems/vcpkg.cmake"
           -DCMAKE_BUILD_TYPE=${{ matrix.build }}
-          -DGGML_OPENBLAS=${{ matrix.blas }}
-          -DCMAKE_LIBRARY_PATH="$env:OPENBLAS_PATH/lib"
+          -DGGML_BLAS=${{ matrix.blas }}
+          -DGGML_BLAS_VENDOR=OpenBLAS
           -DWHISPER_SDL2=${{ matrix.sdl2 }}
 
       - name: Build
@@ -426,9 +533,9 @@ jobs:
           cd ./build
           msbuild ALL_BUILD.vcxproj -t:build -p:configuration=${{ matrix.build }} -p:platform=${{ matrix.arch }}
 
-      - name: Copy libopenblas.dll
+      - name: Copy openblas.dll
         if: matrix.blas == 'ON'
-        run: copy "$env:OPENBLAS_PATH/bin/libopenblas.dll" build/bin/${{ matrix.build }}
+        run: copy "C:/vcpkg/packages/openblas_${{ matrix.s2arc }}-windows/bin/openblas.dll" build/bin/${{ matrix.build }}
 
       - name: Copy SDL2.dll
         if: matrix.sdl2 == 'ON'
@@ -443,7 +550,6 @@ jobs:
 
   windows-cublas:
     runs-on: windows-2019
-
     strategy:
       matrix:
         build: [Release]
@@ -453,12 +559,10 @@ jobs:
         cuda-toolkit: [12.2.0, 11.8.0]
         include:
           - arch: x64
-            s2arc: x64
-          - sdl2: ON
-            s2ver: 2.28.5
-
+            sdl2: ON
+            sdl2_ver: 2.28.5
     steps:
-      - name: Clone
+      - name: Clone repository
         uses: actions/checkout@v4
 
       - name: Add msbuild to PATH
@@ -470,38 +574,43 @@ jobs:
         with:
           cuda: '${{ matrix.cuda-toolkit }}'
 
+      - name: Install 7-Zip
+        run: choco install 7zip -y
+
       - name: Fetch SDL2 and set SDL2_DIR
         if: matrix.sdl2 == 'ON'
         run: |
-          C:/msys64/usr/bin/wget.exe -qO sdl2.zip https://github.com/libsdl-org/SDL/releases/download/release-${{ matrix.s2ver }}/SDL2-devel-${{ matrix.s2ver }}-VC.zip
+          Invoke-WebRequest -Uri https://github.com/libsdl-org/SDL/releases/download/release-${{ matrix.sdl2_ver }}/SDL2-devel-${{ matrix.sdl2_ver }}-VC.zip -OutFile sdl2.zip
           7z x sdl2.zip
-          echo "SDL2_DIR=$env:GITHUB_WORKSPACE/SDL2-${{ matrix.s2ver }}/cmake" >> $env:GITHUB_ENV
-
-      - name: Configure
-        run: >
-          cmake -S . -B ./build -A ${{ matrix.arch }}
-          -DCMAKE_BUILD_TYPE=${{ matrix.build }}
-          -DGGML_CUDA=${{ matrix.cublas }}
-          -DWHISPER_SDL2=${{ matrix.sdl2 }}
-
-      - name: Build ${{ matrix.cuda-toolkit }}
+          echo "SDL2_DIR=${{ github.workspace }}\SDL2-${{ matrix.sdl2_ver }}\cmake" | Out-File -FilePath $env:GITHUB_ENV -Append
+          echo "${{ github.workspace }}\SDL2-${{ matrix.sdl2_ver }}\cmake" > SDL2_PATH.txt
+ 
+      - name: Configure CMake
+        shell: cmd
+        run: |
+          cmake -S . -B ./build -A ${{ matrix.arch }} ^
+            -DCMAKE_BUILD_TYPE=${{ matrix.build }} ^
+            -DGGML_CUDA=${{ matrix.cublas }} ^
+            -DCMAKE_CUDA_ARCHITECTURES=all  ^
+            -DWHISPER_SDL2=${{ matrix.sdl2 }} ^
+            -DSDL2_DIR="%SDL2_DIR%"
+  
+      - name: Build Project
+        shell: cmd
         run: |
           cd ./build
-          cmake --build . --config ${{ matrix.build }}
+          cmake --build . --config ${{ matrix.build }}          
 
       - name: Copy CUDA DLLs
-        run: >
-          Copy-Item -PassThru
-          -Path "${{ steps.cuda-toolkit.outputs.CUDA_PATH }}/bin/*.dll"
-          -Include cudart64_*,cublas64_*,cublasLt64_*
-          -Destination build/bin/${{ matrix.build }}
+        run: |
+          Get-ChildItem "${{ steps.cuda-toolkit.outputs.CUDA_PATH }}/bin/" -Filter "*.dll" | 
+          Copy-Item -Destination "build/bin/${{ matrix.build }}"
 
       - name: Copy SDL2.dll
         if: matrix.sdl2 == 'ON'
-        run: copy "$env:SDL2_DIR/../lib/${{ matrix.s2arc }}/SDL2.dll" build/bin/${{ matrix.build }}
+        run: copy "$env:SDL2_DIR/../lib/${{ matrix.arch }}/SDL2.dll" build/bin/${{ matrix.build }}
 
       - name: Upload binaries
-        if: matrix.sdl2 == 'ON'
         uses: actions/upload-artifact@v4
         with:
           name: whisper-cublas-${{ matrix.cuda-toolkit }}-bin-${{ matrix.arch }}
@@ -529,7 +638,7 @@ jobs:
           emcmake cmake . -DCMAKE_BUILD_TYPE=${{ matrix.build }}
           make
 
-  ios:
+  ios-xcode-build:
     runs-on: macos-latest
 
     strategy:
@@ -537,7 +646,7 @@ jobs:
         build: [Release]
 
     steps:
-      - name: Clone
+      - name: Checkout code
         uses: actions/checkout@v4
 
       - name: Configure
@@ -545,11 +654,34 @@ jobs:
           cp models/for-tests-ggml-base.en.bin models/ggml-base.en.bin
           mkdir models/ggml-base.en-encoder.mlmodelc
 
-      - name: Build objc example
-        run: xcodebuild -project examples/whisper.objc/whisper.objc.xcodeproj -scheme whisper.objc -configuration ${{ matrix.build }} -sdk iphonesimulator build
+      - name: Build
+        id: cmake_build
+        run: |
+          sysctl -a
+          mkdir build
+          cd build
+          cmake -G Xcode .. \
+            -DGGML_METAL_USE_BF16=ON \
+            -DGGML_METAL_EMBED_LIBRARY=ON \
+            -DWHISPER_BUILD_EXAMPLES=OFF \
+            -DWHISPER_BUILD_TESTS=OFF \
+            -DWHISPER_BUILD_SERVER=OFF \
+            -DCMAKE_SYSTEM_NAME=iOS \
+            -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0 \
+            -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
+          cmake --build . --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
+          sudo cmake --install . --config Release
+
+      - name: xcodebuild for swift package
+        id: xcodebuild
+        run: |
+          xcodebuild -scheme whisper-Package -destination 'generic/platform=iOS'
+
+#- name: Build objc example
+#  run: xcodebuild -project examples/whisper.objc/whisper.objc.xcodeproj -scheme whisper.objc -configuration ${{ matrix.build }} -sdk iphoneos build
 
       - name: Build swiftui example
-        run: xcodebuild -project examples/whisper.swiftui/whisper.swiftui.xcodeproj -scheme WhisperCppDemo -configuration ${{ matrix.build }} -sdk iphonesimulator build
+        run: xcodebuild -project examples/whisper.swiftui/whisper.swiftui.xcodeproj -scheme WhisperCppDemo -configuration ${{ matrix.build }} -sdk iphoneos CODE_SIGNING_REQUIRED=NO CODE_SIGN_IDENTITY= -destination 'generic/platform=iOS' build
 
   android:
     runs-on: ubuntu-latest
@@ -560,12 +692,6 @@ jobs:
         with:
           path: whisper
 
-      - name: Clone
-        uses: actions/checkout@v4
-        with:
-          repository: ggerganov/ggml
-          path: ggml
-
       - name: Install Java
         uses: actions/setup-java@v4
         with:
@@ -584,7 +710,7 @@ jobs:
         run: |
           export PATH_TO_GGML=$PWD/ggml
           cd whisper/examples/whisper.android
-          ./gradlew assembleRelease --no-daemon -PGGML_HOME=$PATH_TO_GGML
+          ./gradlew assembleRelease --no-daemon
 
 # TODO: disable because of following fail: https://github.com/ggerganov/whisper.cpp/actions/runs/11019444420/job/30627193602
 #  android_java:
@@ -666,5 +792,6 @@ jobs:
       - name: Test quantize
         run: |
           ./models/download-ggml-model.sh tiny.en
-          make quantize
-          ./quantize models/ggml-tiny.en.bin models/ggml-tiny.en-q4_0.bin q4_0
+          cmake -B build
+          cmake --build build --config Release
+          ./build/bin/quantize models/ggml-tiny.en.bin models/ggml-tiny.en-q4_0.bin q4_0

.github/workflows/docker.yml

@@ -17,7 +17,7 @@ jobs:
     strategy:
       matrix:
         config:
-          - { tag: "main", dockerfile: ".devops/main.Dockerfile", platform: "linux/amd64,linux/arm64" }
+          - { tag: "main", dockerfile: ".devops/main.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" }
@@ -45,7 +45,7 @@ jobs:
         with:
           context: .
           push: true
-          platforms: ${{ matrix.config.platforms }}
+          platforms: ${{ matrix.config.platform }}
           tags: "ghcr.io/${{ github.repository }}:${{ matrix.config.tag }}-${{ env.COMMIT_SHA }}"
           file: ${{ matrix.config.dockerfile }}
 
@@ -54,6 +54,6 @@ jobs:
         with:
           context: .
           push: ${{ github.event_name == 'push' }}
-          platforms: ${{ matrix.config.platforms }}
+          platforms: ${{ matrix.config.platform }}
           tags: "ghcr.io/${{ github.repository }}:${{ matrix.config.tag }}"
           file: ${{ matrix.config.dockerfile }}

.gitignore

@@ -1,5 +1,6 @@
 *.o
 *.a
+*.d
 .cache/
 .coreml/
 .test/
@@ -19,6 +20,9 @@ build-*/
 .swiftpm
 *.metallib
 
+ggml-metal-embed.metal
+ggml-metal-embed.metal.tmp
+
 /main
 /stream
 /command

CMakeLists.txt

@@ -1,6 +1,6 @@
 cmake_minimum_required(VERSION 3.5) # for add_link_options and implicit target directories.
 project("whisper.cpp" C CXX)
-project("whisper.cpp" VERSION 1.7.1)
+project("whisper.cpp" VERSION 1.7.4)
 include(CheckIncludeFileCXX)
 
 set(SOVERSION 1)

Package.swift

@@ -14,47 +14,6 @@ let package = Package(
         .library(name: "whisper", targets: ["whisper"]),
     ],
     targets: [
-        .target(
-            name: "whisper",
-            path: ".",
-            exclude: [
-               "bindings",
-               "cmake",
-               "coreml",
-               "examples",
-               "extra",
-               "models",
-               "samples",
-               "tests",
-               "CMakeLists.txt",
-               "Makefile"
-            ],
-            sources: [
-                "ggml/src/ggml.c",
-                "src/whisper.cpp",
-                "ggml/src/ggml-aarch64.c",
-                "ggml/src/ggml-alloc.c",
-                "ggml/src/ggml-backend.cpp",
-                "ggml/src/ggml-quants.c",
-                "ggml/src/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
+        .systemLibrary(name: "whisper", pkgConfig: "whisper"),
+    ]
 )

README.md

@@ -7,21 +7,22 @@
 [![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: [v1.7.1](https://github.com/ggerganov/whisper.cpp/releases/tag/v1.7.1) / [Roadmap | F.A.Q.](https://github.com/ggerganov/whisper.cpp/discussions/126)
+Stable: [v1.7.4](https://github.com/ggerganov/whisper.cpp/releases/tag/v1.7.4) / [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](https://github.com/ggerganov/whisper.cpp#core-ml-support)
+- Apple Silicon first-class citizen - optimized via ARM NEON, Accelerate framework, Metal and [Core ML](#core-ml-support)
 - AVX intrinsics support for x86 architectures
 - VSX intrinsics support for POWER architectures
 - Mixed F16 / F32 precision
-- [4-bit and 5-bit integer quantization support](https://github.com/ggerganov/whisper.cpp#quantization)
+- [Integer quantization support](#quantization)
 - Zero memory allocations at runtime
+- [Vulkan support](#vulkan-gpu-support)
 - Support for CPU-only inference
-- [Efficient GPU support for NVIDIA](https://github.com/ggerganov/whisper.cpp#nvidia-gpu-support-via-cublas)
-- [OpenVINO Support](https://github.com/ggerganov/whisper.cpp#openvino-support)
-- [Ascend NPU Support](https://github.com/ggerganov/whisper.cpp#ascend-npu-support)
+- [Efficient GPU support for NVIDIA](#nvidia-gpu-support)
+- [OpenVINO Support](#openvino-support)
+- [Ascend NPU Support](#ascend-npu-support)
 - [C-style API](https://github.com/ggerganov/whisper.cpp/blob/master/include/whisper.h)
 
 Supported platforms:
@@ -52,18 +53,6 @@ On Apple Silicon, the inference runs fully on the GPU via Metal:
 
 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/include/ggml.h) / [ggml.c](ggml/src/ggml.c))
-- The transformer model and the high-level C-style API are implemented in C++ ([whisper.h](include/whisper.h) / [whisper.cpp](src/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:
@@ -72,140 +61,38 @@ First clone the repository:
 git clone https://github.com/ggerganov/whisper.cpp.git
 ```
 
+Navigate into the directory:
+
+```
+cd whisper.cpp
+```
+
 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
 ```
 
-Now build the [main](examples/main) example and transcribe an audio file like this:
+Now build the [whisper-cli](examples/cli) example and transcribe an audio file like this:
 
 ```bash
-# build the main example
-make
+# build the project
+cmake -B build
+cmake --build build --config Release
 
 # transcribe an audio file
-./main -f samples/jfk.wav
+./build/bin/whisper-cli -f samples/jfk.wav
 ```
 
 ---
 
-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
-
-
-sh ./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
-```
+For a quick demo, simply run `make base.en`.
 
 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: `./main -h`
+For detailed usage instructions, run: `./build/bin/whisper-cli -h`
 
-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.
+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.
 For example, you can use `ffmpeg` like this:
 
 ```bash
@@ -217,7 +104,7 @@ 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 samples
+make -j samples
 ```
 
 This will download a few more audio files from Wikipedia and convert them to 16-bit WAV format via `ffmpeg`.
@@ -225,18 +112,18 @@ 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 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
-make large-v3-turbo
+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
 ```
 
 ## Memory usage
@@ -258,11 +145,12 @@ Here are the steps for creating and using a quantized model:
 
 ```bash
 # quantize a model with Q5_0 method
-make quantize
-./quantize models/ggml-base.en.bin models/ggml-base.en-q5_0.bin q5_0
+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
 
 # run the examples as usual, specifying the quantized model file
-./main -m models/ggml-base.en-q5_0.bin ./samples/gb0.wav
+./build/bin/whisper-cli -m models/ggml-base.en-q5_0.bin ./samples/gb0.wav
 ```
 
 ## Core ML support
@@ -296,10 +184,6 @@ 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
@@ -308,7 +192,7 @@ speed-up - more than x3 faster compared with CPU-only execution. Here are the in
 - Run the examples as usual. For example:
 
   ```text
-  $ ./main -m models/ggml-base.en.bin -f samples/jfk.wav
+  $ ./build/bin/whisper-cli -m models/ggml-base.en.bin -f samples/jfk.wav
 
   ...
 
@@ -392,7 +276,7 @@ This can result in significant speedup in encoder performance. Here are the inst
 - Run the examples as usual. For example:
 
   ```text
-  $ ./main -m models/ggml-base.en.bin -f samples/jfk.wav
+  $ ./build/bin/whisper-cli -m models/ggml-base.en.bin -f samples/jfk.wav
 
   ...
 
@@ -409,7 +293,7 @@ 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 Core ML implementation please refer to PR [#1037](https://github.com/ggerganov/whisper.cpp/pull/1037).
+For more information about the OpenVINO implementation please refer to PR [#1037](https://github.com/ggerganov/whisper.cpp/pull/1037).
 
 ## NVIDIA GPU support
 
@@ -419,8 +303,18 @@ First, make sure you have installed `cuda`: https://developer.nvidia.com/cuda-do
 Now build `whisper.cpp` with CUDA support:
 
 ```
-make clean
-GGML_CUDA=1 make -j
+cmake -B build -DGGML_CUDA=1
+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
 ```
 
 ## BLAS CPU support via OpenBLAS
@@ -431,28 +325,13 @@ First, make sure you have installed `openblas`: https://www.openblas.net/
 Now build `whisper.cpp` with OpenBLAS support:
 
 ```
-make clean
-GGML_OPENBLAS=1 make -j
-```
-
-## BLAS CPU support via Intel MKL
-
-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
-
-Now build `whisper.cpp` with Intel MKL BLAS support:
-
-```
-source /opt/intel/oneapi/setvars.sh
-mkdir build
-cd build
-cmake -DWHISPER_MKL=ON ..
-WHISPER_MKL=1 make -j
+cmake -B build -DGGML_BLAS=1
+cmake --build build -j --config Release
 ```
 
 ## Ascend NPU support
 
-Ascend NPU provides inference acceleration via [`CANN`](https://www.hiascend.com/en/software/cann) and AI cores. 
+Ascend NPU provides inference acceleration via [`CANN`](https://www.hiascend.com/en/software/cann) and AI cores.
 
 First, check if your Ascend NPU device is supported:
 
@@ -466,16 +345,14 @@ Then, make sure you have installed [`CANN toolkit`](https://www.hiascend.com/en/
 Now build `whisper.cpp` with CANN support:
 
 ```
-mkdir build
-cd build
-cmake .. -D GGML_CANN=on
-make -j
+cmake -B build -DGGML_CANN=1
+cmake --build build -j --config Release
 ```
 
 Run the inference examples as usual, for example:
 
 ```
-./build/bin/main -f samples/jfk.wav -m models/ggml-base.en.bin -t 8
+./build/bin/whisper-cli -f samples/jfk.wav -m models/ggml-base.en.bin -t 8
 ```
 
 *Notes:*
@@ -483,38 +360,6 @@ Run the inference examples as usual, for example:
 - 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`.
 
-## Docker
-
-### Prerequisites
-
-- Docker must be installed and running on your system.
-- Create a folder to store big models & intermediate files (ex. /whisper/models)
-
-### Images
-
-We have two Docker images available for this project:
-
-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
-
-```shell
-# download model and persist it in a local folder
-docker run -it --rm \
-  -v path/to/models:/models \
-  whisper.cpp:main "./models/download-ggml-model.sh base /models"
-# transcribe an audio file
-docker run -it --rm \
-  -v path/to/models:/models \
-  -v path/to/audios:/audios \
-  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 "./main -m /models/ggml-base.bin -f ./samples/jfk.wav"
-```
-
 ## Installing with Conan
 
 You can install pre-built binaries for whisper.cpp or build it from source using [Conan](https://conan.io/). Use the following command:
@@ -529,89 +374,6 @@ 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.
@@ -619,8 +381,9 @@ The [stream](examples/stream) tool samples the audio every half a second and run
 More info is available in [issue #10](https://github.com/ggerganov/whisper.cpp/issues/10).
 
 ```bash
-make stream
-./stream -m ./models/ggml-base.en.bin -t 8 --step 500 --length 5000
+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
 ```
 
 https://user-images.githubusercontent.com/1991296/194935793-76afede7-cfa8-48d8-a80f-28ba83be7d09.mp4
@@ -631,7 +394,7 @@ Adding the `--print-colors` argument will print the transcribed text using an ex
 to highlight words with high or low confidence:
 
 ```bash
-./main -m models/ggml-base.en.bin -f samples/gb0.wav --print-colors
+./build/bin/whisper-cli -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">
@@ -641,7 +404,7 @@ 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
-$ ./main -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -ml 16
+$ ./build/bin/whisper-cli -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -ml 16
 
 whisper_model_load: loading model from './models/ggml-base.en.bin'
 ...
@@ -665,7 +428,7 @@ 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
-$ ./main -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -ml 1
+$ ./build/bin/whisper-cli -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -ml 1
 
 whisper_model_load: loading model from './models/ggml-base.en.bin'
 ...
@@ -712,7 +475,7 @@ Sample usage:
 ./models/download-ggml-model.sh small.en-tdrz
 
 # run as usual, adding the "-tdrz" command-line argument
-./main -f ./samples/a13.wav -m ./models/ggml-small.en-tdrz.bin -tdrz
+./build/bin/whisper-cli -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 ...
 ...
@@ -729,14 +492,14 @@ main: processing './samples/a13.wav' (480000 samples, 30.0 sec), 4 threads, 1 pr
 
 ## Karaoke-style movie generation (experimental)
 
-The [main](examples/main) example provides support for output of karaoke-style movies, where the
+The [whisper-cli](examples/cli) 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
-./main -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -owts
+./build/bin/whisper-cli -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -owts
 source ./samples/jfk.wav.wts
 ffplay ./samples/jfk.wav.mp4
 ```
@@ -746,7 +509,7 @@ https://user-images.githubusercontent.com/1991296/199337465-dbee4b5e-9aeb-48a3-b
 ---
 
 ```bash
-./main -m ./models/ggml-base.en.bin -f ./samples/mm0.wav -owts
+./build/bin/whisper-cli -m ./models/ggml-base.en.bin -f ./samples/mm0.wav -owts
 source ./samples/mm0.wav.wts
 ffplay ./samples/mm0.wav.mp4
 ```
@@ -756,7 +519,7 @@ https://user-images.githubusercontent.com/1991296/199337504-cc8fd233-0cb7-4920-9
 ---
 
 ```bash
-./main -m ./models/ggml-base.en.bin -f ./samples/gb0.wav -owts
+./build/bin/whisper-cli -m ./models/ggml-base.en.bin -f ./samples/gb0.wav -owts
 source ./samples/gb0.wav.wts
 ffplay ./samples/gb0.wav.mp4
 ```
@@ -781,7 +544,7 @@ 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 [bench](examples/bench) tool. The tool simply runs the Encoder part of the model and prints how much time it
+use the [whisper-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/ggerganov/whisper.cpp/issues/89)
@@ -844,13 +607,12 @@ Some of the examples are even ported to run in the browser using WebAssembly. Ch
 
 | Example                                             | Web                                   | Description                                                                                                                     |
 | --------------------------------------------------- | ------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------- |
-| [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-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                                                                                                           |
 | [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                                                                                    |
@@ -858,7 +620,7 @@ Some of the examples are even ported to run in the browser using WebAssembly. Ch
 | [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/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) |
-| [server](examples/server)                           |                                       | HTTP transcription server with OAI-like API                                                                                     |
+| [wchess](examples/wchess)                           | [wchess.wasm](examples/wchess)        | Voice-controlled chess                                                                                                          |
 
 ## [Discussions](https://github.com/ggerganov/whisper.cpp/discussions)
 

Sources/whisper/module.modulemap

@@ -0,0 +1,5 @@
+module whisper [system] {
+    header "whisper.h"
+    link "whisper"
+    export *
+}

Sources/whisper/whisper.h

@@ -0,0 +1,4 @@
+#pragma once
+
+#include <whisper.h>
+

bindings/java/README.md

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

bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperFullParams.java

@@ -181,11 +181,11 @@ public class WhisperFullParams extends Structure {
     }
 
     /** Flag to suppress non-speech tokens. */
-    public CBool suppress_non_speech_tokens;
+    public CBool suppress_nst;
 
     /** Flag to suppress non-speech tokens. */
     public void suppressNonSpeechTokens(boolean enable) {
-        suppress_non_speech_tokens = enable ? CBool.TRUE : CBool.FALSE;
+        suppress_nst = enable ? CBool.TRUE : CBool.FALSE;
     }
 
     /** Initial decoding temperature. */
@@ -315,7 +315,7 @@ public class WhisperFullParams extends Structure {
                 "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",
+                "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",
                 "progress_callback", "progress_callback_user_data",

bindings/javascript/package.json

@@ -1,6 +1,6 @@
 {
   "name": "whisper.cpp",
-  "version": "1.7.1",
+  "version": "1.7.4",
   "description": "Whisper speech recognition",
   "main": "whisper.js",
   "scripts": {

bindings/ruby/.gitignore

@@ -0,0 +1,3 @@
+LICENSE
+pkg/
+lib/whisper.*

bindings/ruby/README.md

@@ -0,0 +1,243 @@
+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
+
+Usage
+-----
+
+```ruby
+require "whisper"
+
+whisper = Whisper::Context.new("base")
+
+params = Whisper::Params.new
+params.language = "en"
+params.offset = 10_000
+params.duration = 60_000
+params.max_text_tokens = 300
+params.translate = true
+params.print_timestamps = false
+params.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:
+
+```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 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 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.
+
+API
+---
+
+### 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)
+
+whisper.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_next_turn?
+  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_next_turn?
+  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)
+whisper.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/ggerganov/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.
+
+License
+-------
+
+The same to [whisper.cpp][].
+
+[whisper.cpp]: https://github.com/ggerganov/whisper.cpp
+[models]: https://github.com/ggerganov/whisper.cpp/tree/master/models

bindings/ruby/Rakefile

@@ -1,12 +1,64 @@
 require 'rake/clean'
-  require 'rubygems/package'
+require "bundler/gem_tasks"
+require "rake/testtask"
+require_relative "extsources"
 
-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)
+SOURCES = FileList[]
+
+EXTSOURCES.each do |src|
+  basename = src.pathmap("%f")
+  dest = basename == "LICENSE" ? basename : src.pathmap("%{../..,ext}p")
+  dir = dest.pathmap("%d")
+  file src
+  directory dir
+  file dest => [src, dir] do |t|
+    cp t.source, t.name
+  end
+  SOURCES.include dest
 end
+
+CLEAN.include SOURCES
+CLEAN.include FileList["ext/*.o", "ext/*.metal", "ext/whisper.{so,bundle,dll}"]
+
+task build: ["ext/Makefile", "ext/ruby_whisper.h", "ext/ruby_whisper.cpp", "whispercpp.gemspec"]
+
+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" => ["ext/extconf.rb", "ext/ruby_whisper.h", "ext/ruby_whisper.cpp"] + SOURCES do |t|
+  Dir.chdir "ext" do
+    ruby "extconf.rb"
+  end
+end
+
+file SO_FILE => "ext/Makefile" do |t|
+  Dir.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 do |t|
+  t.test_files = FileList["tests/test_*.rb"]
+end
+
+TEST_MEMORY_VIEW = "tests/jfk_reader/jfk_reader.#{RbConfig::CONFIG['DLEXT']}"
+file TEST_MEMORY_VIEW => "tests/jfk_reader/jfk_reader.c" do |t|
+  Dir.chdir "tests/jfk_reader" do
+    ruby "extconf.rb"
+    sh "make"
+  end
+end
+CLEAN.include "tests/jfk_reader/jfk_reader.{o,#{RbConfig::CONFIG['DLEXT']}}"
+
+task test: [LIB_FILE, TEST_MEMORY_VIEW]

bindings/ruby/ext/.gitignore

@@ -1,9 +1,13 @@
 Makefile
-ggml.c
-ggml.h
-ggml-alloc.c
-ggml-alloc.h
+whisper.so
 whisper.bundle
-whisper.cpp
-whisper.h
-dr_wav.h
+whisper.dll
+scripts/get-flags.mk
+*.o
+*.c
+*.cpp
+*.h
+*.m
+*.metal
+!ruby_whisper.cpp
+!ruby_whisper.h

bindings/ruby/ext/cpu.mk

@@ -0,0 +1,9 @@
+ggml/src/ggml-cpu/ggml-cpu-cpp.o: \
+	ggml/src/ggml-cpu/ggml-cpu.cpp \
+	ggml/include/ggml-backend.h \
+	ggml/include/ggml.h \
+	ggml/include/ggml-alloc.h \
+	ggml/src/ggml-backend-impl.h \
+	ggml/include/ggml-cpu.h \
+	ggml/src/ggml-impl.h
+	$(CXX) $(CXXFLAGS)   -c $< -o $@

bindings/ruby/ext/extconf.rb

@@ -1,24 +1,10 @@
 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__),'..','..','..','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-aarch64.h')} .")
-system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml-aarch64.c')} .")
-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.cpp')} .")
-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')} .")
-
 
 # need to use c++ compiler flags
-$CXXFLAGS << ' -std=c++11'
+$CXXFLAGS << ' -std=c++17'
+
+$LDFLAGS << ' -lstdc++'
+
 # Set to true when building binary gems
 if enable_config('static-stdlib', false)
   $LDFLAGS << ' -static-libgcc -static-libstdc++'
@@ -29,4 +15,185 @@ if enable_config('march-tune-native', false)
   $CXXFLAGS << ' -march=native -mtune=native'
 end
 
+if ENV['WHISPER_METAL']
+  $GGML_METAL ||= true
+  $DEPRECATE_WARNING ||= true
+end
+
+$UNAME_S = `uname -s`.chomp
+$UNAME_P = `uname -p`.chomp
+$UNAME_M = `uname -m`.chomp
+
+if $UNAME_S == 'Darwin'
+  unless ENV['GGML_NO_METAL']
+    $GGML_METAL ||= true
+  end
+  $GGML_NO_OPENMP ||= true
+end
+
+if $GGML_METAL
+  $GGML_METAL_EMBED_LIBRARY = true
+end
+
+$MK_CPPFLAGS = '-Iggml/include -Iggml/src -Iggml/src/ggml-cpu -Iinclude -Isrc -Iexamples'
+$MK_CFLAGS   = '-std=c11   -fPIC'
+$MK_CXXFLAGS = '-std=c++17 -fPIC'
+$MK_NVCCFLAGS = '-std=c++17'
+$MK_LDFLAGS = ''
+
+$OBJ_GGML = []
+$OBJ_WHISPER = []
+$OBJ_COMMON = []
+$OBJ_SDL = []
+
+$MK_CPPFLAGS << ' -D_XOPEN_SOURCE=600'
+
+if $UNAME_S == 'Linux'
+  $MK_CPPFLAGS << ' -D_GNU_SOURCE'
+end
+
+if $UNAME_S == 'Darwin'
+  $MK_CPPFLAGS << ' -D_DARWIN_C_SOURCE'
+end
+
+if ENV['WHISPER_DEBUG']
+  $MK_CFLAGS    << ' -O0 -g'
+  $MK_CXXFLAGS  << ' -O0 -g'
+  $MK_LDFLAGS   << ' -g'
+  $MK_NVCCFLAGS << ' -O0 -g'
+else
+  $MK_CPPFLAGS   << ' -DNDEBUG'
+  $MK_CFLAGS     << ' -O3'
+  $MK_CXXFLAGS   << ' -O3'
+  $MK_NVCCFLAGS  << ' -O3'
+end
+
+$WARN_FLAGS =
+  ' -Wall' <<
+  ' -Wextra' <<
+  ' -Wpedantic' <<
+  ' -Wcast-qual' <<
+  ' -Wno-unused-function'
+
+$MK_CFLAGS <<
+  $WARN_FLAGS <<
+  ' -Wshadow' <<
+  ' -Wstrict-prototypes' <<
+  ' -Wpointer-arith' <<
+  ' -Wmissing-prototypes' <<
+  ' -Werror=implicit-int' <<
+  ' -Werror=implicit-function-declaration'
+
+$MK_CXXFLAGS <<
+  $WARN_FLAGS <<
+  ' -Wmissing-declarations' <<
+  ' -Wmissing-noreturn'
+
+unless `#{cc_command} #{$LDFLAGS} -Wl,-v 2>&1`.chomp.include? 'dyld-1015.7'
+  $MK_CPPFLAGS << ' -DHAVE_BUGGY_APPLE_LINKER'
+end
+
+if %w[Linux Darwin FreeBSD NetBSD OpenBSD Haiku].include? $UNAME_S
+  $MK_CFLAGS   << ' -pthread'
+  $MK_CXXFLAGS << ' -pthread'
+end
+
+unless $_WIN32
+  $DSO_EXT = '.so'
+else
+  $DSO_EXT = '.dll'
+end
+
+unless ENV['RISCV']
+  if %w[x86_64 i686 amd64].include? $UNAME_M
+    $HOST_CXXFLAGS ||= ''
+
+    $MK_CFLAGS     << ' -march=native -mtune=native'
+    $HOST_CXXFLAGS << ' -march=native -mtune=native'
+  end
+else
+  $MK_CFLAGS   << ' -march=rv64gcv -mabi=lp64d'
+  $MK_CXXFLAGS << ' -march=rv64gcv -mabi=lp64d'
+end
+
+unless ENV['GGML_NO_ACCELERATE']
+  if $UNAME_S == 'Darwin'
+    $MK_CPPFLAGS << ' -DGGML_USE_ACCELERATE -DGGML_USE_BLAS -DGGML_BLAS_USE_ACCELERATE'
+    $MK_CPPFLAGS << ' -DACCELERATE_NEW_LAPACK'
+    $MK_CPPFLAGS << ' -DACCELERATE_LAPACK_ILP64'
+    $MK_LDFLAGS  << ' -framework Accelerate'
+    $OBJ_GGML    << 'ggml/src/ggml-blas/ggml-blas.o'
+  end
+end
+
+if ENV['GGML_OPENBLAS']
+  $MK_CPPFLAGS << " -DGGML_USE_BLAS #{`pkg-config --cflags-only-I openblas`.chomp}"
+  $MK_CFLAGS   << " #{`pkg-config --cflags-only-other openblas)`.chomp}"
+  $MK_LDFLAGS  << " #{`pkg-config --libs openblas`}"
+  $OBJ_GGML    << 'ggml/src/ggml-blas/ggml-blas.o'
+end
+
+if ENV['GGML_OPENBLAS64']
+  $MK_CPPFLAGS << " -DGGML_USE_BLAS #{`pkg-config --cflags-only-I openblas64`.chomp}"
+  $MK_CFLAGS   << " #{`pkg-config --cflags-only-other openblas64)`.chomp}"
+  $MK_LDFLAGS  << " #{`pkg-config --libs openblas64`}"
+  $OBJ_GGML    << 'ggml/src/ggml-blas/ggml-blas.o'
+end
+
+if $GGML_METAL
+  $MK_CPPFLAGS << ' -DGGML_USE_METAL'
+  $MK_LDFLAGS  << ' -framework Foundation -framework Metal -framework MetalKit'
+  $OBJ_GGML    << 'ggml/src/ggml-metal/ggml-metal.o'
+
+  if ENV['GGML_METAL_NDEBUG']
+    $MK_CPPFLAGS << ' -DGGML_METAL_NDEBUG'
+  end
+
+  if $GGML_METAL_EMBED_LIBRARY
+    $MK_CPPFLAGS << ' -DGGML_METAL_EMBED_LIBRARY'
+    $OBJ_GGML    << 'ggml/src/ggml-metal/ggml-metal-embed.o'
+  end
+end
+
+$OBJ_GGML <<
+  'ggml/src/ggml.o' <<
+  'ggml/src/ggml-alloc.o' <<
+  'ggml/src/ggml-backend.o' <<
+  'ggml/src/ggml-backend-reg.o' <<
+  'ggml/src/ggml-opt.o' <<
+  'ggml/src/ggml-quants.o' <<
+  'ggml/src/ggml-threading.o' <<
+  'ggml/src/ggml-cpu/ggml-cpu.o' <<
+  'ggml/src/ggml-cpu/ggml-cpu-cpp.o' <<
+  'ggml/src/ggml-cpu/ggml-cpu-aarch64.o' <<
+  'ggml/src/ggml-cpu/ggml-cpu-hbm.o' <<
+  'ggml/src/ggml-cpu/ggml-cpu-quants.o' <<
+  'ggml/src/ggml-cpu/ggml-cpu-traits.o'
+
+$OBJ_WHISPER <<
+  'src/whisper.o'
+
+$objs = $OBJ_GGML + $OBJ_WHISPER + $OBJ_COMMON + $OBJ_SDL
+$objs << "ruby_whisper.o"
+
+$CPPFLAGS  = "#{$MK_CPPFLAGS} #{$CPPFLAGS}"
+$CFLAGS    = "#{$CPPFLAGS} #{$MK_CFLAGS} #{$GF_CFLAGS} #{$CFLAGS}"
+$BASE_CXXFLAGS = "#{$MK_CXXFLAGS} #{$CXXFLAGS}"
+$CXXFLAGS  = "#{$BASE_CXXFLAGS} #{$HOST_CXXFLAGS} #{$GF_CXXFLAGS} #{$CPPFLAGS}"
+$NVCCFLAGS = "#{$MK_NVCCFLAGS} #{$NVCCFLAGS}"
+$LDFLAGS   = "#{$MK_LDFLAGS} #{$LDFLAGS}"
+
 create_makefile('whisper')
+
+File.open 'Makefile', 'a' do |file|
+  file.puts 'include scripts/get-flags.mk'
+  file.puts 'include cpu.mk'
+
+  if $GGML_METAL
+    file.puts 'include metal.mk'
+
+    if $GGML_METAL_EMBED_LIBRARY
+      file.puts 'include metal-embed.mk'
+    end
+  end
+end

bindings/ruby/ext/ggml-backend-impl.h

@@ -1,141 +0,0 @@
-#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

bindings/ruby/ext/ggml-backend.h

@@ -1,233 +0,0 @@
-#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

bindings/ruby/ext/ggml-cuda.h

@@ -1,43 +0,0 @@
-#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

bindings/ruby/ext/ggml-impl.h

@@ -1,272 +0,0 @@
-#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

bindings/ruby/ext/ggml-kompute.h

@@ -1,46 +0,0 @@
-#pragma once
-
-#include "ggml.h"
-#include "ggml-backend.h"
-
-#include <stdbool.h>
-#include <stddef.h>
-#include <stdint.h>
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-struct ggml_vk_device {
-    int index;
-    int type; // same as VkPhysicalDeviceType
-    size_t heapSize;
-    const char * name;
-    const char * vendor;
-    int subgroupSize;
-    uint64_t bufferAlignment;
-    uint64_t maxAlloc;
-};
-
-struct ggml_vk_device * ggml_vk_available_devices(size_t memoryRequired, size_t * count);
-bool ggml_vk_get_device(struct ggml_vk_device * device, size_t memoryRequired, const char * name);
-bool ggml_vk_has_vulkan(void);
-bool ggml_vk_has_device(void);
-struct ggml_vk_device ggml_vk_current_device(void);
-
-//
-// backend API
-//
-
-// forward declaration
-typedef struct ggml_backend * ggml_backend_t;
-
-GGML_API ggml_backend_t ggml_backend_kompute_init(int device);
-
-GGML_API bool ggml_backend_is_kompute(ggml_backend_t backend);
-
-GGML_API ggml_backend_buffer_type_t ggml_backend_kompute_buffer_type(int device);
-
-#ifdef __cplusplus
-}
-#endif

bindings/ruby/ext/ggml-metal.h

@@ -1,66 +0,0 @@
-// 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, OpenCL, etc.)
-//
-// How it works?
-//
-// As long as your program can create and evaluate a ggml_cgraph on the CPU, you can use this
-// interface to evaluate the same graph on the GPU. Instead of using ggml_graph_compute(), you
-// use ggml_metal_graph_compute() (or ggml_vulkan_graph_compute(), etc.)
-//
-// You only need to make sure that all memory buffers that you used during the graph creation
-// are mapped to the device memory with the ggml_metal_add_buffer() function. This mapping is
-// used during the graph evaluation to determine the arguments of the compute kernels.
-//
-// Synchronization between device and host memory (for example for input and output tensors)
-// is done with the ggml_metal_set_tensor() and ggml_metal_get_tensor() functions.
-//
-
-#pragma once
-
-#include "ggml.h"
-#include "ggml-backend.h"
-
-#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;
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-//
-// backend API
-// user-code should use only these functions
-//
-
-GGML_API void ggml_backend_metal_log_set_callback(ggml_log_callback log_callback, void * user_data);
-
-GGML_API ggml_backend_t ggml_backend_metal_init(void);
-
-GGML_API bool ggml_backend_is_metal(ggml_backend_t backend);
-
-GGML_API GGML_CALL ggml_backend_buffer_t ggml_backend_metal_buffer_from_ptr(void * data, size_t size, size_t max_size);
-
-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_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_API void ggml_backend_metal_capture_next_compute(ggml_backend_t backend);
-
-#ifdef __cplusplus
-}
-#endif
-

bindings/ruby/ext/ggml-opencl.h

@@ -1,36 +0,0 @@
-#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

bindings/ruby/ext/ggml-quants.h

@@ -1,133 +0,0 @@
-#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
-

bindings/ruby/ext/ggml-sycl.h

@@ -1,49 +0,0 @@
-//
-//  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

bindings/ruby/ext/ggml-vulkan.h

@@ -1,29 +0,0 @@
-#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

bindings/ruby/ext/metal-embed.mk

@@ -0,0 +1,17 @@
+ggml/src/ggml-metal/ggml-metal-embed.o: \
+	ggml/src/ggml-metal/ggml-metal.metal \
+	ggml/src/ggml-metal/ggml-metal-impl.h \
+	ggml/src/ggml-common.h
+	@echo "Embedding Metal library"
+	@sed -e '/__embed_ggml-common.h__/r      ggml/src/ggml-common.h'                -e '/__embed_ggml-common.h__/d'      < ggml/src/ggml-metal/ggml-metal.metal           > ggml/src/ggml-metal/ggml-metal-embed.metal.tmp
+	@sed -e '/#include "ggml-metal-impl.h"/r ggml/src/ggml-metal/ggml-metal-impl.h' -e '/#include "ggml-metal-impl.h"/d' < ggml/src/ggml-metal/ggml-metal-embed.metal.tmp > ggml/src/ggml-metal/ggml-metal-embed.metal
+	$(eval TEMP_ASSEMBLY=$(shell mktemp -d))
+	@echo ".section __DATA, __ggml_metallib"                       >  $(TEMP_ASSEMBLY)/ggml-metal-embed.s
+	@echo ".globl _ggml_metallib_start"                            >> $(TEMP_ASSEMBLY)/ggml-metal-embed.s
+	@echo "_ggml_metallib_start:"                                  >> $(TEMP_ASSEMBLY)/ggml-metal-embed.s
+	@echo ".incbin \"ggml/src/ggml-metal/ggml-metal-embed.metal\"" >> $(TEMP_ASSEMBLY)/ggml-metal-embed.s
+	@echo ".globl _ggml_metallib_end"                              >> $(TEMP_ASSEMBLY)/ggml-metal-embed.s
+	@echo "_ggml_metallib_end:"                                    >> $(TEMP_ASSEMBLY)/ggml-metal-embed.s
+	$(CC) $(CFLAGS) -c $(TEMP_ASSEMBLY)/ggml-metal-embed.s -o $@
+	@rm -f ${TEMP_ASSEMBLY}/ggml-metal-embed.s
+	@rmdir ${TEMP_ASSEMBLY}

bindings/ruby/ext/metal.mk

@@ -0,0 +1,6 @@
+ggml/src/ggml-metal/ggml-metal.o: \
+	ggml/src/ggml-metal/ggml-metal.m \
+	ggml/src/ggml-metal/ggml-metal-impl.h \
+	ggml/include/ggml-metal.h \
+	ggml/include/ggml.h
+	$(CC) $(CFLAGS) -c $< -o $@

bindings/ruby/ext/ruby_whisper.h

@@ -1,8 +1,15 @@
-#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;
@@ -10,6 +17,9 @@ 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 *abort_callback_container;
 } ruby_whisper_params;
 
 #endif

bindings/ruby/extsources.rb

@@ -0,0 +1,6 @@
+require "yaml"
+
+sources = `git ls-files -z ../..`.split("\x0")
+paths = YAML.load_file("../../.github/workflows/bindings-ruby.yml")[true]["push"]["paths"]
+paths.delete "bindings/ruby/**"
+EXTSOURCES = (Dir.glob(paths, base: "../..").collect {|path| "../../#{path}"} << "../../LICENSE") & sources

bindings/ruby/lib/whisper/model/uri.rb

@@ -0,0 +1,163 @@
+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") ? 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
+              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
+      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
+
+    @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")
+    }
+
+    class << self
+      attr_reader :pre_converted_models
+    end
+  end
+end

bindings/ruby/sig/whisper.rbs

@@ -0,0 +1,153 @@
+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 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) -> log_callback
+  def self.finalize_log_callback: (void) -> void # Second argument of ObjectSpace.define_finalizer
+
+  class Context
+    def initialize: (string | _ToPath | ::URI::HTTP ) -> void
+    def transcribe: (string, Params) -> void
+                  | (string, Params) { (String) -> void } -> void
+    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
+    def full_n_segments: () -> Integer
+    def full_lang_id: () -> Integer
+    def full_get_segment_t0: (Integer) -> Integer
+    def full_get_segment_t1: (Integer) -> Integer
+    def full_get_segment_speaker_turn_next: (Integer) -> (true | false)
+    def full_get_segment_text: (Integer) -> String
+    def full_get_segment_no_speech_prob: (Integer) -> Float
+    def full: (Params, Array[Float], ?Integer) -> void
+            | (Params, _Samples, ?Integer) -> void
+    def full_parallel: (Params, Array[Float], ?Integer) -> void
+                     | (Params, _Samples, ?Integer) -> void
+                     | (Params, _Samples, ?Integer?, Integer) -> void
+    def each_segment: { (Segment) -> void } -> void
+                    | () -> Enumerator[Segment]
+    def model: () -> Model
+  end
+
+  class Params
+    def initialize: () -> void
+    def language=: (String) -> String # TODO: Enumerate lang names
+    def language: () -> String
+    def translate=: (boolish) -> boolish
+    def translate: () -> (true | false)
+    def no_context=: (boolish) -> boolish
+    def no_context: () -> (true | false)
+    def single_segment=: (boolish) -> boolish
+    def single_segment: () -> (true | false)
+    def print_special=: (boolish) -> boolish
+    def print_special: () -> (true | false)
+    def print_progress=: (boolish) -> boolish
+    def print_progress: () -> (true | false)
+    def print_realtime=: (boolish) -> boolish
+    def print_realtime: () -> (true | false)
+    def print_timestamps=: (boolish) -> boolish
+    def print_timestamps: () -> (true | false)
+    def suppress_blank=: (boolish) -> boolish
+    def suppress_blank: () -> (true | false)
+    def suppress_nst=: (boolish) -> boolish
+    def suppress_nst: () -> (true | false)
+    def token_timestamps=: (boolish) -> boolish
+    def token_timestamps: () -> (true | false)
+    def split_on_word=: (boolish) -> boolish
+    def split_on_word: () -> (true | false)
+    def initial_prompt=: (_ToS) -> _ToS
+    def initial_prompt: () -> String
+    def diarize=: (boolish) -> boolish
+    def diarize: () -> (true | false)
+    def offset=: (Integer) -> Integer
+    def offset: () -> Integer
+    def duration=: (Integer) -> Integer
+    def duration: () -> Integer
+    def max_text_tokens=: (Integer) -> Integer
+    def max_text_tokens: () -> Integer
+    def temperature=: (Float) -> Float
+    def temperature: () -> Float
+    def max_initial_ts=: (Float) -> Float
+    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
+    def entropy_thold: () -> Float
+    def logprob_thold=: (Float) -> Float
+    def logprob_thold: () -> Float
+    def no_speech_thold=: (Float) -> Float
+    def no_speech_thold: () -> Float
+    def new_segment_callback=: (new_segment_callback) -> new_segment_callback
+    def new_segment_callback_user_data=: (Object) -> Object
+    def progress_callback=: (progress_callback) -> progress_callback
+    def progress_callback_user_data=: (Object) -> Object
+    def abort_callback=: (abort_callback) -> abort_callback
+    def abort_callback_user_data=: (Object) -> Object
+    def on_new_segment: { (Segment) -> void } -> void
+    def on_progress: { (Integer) -> void } -> void
+    def abort_on: { (Object) -> boolish } -> void
+  end
+
+  class Model
+    def self.pre_converted_models: () -> Hash[String, Model::URI]
+    def initialize: () -> void
+    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 initialize: (string | ::URI::HTTP) -> void
+      def to_path: -> String
+      def clear_cache: -> void
+    end
+  end
+
+  class Segment
+    def initialize: () -> void
+    def start_time: () -> Integer
+    def end_time: () -> Integer
+    def speaker_next_turn?: () -> (true | false)
+    def text: () -> String
+    def no_speech_prob: () -> Float
+  end
+
+  class Error < StandardError
+    attr_reader code: Integer
+
+    def initialize: (Integer) -> void
+  end
+end

bindings/ruby/tests/helper.rb

@@ -0,0 +1,24 @@
+require "test/unit"
+require "whisper"
+require_relative "jfk_reader/jfk_reader"
+
+class TestBase < Test::Unit::TestCase
+  AUDIO = File.join(__dir__, "..", "..", "..", "samples", "jfk.wav")
+
+  class << self
+    attr_reader :whisper
+
+    def startup
+      @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

bindings/ruby/tests/jfk_reader/.gitignore

@@ -0,0 +1,5 @@
+Makefile
+jfk_reader.o
+jfk_reader.so
+jfk_reader.bundle
+jfk_reader.dll

bindings/ruby/tests/jfk_reader/extconf.rb

@@ -0,0 +1,3 @@
+require "mkmf"
+
+create_makefile("jfk_reader")

bindings/ruby/tests/jfk_reader/jfk_reader.c

@@ -0,0 +1,68 @@
+#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);
+}

bindings/ruby/tests/test_callback.rb

@@ -0,0 +1,160 @@
+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_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

bindings/ruby/tests/test_error.rb

@@ -0,0 +1,20 @@
+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

bindings/ruby/tests/test_model.rb

@@ -0,0 +1,109 @@
+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
+end

bindings/ruby/tests/test_package.rb

@@ -0,0 +1,31 @@
+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
+      match_data = `rake -Tbuild`.match(/(whispercpp-(.+)\.gem)/)
+      filename = match_data[1]
+      version = match_data[2]
+      basename = "whisper.#{RbConfig::CONFIG["DLEXT"]}"
+      Dir.mktmpdir do |dir|
+        system "gem", "install", "--install-dir", dir.shellescape, "--no-document", "pkg/#{filename.shellescape}", exception: true
+        assert_path_exist File.join(dir, "gems/whispercpp-#{version}/lib", basename)
+      end
+    end
+  end
+end

bindings/ruby/tests/test_params.rb

@@ -0,0 +1,160 @@
+require_relative "helper"
+
+class TestParams < TestBase
+  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
+end

bindings/ruby/tests/test_segment.rb

@@ -0,0 +1,74 @@
+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
+end

bindings/ruby/tests/test_whisper.rb

@@ -1,131 +1,217 @@
-TOPDIR = File.expand_path(File.join(File.dirname(__FILE__), '..'))
-EXTDIR = File.join(TOPDIR, 'ext')
-#$LIBDIR = File.join(TOPDIR, 'lib')
-#$:.unshift(LIBDIR)
-$:.unshift(EXTDIR)
+require_relative "helper"
+require "stringio"
+require "etc"
 
-require 'whisper'
-require 'test/unit'
+# Exists to detect memory-related bug
+Whisper.log_set ->(level, buffer, user_data) {}, nil
 
-class TestWhisper < Test::Unit::TestCase
+class TestWhisper < TestBase
   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_non_speech_tokens
-    @params.suppress_non_speech_tokens = true
-    assert @params.suppress_non_speech_tokens
-    @params.suppress_non_speech_tokens = false
-    assert !@params.suppress_non_speech_tokens
-  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_whisper
-    @whisper = Whisper::Context.new(File.join(TOPDIR, '..', '..', 'models', 'ggml-base.en.bin'))
+    @whisper = Whisper::Context.new("base.en")
     params  = Whisper::Params.new
     params.print_timestamps = false
 
-    jfk = File.join(TOPDIR, '..', '..', 'samples', 'jfk.wav')
-    @whisper.transcribe(jfk, params) {|text|
+    @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
 
+  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_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_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
+      @whisper.full_parallel(@params, @samples, @samples.length, Etc.nprocessors)
+
+      assert_equal Etc.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
+      samples = JFKReader.new(AUDIO)
+      @whisper.full_parallel(@params, samples, nil, Etc.nprocessors)
+
+      assert_equal Etc.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
+      @whisper.full_parallel(@params, @samples, nil, Etc.nprocessors)
+
+      assert_equal Etc.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
 end

bindings/ruby/whispercpp.gemspec

@@ -1,28 +1,36 @@
+require_relative "extsources"
+
 Gem::Specification.new do |s|
   s.name    = "whispercpp"
   s.authors = ["Georgi Gerganov", "Todd A. Fisher"]
-  s.version = '1.3.0'
-  s.date    = '2024-05-14'
+  s.version = '1.3.1'
+  s.date    = '2024-12-19'
   s.description = %q{High-performance inference of OpenAI's Whisper automatic speech recognition (ASR) model via Ruby}
   s.email   = 'todd.fisher@gmail.com'
   s.extra_rdoc_files = ['LICENSE', 'README.md']
-  
-  s.files = ["LICENSE", "README.md", "Rakefile", "ext/extconf.rb", "ext/ggml.c", "ext/ruby_whisper.cpp", "ext/whisper.cpp", "ext/dr_wav.h", "ext/ggml.h", "ext/ruby_whisper.h", "ext/whisper.h"]
 
-  #### Load-time details
-  s.require_paths = ['lib','ext']
+  s.files = `git ls-files . -z`.split("\x0") +
+              EXTSOURCES.collect {|file|
+                basename = File.basename(file)
+                if s.extra_rdoc_files.include?(basename)
+                  basename
+                else
+                  file.sub("../..", "ext")
+                end
+              }
+
   s.summary = %q{Ruby whisper.cpp bindings}
-  s.test_files = ["tests/test_whisper.rb"]
-  
+  s.test_files = s.files.select {|file| file.start_with? "tests/"}
+
   s.extensions << 'ext/extconf.rb'
-  
+  s.required_ruby_version = '>= 3.1.0'
 
   #### Documentation and testing.
   s.homepage = 'https://github.com/ggerganov/whisper.cpp'
-  s.rdoc_options = ['--main', '../../README.md']
+  s.rdoc_options = ['--main', 'README.md']
+
 
-  
     s.platform = Gem::Platform::RUBY
-  
+
   s.licenses = ['MIT']
 end

cmake/DefaultTargetOptions.cmake

@@ -13,5 +13,4 @@ set_target_properties(${TARGET}
     PROPERTIES
         EXPORT_COMPILE_COMMANDS ON
         RUNTIME_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}/bin"
-        INSTALL_RPATH            "${CMAKE_INSTALL_PREFIX}/lib"
 )

cmake/whisper.pc.in

@@ -1,10 +1,10 @@
 prefix=@CMAKE_INSTALL_PREFIX@
 exec_prefix=${prefix}
-libdir=@CMAKE_INSTALL_FULL_LIBDIR@
+libdir=${exec_prefix}/lib
 includedir=${prefix}/include
 
 Name: whisper
 Description: Port of OpenAI's Whisper model in C/C++
 Version: @PROJECT_VERSION@
-Libs: -L${libdir} -lwhisper
+Libs: -L${libdir} -lggml  -lggml-base -lwhisper
 Cflags: -I${includedir}

examples/CMakeLists.txt

@@ -97,52 +97,29 @@ include_directories(${CMAKE_CURRENT_SOURCE_DIR})
 
 if (EMSCRIPTEN)
     add_subdirectory(whisper.wasm)
-    set_target_properties(libmain PROPERTIES FOLDER "libs")
     add_subdirectory(stream.wasm)
-    set_target_properties(libstream PROPERTIES FOLDER "libs")
     add_subdirectory(command.wasm)
-    set_target_properties(libcommand PROPERTIES FOLDER "libs")
-    #add_subdirectory(talk.wasm)
-    #set_target_properties(libtalk PROPERTIES FOLDER "libs")
     add_subdirectory(bench.wasm)
-    set_target_properties(libbench PROPERTIES FOLDER "libs")
 elseif(CMAKE_JS_VERSION)
     add_subdirectory(addon.node)
-    set_target_properties(addon.node PROPERTIES FOLDER "examples")
 else()
-    add_subdirectory(main)
-    set_target_properties(main PROPERTIES FOLDER "examples")
-if (WHISPER_SDL2)
-    add_subdirectory(stream)
-    set_target_properties(stream PROPERTIES FOLDER "examples")
-endif (WHISPER_SDL2)
-    add_subdirectory(server)
-    set_target_properties(server PROPERTIES FOLDER "examples")
-if (WHISPER_SDL2)
-    add_subdirectory(command)
-    set_target_properties(command PROPERTIES FOLDER "examples")
-endif (WHISPER_SDL2)
+    add_subdirectory(cli)
     add_subdirectory(bench)
-    set_target_properties(bench PROPERTIES FOLDER "examples")
+    add_subdirectory(server)
     add_subdirectory(quantize)
-    set_target_properties(quantize PROPERTIES FOLDER "examples")
-if (WHISPER_SDL2)
-    # TODO: disabled until update
-    #       https://github.com/ggerganov/whisper.cpp/issues/1818
-    #add_subdirectory(talk)
-    #set_target_properties(talk PROPERTIES FOLDER "examples")
-    add_subdirectory(talk-llama)
-    set_target_properties(talk-llama PROPERTIES FOLDER "examples")
-    add_subdirectory(lsp)
-    set_target_properties(lsp PROPERTIES FOLDER "examples")
-    if (GGML_SYCL)
-        add_subdirectory(sycl)
-        set_target_properties(sycl PROPERTIES FOLDER "examples")
-    endif()
-endif (WHISPER_SDL2)
+    if (WHISPER_SDL2)
+        add_subdirectory(stream)
+        add_subdirectory(command)
+        add_subdirectory(talk-llama)
+        add_subdirectory(lsp)
+        if (GGML_SYCL)
+            add_subdirectory(sycl)
+        endif()
+    endif (WHISPER_SDL2)
+
+    add_subdirectory(deprecation-warning)
 endif()
 
 if (WHISPER_SDL2)
     add_subdirectory(wchess)
-    set_target_properties(wchess PROPERTIES FOLDER "examples")
 endif (WHISPER_SDL2)

examples/bench/CMakeLists.txt

@@ -1,6 +1,8 @@
-set(TARGET bench)
+set(TARGET whisper-bench)
 add_executable(${TARGET} bench.cpp)
 
 include(DefaultTargetOptions)
 
 target_link_libraries(${TARGET} PRIVATE whisper ${CMAKE_THREAD_LIBS_INIT})
+
+install(TARGETS ${TARGET} RUNTIME)

examples/bench/README.md

@@ -1,4 +1,4 @@
-# bench
+# whisper.cpp/examples/bench
 
 A very basic tool for benchmarking the inference performance on your device. The tool simply runs the Encoder part of
 the transformer on some random audio data and records the execution time. This way we can have an objective comparison
@@ -7,11 +7,8 @@ of the performance of the model for various setups.
 Benchmark results are tracked in the following Github issue: https://github.com/ggerganov/whisper.cpp/issues/89
 
 ```bash
-# build the bench tool
-$ make bench
-
-# run it on the small.en model using 4 threads
-$ ./bench -m ./models/ggml-small.en.bin -t 4
+# run the bench too on the small.en model using 4 threads
+$ ./build/bin/whisper-bench -m ./models/ggml-small.en.bin -t 4
 
 whisper_model_load: loading model from './models/ggml-small.en.bin'
 whisper_model_load: n_vocab       = 51864

examples/cli/CMakeLists.txt

@@ -1,6 +1,8 @@
-set(TARGET main)
-add_executable(${TARGET} main.cpp)
+set(TARGET whisper-cli)
+add_executable(${TARGET} cli.cpp)
 
 include(DefaultTargetOptions)
 
 target_link_libraries(${TARGET} PRIVATE common whisper ${FFMPEG_LIBRARIES} ${CMAKE_THREAD_LIBS_INIT})
+
+install(TARGETS ${TARGET} RUNTIME)

examples/cli/README.md

@@ -1,12 +1,12 @@
-# main
+# whisper.cpp/examples/cli
 
 This is the main example demonstrating most of the functionality of the Whisper model.
 It can be used as a reference for using the `whisper.cpp` library in other projects.
 
 ```
-./main -h
+./build/bin/whisper-cli -h
 
-usage: ./main [options] file0.wav file1.wav ...
+usage: ./build-pkg/bin/whisper-cli [options] file0.wav file1.wav ...
 
 options:
   -h,        --help              [default] show this help message and exit
@@ -20,9 +20,12 @@ options:
   -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
+  -ac N,     --audio-ctx N       [0      ] audio context size (0 - all)
   -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
+  -tp,       --temperature N     [0.00   ] The sampling temperature, between 0 and 1
+  -tpi,      --temperature-inc N [0.20   ] The increment of temperature, between 0 and 1
   -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
@@ -38,16 +41,23 @@ options:
   -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)
+  -np,       --no-prints         [false  ] do not print anything other than the results
   -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
+             --prompt PROMPT     [       ] initial prompt (max n_text_ctx/2 tokens)
   -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
+  -dtw MODEL --dtw MODEL         [       ] compute token-level timestamps
   -ls,       --log-score         [false  ] log best decoder scores of tokens
   -ng,       --no-gpu            [false  ] disable GPU
+  -fa,       --flash-attn        [false  ] flash attention
+  --suppress-regex REGEX         [       ] regular expression matching tokens to suppress
+  --grammar GRAMMAR              [       ] GBNF grammar to guide decoding
+  --grammar-rule RULE            [       ] top-level GBNF grammar rule name
+  --grammar-penalty N            [100.0  ] scales down logits of nongrammar tokens
 ```

examples/cli/cli.cpp

@@ -43,6 +43,7 @@ struct whisper_params {
     float word_thold      =  0.01f;
     float entropy_thold   =  2.40f;
     float logprob_thold   = -1.00f;
+    float no_speech_thold =  0.6f;
     float grammar_penalty = 100.0f;
     float temperature     = 0.0f;
     float temperature_inc = 0.2f;
@@ -70,6 +71,7 @@ struct whisper_params {
     bool log_score       = false;
     bool use_gpu         = true;
     bool flash_attn      = false;
+    bool suppress_nst    = false;
 
     std::string language  = "en";
     std::string prompt;
@@ -104,6 +106,11 @@ static char * whisper_param_turn_lowercase(char * in){
     return in;
 }
 
+static char * requires_value_error(const std::string & arg) {
+    fprintf(stderr, "error: argument %s requires value\n", arg.c_str());
+    exit(0);
+}
+
 static bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
     for (int i = 1; i < argc; i++) {
         std::string arg = argv[i];
@@ -122,21 +129,23 @@ static bool whisper_params_parse(int argc, char ** argv, whisper_params & params
             whisper_print_usage(argc, argv, params);
             exit(0);
         }
-        else if (arg == "-t"    || arg == "--threads")         { params.n_threads       = std::stoi(argv[++i]); }
-        else if (arg == "-p"    || arg == "--processors")      { params.n_processors    = std::stoi(argv[++i]); }
-        else if (arg == "-ot"   || arg == "--offset-t")        { params.offset_t_ms     = std::stoi(argv[++i]); }
-        else if (arg == "-on"   || arg == "--offset-n")        { params.offset_n        = std::stoi(argv[++i]); }
-        else if (arg == "-d"    || arg == "--duration")        { params.duration_ms     = std::stoi(argv[++i]); }
-        else if (arg == "-mc"   || arg == "--max-context")     { params.max_context     = std::stoi(argv[++i]); }
-        else if (arg == "-ml"   || arg == "--max-len")         { params.max_len         = std::stoi(argv[++i]); }
-        else if (arg == "-bo"   || arg == "--best-of")         { params.best_of         = std::stoi(argv[++i]); }
-        else if (arg == "-bs"   || arg == "--beam-size")       { params.beam_size       = std::stoi(argv[++i]); }
-        else if (arg == "-ac"   || arg == "--audio-ctx")       { params.audio_ctx       = std::stoi(argv[++i]); }
-        else if (arg == "-wt"   || arg == "--word-thold")      { params.word_thold      = std::stof(argv[++i]); }
-        else if (arg == "-et"   || arg == "--entropy-thold")   { params.entropy_thold   = std::stof(argv[++i]); }
-        else if (arg == "-lpt"  || arg == "--logprob-thold")   { params.logprob_thold   = std::stof(argv[++i]); }
-        else if (arg == "-tp"   || arg == "--temperature")     { params.temperature     = std::stof(argv[++i]); }
-        else if (arg == "-tpi"  || arg == "--temperature-inc") { params.temperature_inc = std::stof(argv[++i]); }
+        #define ARGV_NEXT (((i + 1) < argc) ? argv[++i] : requires_value_error(arg))
+        else if (arg == "-t"    || arg == "--threads")         { params.n_threads       = std::stoi(ARGV_NEXT); }
+        else if (arg == "-p"    || arg == "--processors")      { params.n_processors    = std::stoi(ARGV_NEXT); }
+        else if (arg == "-ot"   || arg == "--offset-t")        { params.offset_t_ms     = std::stoi(ARGV_NEXT); }
+        else if (arg == "-on"   || arg == "--offset-n")        { params.offset_n        = std::stoi(ARGV_NEXT); }
+        else if (arg == "-d"    || arg == "--duration")        { params.duration_ms     = std::stoi(ARGV_NEXT); }
+        else if (arg == "-mc"   || arg == "--max-context")     { params.max_context     = std::stoi(ARGV_NEXT); }
+        else if (arg == "-ml"   || arg == "--max-len")         { params.max_len         = std::stoi(ARGV_NEXT); }
+        else if (arg == "-bo"   || arg == "--best-of")         { params.best_of         = std::stoi(ARGV_NEXT); }
+        else if (arg == "-bs"   || arg == "--beam-size")       { params.beam_size       = std::stoi(ARGV_NEXT); }
+        else if (arg == "-ac"   || arg == "--audio-ctx")       { params.audio_ctx       = std::stoi(ARGV_NEXT); }
+        else if (arg == "-wt"   || arg == "--word-thold")      { params.word_thold      = std::stof(ARGV_NEXT); }
+        else if (arg == "-et"   || arg == "--entropy-thold")   { params.entropy_thold   = std::stof(ARGV_NEXT); }
+        else if (arg == "-lpt"  || arg == "--logprob-thold")   { params.logprob_thold   = std::stof(ARGV_NEXT); }
+        else if (arg == "-nth"  || arg == "--no-speech-thold") { params.no_speech_thold = std::stof(ARGV_NEXT); }
+        else if (arg == "-tp"   || arg == "--temperature")     { params.temperature     = std::stof(ARGV_NEXT); }
+        else if (arg == "-tpi"  || arg == "--temperature-inc") { params.temperature_inc = std::stof(ARGV_NEXT); }
         else if (arg == "-debug"|| arg == "--debug-mode")      { params.debug_mode      = true; }
         else if (arg == "-tr"   || arg == "--translate")       { params.translate       = true; }
         else if (arg == "-di"   || arg == "--diarize")         { params.diarize         = true; }
@@ -148,30 +157,31 @@ static bool whisper_params_parse(int argc, char ** argv, whisper_params & params
         else if (arg == "-osrt" || arg == "--output-srt")      { params.output_srt      = true; }
         else if (arg == "-owts" || arg == "--output-words")    { params.output_wts      = true; }
         else if (arg == "-olrc" || arg == "--output-lrc")      { params.output_lrc      = true; }
-        else if (arg == "-fp"   || arg == "--font-path")       { params.font_path       = argv[++i]; }
+        else if (arg == "-fp"   || arg == "--font-path")       { params.font_path       = ARGV_NEXT; }
         else if (arg == "-ocsv" || arg == "--output-csv")      { params.output_csv      = true; }
         else if (arg == "-oj"   || arg == "--output-json")     { params.output_jsn      = true; }
         else if (arg == "-ojf"  || arg == "--output-json-full"){ params.output_jsn_full = params.output_jsn = true; }
-        else if (arg == "-of"   || arg == "--output-file")     { params.fname_out.emplace_back(argv[++i]); }
+        else if (arg == "-of"   || arg == "--output-file")     { params.fname_out.emplace_back(ARGV_NEXT); }
         else if (arg == "-np"   || arg == "--no-prints")       { params.no_prints       = true; }
         else if (arg == "-ps"   || arg == "--print-special")   { params.print_special   = true; }
         else if (arg == "-pc"   || arg == "--print-colors")    { params.print_colors    = true; }
         else if (arg == "-pp"   || arg == "--print-progress")  { params.print_progress  = true; }
         else if (arg == "-nt"   || arg == "--no-timestamps")   { params.no_timestamps   = true; }
-        else if (arg == "-l"    || arg == "--language")        { params.language        = whisper_param_turn_lowercase(argv[++i]); }
+        else if (arg == "-l"    || arg == "--language")        { params.language        = whisper_param_turn_lowercase(ARGV_NEXT); }
         else if (arg == "-dl"   || arg == "--detect-language") { params.detect_language = true; }
-        else if (                  arg == "--prompt")          { params.prompt          = argv[++i]; }
-        else if (arg == "-m"    || arg == "--model")           { params.model           = argv[++i]; }
-        else if (arg == "-f"    || arg == "--file")            { params.fname_inp.emplace_back(argv[++i]); }
-        else if (arg == "-oved" || arg == "--ov-e-device")     { params.openvino_encode_device = argv[++i]; }
-        else if (arg == "-dtw"  || arg == "--dtw")             { params.dtw             = argv[++i]; }
+        else if (                  arg == "--prompt")          { params.prompt          = ARGV_NEXT; }
+        else if (arg == "-m"    || arg == "--model")           { params.model           = ARGV_NEXT; }
+        else if (arg == "-f"    || arg == "--file")            { params.fname_inp.emplace_back(ARGV_NEXT); }
+        else if (arg == "-oved" || arg == "--ov-e-device")     { params.openvino_encode_device = ARGV_NEXT; }
+        else if (arg == "-dtw"  || arg == "--dtw")             { params.dtw             = ARGV_NEXT; }
         else if (arg == "-ls"   || arg == "--log-score")       { params.log_score       = true; }
         else if (arg == "-ng"   || arg == "--no-gpu")          { params.use_gpu         = false; }
         else if (arg == "-fa"   || arg == "--flash-attn")      { params.flash_attn      = true; }
-        else if (                  arg == "--suppress-regex")  { params.suppress_regex  = argv[++i]; }
-        else if (                  arg == "--grammar")         { params.grammar         = argv[++i]; }
-        else if (                  arg == "--grammar-rule")    { params.grammar_rule    = argv[++i]; }
-        else if (                  arg == "--grammar-penalty") { params.grammar_penalty = std::stof(argv[++i]); }
+        else if (arg == "-sns"  || arg == "--suppress-nst")    { params.suppress_nst    = true; }
+        else if (                  arg == "--suppress-regex")  { params.suppress_regex  = ARGV_NEXT; }
+        else if (                  arg == "--grammar")         { params.grammar         = ARGV_NEXT; }
+        else if (                  arg == "--grammar-rule")    { params.grammar_rule    = ARGV_NEXT; }
+        else if (                  arg == "--grammar-penalty") { params.grammar_penalty = std::stof(ARGV_NEXT); }
         else {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
             whisper_print_usage(argc, argv, params);
@@ -202,6 +212,7 @@ static void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params
     fprintf(stderr, "  -wt N,     --word-thold N      [%-7.2f] word timestamp probability threshold\n",         params.word_thold);
     fprintf(stderr, "  -et N,     --entropy-thold N   [%-7.2f] entropy threshold for decoder fail\n",           params.entropy_thold);
     fprintf(stderr, "  -lpt N,    --logprob-thold N   [%-7.2f] log probability threshold for decoder fail\n",   params.logprob_thold);
+    fprintf(stderr, "  -nth N,    --no-speech-thold N [%-7.2f] no speech threshold\n",                          params.no_speech_thold);
     fprintf(stderr, "  -tp,       --temperature N     [%-7.2f] The sampling temperature, between 0 and 1\n",    params.temperature);
     fprintf(stderr, "  -tpi,      --temperature-inc N [%-7.2f] The increment of temperature, between 0 and 1\n",params.temperature_inc);
     fprintf(stderr, "  -debug,    --debug-mode        [%-7s] enable debug mode (eg. dump log_mel)\n",           params.debug_mode ? "true" : "false");
@@ -234,6 +245,7 @@ static void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params
     fprintf(stderr, "  -ls,       --log-score         [%-7s] log best decoder scores of tokens\n",              params.log_score?"true":"false");
     fprintf(stderr, "  -ng,       --no-gpu            [%-7s] disable GPU\n",                                    params.use_gpu ? "false" : "true");
     fprintf(stderr, "  -fa,       --flash-attn        [%-7s] flash attention\n",                                params.flash_attn ? "true" : "false");
+    fprintf(stderr, "  -sns,      --suppress-nst      [%-7s] suppress non-speech tokens\n",                     params.suppress_nst ? "true" : "false");
     fprintf(stderr, "  --suppress-regex REGEX         [%-7s] regular expression matching tokens to suppress\n", params.suppress_regex.c_str());
     fprintf(stderr, "  --grammar GRAMMAR              [%-7s] GBNF grammar to guide decoding\n",                 params.grammar.c_str());
     fprintf(stderr, "  --grammar-rule RULE            [%-7s] top-level GBNF grammar rule name\n",               params.grammar_rule.c_str());
@@ -997,6 +1009,7 @@ int main(int argc, char ** argv) {
         if (params.dtw == "large.v1")  cparams.dtw_aheads_preset = WHISPER_AHEADS_LARGE_V1;
         if (params.dtw == "large.v2")  cparams.dtw_aheads_preset = WHISPER_AHEADS_LARGE_V2;
         if (params.dtw == "large.v3")  cparams.dtw_aheads_preset = WHISPER_AHEADS_LARGE_V3;
+        if (params.dtw == "large.v3.turbo")  cparams.dtw_aheads_preset = WHISPER_AHEADS_LARGE_V3_TURBO;
 
         if (cparams.dtw_aheads_preset == WHISPER_AHEADS_NONE) {
             fprintf(stderr, "error: unknown DTW preset '%s'\n", params.dtw.c_str());
@@ -1120,9 +1133,12 @@ int main(int argc, char ** argv) {
 
             wparams.entropy_thold    = params.entropy_thold;
             wparams.logprob_thold    = params.logprob_thold;
+            wparams.no_speech_thold  = params.no_speech_thold;
 
             wparams.no_timestamps    = params.no_timestamps;
 
+            wparams.suppress_nst     = params.suppress_nst;
+
             whisper_print_user_data user_data = { &params, &pcmf32s, 0 };
 
             const auto & grammar_parsed = params.grammar_parsed;

examples/command/CMakeLists.txt

@@ -1,9 +1,10 @@
 if (WHISPER_SDL2)
-    # command
-    set(TARGET command)
+    set(TARGET whisper-command)
     add_executable(${TARGET} command.cpp)
 
     include(DefaultTargetOptions)
 
     target_link_libraries(${TARGET} PRIVATE common common-sdl whisper ${CMAKE_THREAD_LIBS_INIT})
+
+    install(TARGETS ${TARGET} RUNTIME)
 endif ()

examples/command/README.md

@@ -1,14 +1,14 @@
-# command
+# whisper.cpp/examples/command
 
 This is a basic Voice Assistant example that accepts voice commands from the microphone.
 More info is available in [issue #171](https://github.com/ggerganov/whisper.cpp/issues/171).
 
 ```bash
 # Run with default arguments and small model
-./command -m ./models/ggml-small.en.bin -t 8
+./whisper-command -m ./models/ggml-small.en.bin -t 8
 
 # On Raspberry Pi, use tiny or base models + "-ac 768" for better performance
-./command -m ./models/ggml-tiny.en.bin -ac 768 -t 3 -c 0
+./whisper-command -m ./models/ggml-tiny.en.bin -ac 768 -t 3 -c 0
 ```
 
 https://user-images.githubusercontent.com/1991296/204038393-2f846eae-c255-4099-a76d-5735c25c49da.mp4
@@ -23,10 +23,10 @@ Initial tests show that this approach might be extremely efficient in terms of p
 
 ```bash
 # Run in guided mode, the list of allowed commands is in commands.txt
-./command -m ./models/ggml-base.en.bin -cmd ./examples/command/commands.txt
+./whisper-command -m ./models/ggml-base.en.bin -cmd ./examples/command/commands.txt
 
 # On Raspberry Pi, in guided mode you can use "-ac 128" for extra performance
-./command -m ./models/ggml-tiny.en.bin -cmd ./examples/command/commands.txt -ac 128 -t 3 -c 0
+./whisper-command -m ./models/ggml-tiny.en.bin -cmd ./examples/command/commands.txt -ac 128 -t 3 -c 0
 ```
 
 https://user-images.githubusercontent.com/1991296/207435352-8fc4ed3f-bde5-4555-9b8b-aeeb76bee969.mp4
@@ -34,7 +34,7 @@ https://user-images.githubusercontent.com/1991296/207435352-8fc4ed3f-bde5-4555-9
 
 ## Building
 
-The `command` tool depends on SDL2 library to capture audio from the microphone. You can build it like this:
+The `whisper-command` tool depends on SDL2 library to capture audio from the microphone. You can build it like this:
 
 ```bash
 # Install SDL2
@@ -47,5 +47,6 @@ sudo dnf install SDL2 SDL2-devel
 # Install SDL2 on Mac OS
 brew install sdl2
 
-make command
+cmake -B build -DWHISPER_SDL2=ON
+cmake --build build --config Release
 ```

examples/common-ggml.cpp

@@ -72,9 +72,6 @@ bool ggml_common_quantize_0(
         case GGML_FTYPE_MOSTLY_IQ4_XS:
         case GGML_FTYPE_MOSTLY_IQ1_M:
         case GGML_FTYPE_MOSTLY_BF16:
-        case GGML_FTYPE_MOSTLY_Q4_0_4_4:
-        case GGML_FTYPE_MOSTLY_Q4_0_4_8:
-        case GGML_FTYPE_MOSTLY_Q4_0_8_8:
                 {
                     fprintf(stderr, "%s: invalid model type %d\n", __func__, ftype);
                     return false;
@@ -212,9 +209,6 @@ bool ggml_common_quantize_0(
                 case GGML_TYPE_IQ4_XS:
                 case GGML_TYPE_IQ1_M:
                 case GGML_TYPE_BF16:
-                case GGML_TYPE_Q4_0_4_4:
-                case GGML_TYPE_Q4_0_4_8:
-                case GGML_TYPE_Q4_0_8_8:
                 case GGML_TYPE_TQ1_0:
                 case GGML_TYPE_TQ2_0:
                 case GGML_TYPE_COUNT:

examples/common-sdl.cpp

@@ -1,5 +1,7 @@
 #include "common-sdl.h"
 
+#include <cstdio>
+
 audio_async::audio_async(int len_ms) {
     m_len_ms = len_ms;
 

examples/deprecation-warning/CMakeLists.txt

@@ -0,0 +1,4 @@
+add_executable(main ./deprecation-warning.cpp)
+add_executable(bench ./deprecation-warning.cpp)
+add_executable(stream ./deprecation-warning.cpp)
+add_executable(command ./deprecation-warning.cpp)

examples/deprecation-warning/README.md

@@ -0,0 +1,17 @@
+# Migration notice for binary filenames
+
+> [!IMPORTANT]
+[2024 Dec 20] Binaries have been renamed w/ a `whisper-` prefix. `main` is now `whisper-cli`, `server` is `whisper-server`, etc (https://github.com/ggerganov/whisper.cpp/pull/2648)
+
+This migration was important, but it is a breaking change that may not always be immediately obvious to users.
+
+Please update all scripts and workflows to use the new binary names.
+
+| Old Filename | New Filename |
+| ---- | ---- |
+| main | whisper-cli |
+| bench | whisper-bench |
+| stream | whisper-stream |
+| command | whisper-command |
+| server | whisper-server |
+| talk-llama | whisper-talk-llama |

examples/deprecation-warning/deprecation-warning.cpp

@@ -0,0 +1,38 @@
+// Warns users that this filename was deprecated, and provides a link for more information.
+
+#include <cstdio>
+#include <string>
+
+// Main
+int main(int argc, char** argv) {
+    std::string filename = "main";
+    if (argc >= 1) {
+        filename = argv[0];
+    }
+
+    // Get only the program name from the full path
+    size_t pos = filename.find_last_of("/\\");
+    if (pos != std::string::npos) {
+        filename = filename.substr(pos+1);
+    }
+
+    // Append "whisper-" to the beginning of filename to get the replacemnt filename
+    std::string replacement_filename = "whisper-" + filename;
+
+    // The exception is if the filename is "main", then our replacement filename is "whisper-cli"
+    if (filename == "main") {
+        replacement_filename = "whisper-cli";
+    }
+
+    if (filename == "main.exe") {
+        replacement_filename = "whisper-cli.exe";
+    }
+
+    fprintf(stdout, "\n");
+    fprintf(stdout, "WARNING: The binary '%s' is deprecated.\n", filename.c_str());
+    fprintf(stdout, " Please use '%s' instead.\n", replacement_filename.c_str());
+    fprintf(stdout, " See https://github.com/ggerganov/whisper.cpp/tree/master/examples/deprecation-warning/README.md for more information.\n");
+    fprintf(stdout, "\n");
+
+    return EXIT_FAILURE;
+}

examples/ffmpeg-transcode.cpp

@@ -204,8 +204,6 @@ static int decode_audio(struct audio_buffer *audio_buf, s16 **data, int *size)
     const size_t errbuffsize = 1024;
     char errbuff[errbuffsize];
 
-    av_register_all(); // from avformat. Still a must-have call for ffmpeg v3! (can be skipped for later versions)
-
     fmt_ctx = avformat_alloc_context();
     avio_ctx_buffer = (u8*)av_malloc(AVIO_CTX_BUF_SZ);
     LOG("Creating an avio context: AVIO_CTX_BUF_SZ=%d\n", AVIO_CTX_BUF_SZ);

examples/generate-karaoke.sh

@@ -11,7 +11,7 @@
 # Press Ctrl+C to stop recording
 #
 
-executable="./main"
+executable="./build/bin/whisper-cli"
 model="base.en"
 model_path="models/ggml-$model.bin"
 
@@ -46,7 +46,7 @@ ffmpeg -y -i ./rec.wav -ar 16000 -ac 1 -c:a pcm_s16le ./rec16.wav > /dev/null 2>
 
 # run Whisper
 echo "Processing ..."
-./main -m models/ggml-base.en.bin rec16.wav -owts > /dev/null 2>&1
+${executable} -m models/ggml-base.en.bin rec16.wav -owts > /dev/null 2>&1
 
 # generate Karaoke video
 echo "Generating video ..."

examples/livestream.sh

@@ -14,7 +14,7 @@ model="base.en"
 
 check_requirements()
 {
-    if ! command -v ./main &>/dev/null; then
+    if ! command -v ./build/bin/whisper-cli &>/dev/null; then
         echo "whisper.cpp main executable is required (make)"
         exit 1
     fi
@@ -100,7 +100,7 @@ while [ $running -eq 1 ]; do
         err=$(cat /tmp/whisper-live.err | wc -l)
     done
 
-    ./main -t 8 -m ./models/ggml-${model}.bin -f /tmp/whisper-live.wav --no-timestamps -otxt 2> /tmp/whispererr | tail -n 1
+    ./build/bin/whisper-cli -t 8 -m ./models/ggml-${model}.bin -f /tmp/whisper-live.wav --no-timestamps -otxt 2> /tmp/whispererr | tail -n 1
 
     while [ $SECONDS -lt $((($i+1)*$step_s)) ]; do
         sleep 1
@@ -109,4 +109,4 @@ while [ $running -eq 1 ]; do
 done
 
 killall -v ffmpeg
-killall -v main
+killall -v whisper-cli

examples/lsp/lsp.cpp

@@ -181,7 +181,7 @@ static json unguided_transcription(struct whisper_context * ctx, audio_async &au
     wparams.n_threads        = params.n_threads;
 
     wparams.audio_ctx        = params.audio_ctx;
-    wparams.suppress_non_speech_tokens = true;
+    wparams.suppress_nst     = true;
     // run the transformer and a single decoding pass
     if (whisper_full(ctx, wparams, pcmf32.data(), pcmf32.size()) != 0) {
         fprintf(stderr, "%s: ERROR: whisper_full() failed\n", __func__);
@@ -225,7 +225,7 @@ static json guided_transcription(struct whisper_context * ctx, audio_async &audi
     wparams.prompt_tokens    = cs.prompt_tokens.data();
     wparams.prompt_n_tokens  = cs.prompt_tokens.size();
     // TODO: properly expose as option
-    wparams.suppress_non_speech_tokens = true;
+    wparams.suppress_nst     = true;
 
     // run the transformer and a single decoding pass
     if (whisper_full(ctx, wparams, pcmf32.data(), pcmf32.size()) != 0) {

examples/server/CMakeLists.txt

@@ -1,4 +1,4 @@
-set(TARGET server)
+set(TARGET whisper-server)
 add_executable(${TARGET} server.cpp httplib.h)
 
 include(DefaultTargetOptions)
@@ -8,3 +8,5 @@ target_link_libraries(${TARGET} PRIVATE common json_cpp whisper ${CMAKE_THREAD_L
 if (WIN32)
     target_link_libraries(${TARGET} PRIVATE ws2_32)
 endif()
+
+install(TARGETS ${TARGET} RUNTIME)

examples/server/README.md

@@ -1,4 +1,4 @@
-# whisper.cpp http server
+# whisper.cpp/examples/server
 
 Simple http server. WAV Files are passed to the inference model via http requests.
 
@@ -7,9 +7,9 @@ https://github.com/ggerganov/whisper.cpp/assets/1991296/e983ee53-8741-4eb5-9048-
 ## Usage
 
 ```
-./server -h
+./build/bin/whisper-server -h
 
-usage: ./bin/server [options]
+usage: ./build/bin/whisper-server [options]
 
 options:
   -h,        --help              [default] show this help message and exit

examples/server/server.cpp

@@ -61,6 +61,7 @@ struct whisper_params {
     float logprob_thold   = -1.00f;
     float temperature     =  0.00f;
     float temperature_inc =  0.20f;
+    float no_speech_thold = 0.6f;
 
     bool debug_mode      = false;
     bool translate       = false;
@@ -76,6 +77,7 @@ struct whisper_params {
     bool no_timestamps   = false;
     bool use_gpu         = true;
     bool flash_attn      = false;
+    bool suppress_nst    = false;
 
     std::string language        = "en";
     std::string prompt          = "";
@@ -134,7 +136,9 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     fprintf(stderr, "  --public PATH,                 [%-7s] Path to the public folder\n", sparams.public_path.c_str());
     fprintf(stderr, "  --request-path PATH,           [%-7s] Request path for all requests\n", sparams.request_path.c_str());
     fprintf(stderr, "  --inference-path PATH,         [%-7s] Inference path for all requests\n", sparams.inference_path.c_str());
-    fprintf(stderr, "  --convert,                     [%-7s] Convert audio to WAV, requires ffmpeg on the server", sparams.ffmpeg_converter ? "true" : "false");
+    fprintf(stderr, "  --convert,                     [%-7s] Convert audio to WAV, requires ffmpeg on the server\n", sparams.ffmpeg_converter ? "true" : "false");
+    fprintf(stderr, "  -sns,      --suppress-nst      [%-7s] suppress non-speech tokens\n", params.suppress_nst ? "true" : "false");
+    fprintf(stderr, "  -nth N,    --no-speech-thold N [%-7.2f] no speech threshold\n",   params.no_speech_thold);
     fprintf(stderr, "\n");
 }
 
@@ -179,6 +183,9 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params, serve
         else if (arg == "-dtw"  || arg == "--dtw")             { params.dtw             = argv[++i]; }
         else if (arg == "-ng"   || arg == "--no-gpu")          { params.use_gpu         = false; }
         else if (arg == "-fa"   || arg == "--flash-attn")      { params.flash_attn      = true; }
+        else if (arg == "-sns"  || arg == "--suppress-nst")    { params.suppress_nst    = true; }
+        else if (arg == "-nth"  || arg == "--no-speech-thold") { params.no_speech_thold = std::stof(argv[++i]); }
+
         // server params
         else if (                  arg == "--port")            { sparams.port        = std::stoi(argv[++i]); }
         else if (                  arg == "--host")            { sparams.hostname    = argv[++i]; }
@@ -472,6 +479,14 @@ void get_req_parameters(const Request & req, whisper_params & params)
     {
         params.temperature_inc = std::stof(req.get_file_value("temperature_inc").content);
     }
+    if (req.has_file("suppress_non_speech"))
+    {
+        params.suppress_nst = parse_str_to_bool(req.get_file_value("suppress_non_speech").content);
+    }
+    if (req.has_file("suppress_nst"))
+    {
+        params.suppress_nst = parse_str_to_bool(req.get_file_value("suppress_nst").content);
+    }
 }
 
 }  // namespace
@@ -677,7 +692,8 @@ int main(int argc, char ** argv) {
         if (sparams.ffmpeg_converter) {
             // if file is not wav, convert to wav
             // write to temporary file
-            const std::string temp_filename_base = std::tmpnam(nullptr);
+            //const std::string temp_filename_base = std::tmpnam(nullptr);
+            const std::string temp_filename_base = "whisper-server-tmp"; // TODO: this is a hack, remove when the mutext is removed
             const std::string temp_filename = temp_filename_base + ".wav";
             std::ofstream temp_file{temp_filename, std::ios::binary};
             temp_file << audio_file.content;
@@ -711,7 +727,6 @@ int main(int argc, char ** argv) {
             }
         }
 
-
         printf("Successfully loaded %s\n", filename.c_str());
 
         // print system information
@@ -779,6 +794,7 @@ int main(int argc, char ** argv) {
             wparams.beam_search.beam_size = params.beam_size;
 
             wparams.temperature      = params.temperature;
+            wparams.no_speech_thold = params.no_speech_thold;
             wparams.temperature_inc  = params.temperature_inc;
             wparams.entropy_thold    = params.entropy_thold;
             wparams.logprob_thold    = params.logprob_thold;
@@ -786,6 +802,8 @@ int main(int argc, char ** argv) {
             wparams.no_timestamps    = params.no_timestamps;
             wparams.token_timestamps = !params.no_timestamps && params.response_format == vjson_format;
 
+            wparams.suppress_nst     = params.suppress_nst;
+
             whisper_print_user_data user_data = { &params, &pcmf32s, 0 };
 
             // this callback is called on each new segment
@@ -929,7 +947,7 @@ int main(int argc, char ** argv) {
 
                 // TODO compression_ratio and no_speech_prob are not implemented yet
                 // segment["compression_ratio"] = 0;
-                // segment["no_speech_prob"] = 0;
+                segment["no_speech_prob"] = whisper_full_get_segment_no_speech_prob(ctx, i);
 
                 jres["segments"].push_back(segment);
             }

examples/stream/CMakeLists.txt

@@ -1,9 +1,10 @@
 if (WHISPER_SDL2)
-    # stream
-    set(TARGET stream)
+    set(TARGET whisper-stream)
     add_executable(${TARGET} stream.cpp)
 
     include(DefaultTargetOptions)
 
     target_link_libraries(${TARGET} PRIVATE common common-sdl whisper ${CMAKE_THREAD_LIBS_INIT})
+
+    install(TARGETS ${TARGET} RUNTIME)
 endif ()

examples/stream/README.md

@@ -1,11 +1,11 @@
-# stream
+# whisper.cpp/examples/stream
 
 This is a naive example of performing real-time inference on audio from your microphone.
-The `stream` tool samples the audio every half a second and runs the transcription continously.
+The `whisper-stream` tool samples the audio every half a second and runs the transcription continously.
 More info is available in [issue #10](https://github.com/ggerganov/whisper.cpp/issues/10).
 
 ```bash
-./stream -m ./models/ggml-base.en.bin -t 8 --step 500 --length 5000
+./build/bin/whisper-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
@@ -15,7 +15,7 @@ https://user-images.githubusercontent.com/1991296/194935793-76afede7-cfa8-48d8-a
 Setting the `--step` argument to `0` enables the sliding window mode:
 
 ```bash
- ./stream -m ./models/ggml-small.en.bin -t 6 --step 0 --length 30000 -vth 0.6
+ ./build/bin/whisper-stream -m ./models/ggml-base.en.bin -t 6 --step 0 --length 30000 -vth 0.6
 ```
 
 In this mode, the tool will transcribe only after some speech activity is detected. A very
@@ -27,7 +27,7 @@ a transcription block that is suitable for parsing.
 
 ## Building
 
-The `stream` tool depends on SDL2 library to capture audio from the microphone. You can build it like this:
+The `whisper-stream` tool depends on SDL2 library to capture audio from the microphone. You can build it like this:
 
 ```bash
 # Install SDL2
@@ -40,21 +40,10 @@ sudo dnf install SDL2 SDL2-devel
 # Install SDL2 on Mac OS
 brew install sdl2
 
-make stream
-```
+cmake -B build -DWHISPER_SDL2=ON
+cmake --build build --config Release
 
-Ensure you are at the root of the repo when running `make stream`. Not within the `examples/stream` dir
-as the libraries needed like `common-sdl.h` are located within `examples`. Attempting to compile within
-`examples/steam` means your compiler cannot find them and it gives an error it cannot find the file.
-
-```bash
-whisper.cpp/examples/stream$ make stream
-g++     stream.cpp   -o stream
-stream.cpp:6:10: fatal error: common/sdl.h: No such file or directory
-    6 | #include "common/sdl.h"
-      |          ^~~~~~~~~~~~~~
-compilation terminated.
-make: *** [<builtin>: stream] Error 1
+./build/bin/whisper-stream
 ```
 
 ## Web version

examples/sycl/CMakeLists.txt

@@ -5,5 +5,5 @@
 set(TARGET ls-sycl-device)
 add_executable(${TARGET} ls-sycl-device.cpp)
 install(TARGETS ${TARGET} RUNTIME)
-target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_17)
+target_link_libraries(${TARGET} PRIVATE common whisper ${CMAKE_THREAD_LIBS_INIT})
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/sycl/build.sh

@@ -7,13 +7,16 @@ 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 # faster for long-prompt inference
+#cmake .. -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DWHISPER_SYCL_F16=ON # faster for long-prompt inference
 
 #for FP32
-cmake .. -DWHISPER_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx
+cmake .. -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx
+
+#for other features from the examples, e.g. stream and talk link with SDL2:
+#cmake .. -DGGML_SYCL=ON -DWHISPER_SDL2=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
+cmake --build . --config Release -v

examples/talk-llama/CMakeLists.txt

@@ -1,6 +1,5 @@
 if (WHISPER_SDL2)
-    # talk-llama
-    set(TARGET talk-llama)
+    set(TARGET whisper-talk-llama)
     add_executable(${TARGET} talk-llama.cpp
         llama.cpp
         llama-vocab.cpp

examples/talk-llama/README.md

@@ -1,4 +1,4 @@
-# talk-llama
+# whisper.cpp/examples/talk-llama
 
 Talk with an LLaMA AI in your terminal
 
@@ -12,7 +12,7 @@ https://github.com/ggerganov/whisper.cpp/assets/1991296/d97a3788-bf2a-4756-9a43-
 
 ## Building
 
-The `talk-llama` tool depends on SDL2 library to capture audio from the microphone. You can build it like this:
+The `whisper-talk-llama` tool depends on SDL2 library to capture audio from the microphone. You can build it like this:
 
 ```bash
 # Install SDL2
@@ -25,11 +25,12 @@ sudo dnf install SDL2 SDL2-devel
 # Install SDL2 on Mac OS
 brew install sdl2
 
-# Build the "talk-llama" executable
-make talk-llama
+# Build the "whisper-talk-llama" executable
+cmake -B build -S . -DWHISPER_SDL2=ON
+cmake --build build --config Release
 
 # Run it
-./talk-llama -mw ./models/ggml-small.en.bin -ml ../llama.cpp/models/llama-13b/ggml-model-q4_0.gguf -p "Georgi" -t 8
+./build/bin/whisper-talk-llama -mw ./models/ggml-small.en.bin -ml ../llama.cpp/models/llama-13b/ggml-model-q4_0.gguf -p "Georgi" -t 8
 ```
 
 - The `-mw` argument specifies the Whisper model that you would like to use. Recommended `base` or `small` for real-time experience
@@ -37,16 +38,16 @@ make talk-llama
 
 ## Session
 
-The `talk-llama` tool supports session management to enable more coherent and continuous conversations. By maintaining context from previous interactions, it can better understand and respond to user requests in a more natural way.
+The `whisper-talk-llama` tool supports session management to enable more coherent and continuous conversations. By maintaining context from previous interactions, it can better understand and respond to user requests in a more natural way.
 
-To enable session support, use the `--session FILE` command line option when running the program. The `talk-llama` model state will be saved to the specified file after each interaction. If the file does not exist, it will be created. If the file exists, the model state will be loaded from it, allowing you to resume a previous session.
+To enable session support, use the `--session FILE` command line option when running the program. The `whisper-talk-llama` model state will be saved to the specified file after each interaction. If the file does not exist, it will be created. If the file exists, the model state will be loaded from it, allowing you to resume a previous session.
 
 This feature is especially helpful for maintaining context in long conversations or when interacting with the AI assistant across multiple sessions. It ensures that the assistant remembers the previous interactions and can provide more relevant and contextual responses.
 
 Example usage:
 
 ```bash
-./talk-llama --session ./my-session-file -mw ./models/ggml-small.en.bin -ml ../llama.cpp/models/llama-13b/ggml-model-q4_0.gguf -p "Georgi" -t 8
+./build/bin/whisper-talk-llama --session ./my-session-file -mw ./models/ggml-small.en.bin -ml ../llama.cpp/models/llama-13b/ggml-model-q4_0.gguf -p "Georgi" -t 8
 ```
 
 ## TTS

examples/talk-llama/llama-sampling.h

@@ -4,8 +4,6 @@
 
 #include "llama-grammar.h"
 
-#include <unordered_map>
-
 struct llama_vocab;
 struct llama_grammar;
 
@@ -27,3 +25,24 @@ struct llama_sampler * llama_sampler_init_grammar_impl(
         const struct llama_vocab & vocab,
                       const char * grammar_str,
                       const char * grammar_root);
+
+struct llama_sampler * llama_sampler_init_infill_impl(
+        const struct llama_vocab & vocab);
+
+struct llama_sampler * llama_sampler_init_dry_impl(
+        const struct llama_vocab &  vocab,
+                         int32_t    context_size,
+                           float    dry_multiplier,
+                           float    dry_base,
+                         int32_t    dry_allowed_length,
+                         int32_t    dry_penalty_last_n,
+                      const char ** seq_breakers,
+                          size_t    num_breakers);
+
+struct llama_sampler * llama_sampler_init_dry_testing(
+                         int32_t   context_size,
+                           float   dry_multiplier,
+                           float   dry_base,
+                         int32_t   dry_allowed_length,
+                         int32_t   dry_penalty_last_n,
+  const std::vector<std::vector<llama_token>>& seq_breakers);

examples/talk-llama/llama-vocab.cpp

@@ -221,7 +221,7 @@ struct llm_tokenizer_spm_session {
         }
 
         // seed the work queue with all possible 2-character tokens.
-        for (size_t i = 1; i < symbols.size(); ++i) {
+        for (int i = 1; i < (int) symbols.size(); ++i) {
             try_add_bigram(i - 1, i);
         }
 
@@ -418,6 +418,7 @@ struct llm_tokenizer_bpe : llm_tokenizer {
             case LLAMA_VOCAB_PRE_TYPE_SMOLLM:
             case LLAMA_VOCAB_PRE_TYPE_CODESHELL:
             case LLAMA_VOCAB_PRE_TYPE_EXAONE:
+            case LLAMA_VOCAB_PRE_TYPE_MINERVA:
                 regex_exprs = {
                     "\\p{N}",
                     "'s|'t|'re|'ve|'m|'ll|'d| ?\\p{L}+| ?\\p{N}+| ?[^\\s\\p{L}\\p{N}]+|\\s+(?!\\S)",
@@ -563,7 +564,7 @@ struct llm_tokenizer_bpe_session {
                 index++;
                 symbols.emplace_back(sym);
             }
-            for (size_t i = 1; i < symbols.size(); ++i) {
+            for (int i = 1; i < (int) symbols.size(); ++i) {
                 add_new_bigram(i - 1, i);
             }
 
@@ -737,7 +738,7 @@ struct llm_tokenizer_wpm_session {
         std::vector<std::string> words(1, "");
 
         for (const uint32_t cpt : cpts_nfd) {
-            const auto flags = unicode_cpt_flags(cpt);
+            const auto flags = unicode_cpt_flags_from_cpt(cpt);
 
             if (flags.is_whitespace) {
                 if (words.back().size()) {  // finish previous word if any
@@ -1663,6 +1664,14 @@ llama_token llama_token_eos_impl(const struct llama_vocab & vocab) {
     return vocab.special_eos_id;
 }
 
+llama_token llama_token_eot_impl(const struct llama_vocab & vocab) {
+    return vocab.special_eot_id;
+}
+
+llama_token llama_token_eom_impl(const struct llama_vocab & vocab) {
+    return vocab.special_eom_id;
+}
+
 llama_token llama_token_cls_impl(const struct llama_vocab & vocab) {
     return vocab.special_cls_id;
 }
@@ -1688,23 +1697,39 @@ bool llama_add_eos_token_impl(const struct llama_vocab & vocab) {
 }
 
 llama_token llama_token_prefix_impl(const struct llama_vocab & vocab) {
-    return vocab.special_prefix_id;
+    return vocab.special_fim_pre_id;
 }
 
 llama_token llama_token_middle_impl(const struct llama_vocab & vocab) {
-    return vocab.special_middle_id;
+    return vocab.special_fim_mid_id;
 }
 
 llama_token llama_token_suffix_impl(const struct llama_vocab & vocab) {
-    return vocab.special_suffix_id;
+    return vocab.special_fim_suf_id;
 }
 
-llama_token llama_token_eot_impl(const struct llama_vocab & vocab) {
-    return vocab.special_eot_id;
+llama_token llama_token_fim_pre_impl(const struct llama_vocab & vocab) {
+    return vocab.special_fim_pre_id;
 }
 
-llama_token llama_token_eom_impl(const struct llama_vocab & vocab) {
-    return vocab.special_eom_id;
+llama_token llama_token_fim_suf_impl(const struct llama_vocab & vocab) {
+    return vocab.special_fim_suf_id;
+}
+
+llama_token llama_token_fim_mid_impl(const struct llama_vocab & vocab) {
+    return vocab.special_fim_mid_id;
+}
+
+llama_token llama_token_fim_pad_impl(const struct llama_vocab & vocab) {
+    return vocab.special_fim_pad_id;
+}
+
+llama_token llama_token_fim_rep_impl(const struct llama_vocab & vocab) {
+    return vocab.special_fim_rep_id;
+}
+
+llama_token llama_token_fim_sep_impl(const struct llama_vocab & vocab) {
+    return vocab.special_fim_sep_id;
 }
 
 int32_t llama_tokenize_impl(
@@ -1942,3 +1967,19 @@ int32_t llama_detokenize_impl(
 
     return total <= text_len_max ? total : -total;
 }
+
+std::string llama_detokenize(const struct llama_vocab & vocab, const std::vector<llama_token> & tokens, bool special) {
+    std::string text;
+    text.resize(std::max(text.capacity(), tokens.size()));
+    int32_t n_chars = llama_detokenize_impl(vocab, tokens.data(), (int32_t)tokens.size(), &text[0], (int32_t)text.size(), false, special);
+    if (n_chars < 0) {
+        text.resize(-n_chars);
+        n_chars = llama_detokenize_impl(vocab, tokens.data(), (int32_t)tokens.size(), &text[0], (int32_t)text.size(), false, special);
+        GGML_ASSERT(n_chars <= (int32_t)text.size());  // whitespace trimming is performed after per-token detokenization
+    }
+
+    text.resize(n_chars);
+
+    // NOTE: the original tokenizer decodes bytes after collecting the pieces.
+    return text;
+}

examples/talk-llama/llama-vocab.h

@@ -37,20 +37,26 @@ struct llama_vocab {
     std::map<std::pair<std::string, std::string>, int> bpe_ranks;
 
     // default LLaMA special tokens
+    // TODO: should we set all of these to LLAMA_TOKEN_NULL?
     id special_bos_id  = 1;
     id special_eos_id  = 2;
+    id special_eot_id  = LLAMA_TOKEN_NULL;
+    id special_eom_id  = LLAMA_TOKEN_NULL;
     id special_unk_id  = 0;
-    id special_sep_id  = -1;
-    id special_pad_id  = -1;
-    id special_cls_id  = -1;
-    id special_mask_id = -1;
+    id special_sep_id  = LLAMA_TOKEN_NULL;
+    id special_pad_id  = LLAMA_TOKEN_NULL;
+    id special_cls_id  = LLAMA_TOKEN_NULL;
+    id special_mask_id = LLAMA_TOKEN_NULL;
 
-    id linefeed_id       = 13;
-    id special_prefix_id = -1;
-    id special_suffix_id = -1;
-    id special_middle_id = -1;
-    id special_eot_id    = -1; // TODO: move above after "eos_id", and here add "file separator" token
-    id special_eom_id    = -1;
+    id linefeed_id = 13;
+
+    // fim tokens
+    id special_fim_pre_id = LLAMA_TOKEN_NULL;
+    id special_fim_suf_id = LLAMA_TOKEN_NULL;
+    id special_fim_mid_id = LLAMA_TOKEN_NULL;
+    id special_fim_pad_id = LLAMA_TOKEN_NULL;
+    id special_fim_rep_id = LLAMA_TOKEN_NULL; // repo
+    id special_fim_sep_id = LLAMA_TOKEN_NULL; // file separator
 
     // set of all tokens that cause "end of generation"
     std::set<id> special_eog_ids;
@@ -104,19 +110,26 @@ bool llama_token_is_control_impl(const struct llama_vocab & vocab, llama_token t
 
 llama_token llama_token_bos_impl(const struct llama_vocab & vocab);
 llama_token llama_token_eos_impl(const struct llama_vocab & vocab);
+llama_token llama_token_eot_impl(const struct llama_vocab & vocab);
+llama_token llama_token_eom_impl(const struct llama_vocab & vocab);
 llama_token llama_token_cls_impl(const struct llama_vocab & vocab);
 llama_token llama_token_sep_impl(const struct llama_vocab & vocab);
 llama_token llama_token_nl_impl (const struct llama_vocab & vocab);
 llama_token llama_token_pad_impl(const struct llama_vocab & vocab);
 
-bool llama_add_bos_token_impl(const struct llama_vocab & vocab);
-bool llama_add_eos_token_impl(const struct llama_vocab & vocab);
-
 llama_token llama_token_prefix_impl(const struct llama_vocab & vocab);
 llama_token llama_token_middle_impl(const struct llama_vocab & vocab);
 llama_token llama_token_suffix_impl(const struct llama_vocab & vocab);
-llama_token llama_token_eot_impl   (const struct llama_vocab & vocab);
-llama_token llama_token_eom_impl   (const struct llama_vocab & vocab);
+
+llama_token llama_token_fim_pre_impl(const struct llama_vocab & vocab);
+llama_token llama_token_fim_suf_impl(const struct llama_vocab & vocab);
+llama_token llama_token_fim_mid_impl(const struct llama_vocab & vocab);
+llama_token llama_token_fim_pad_impl(const struct llama_vocab & vocab);
+llama_token llama_token_fim_rep_impl(const struct llama_vocab & vocab);
+llama_token llama_token_fim_sep_impl(const struct llama_vocab & vocab);
+
+bool llama_add_bos_token_impl(const struct llama_vocab & vocab);
+bool llama_add_eos_token_impl(const struct llama_vocab & vocab);
 
 int32_t llama_tokenize_impl(
         const struct llama_vocab & vocab,
@@ -136,6 +149,12 @@ int32_t llama_token_to_piece_impl(
                          int32_t   lstrip,
                             bool   special);
 
+// check if token0 is contained as a prefix in token1
+bool llama_token_is_prefix_impl(
+        const struct llama_vocab & vocab,
+                     llama_token   token0,
+                     llama_token   token1);
+
 int32_t llama_detokenize_impl(
         const struct llama_vocab & vocab,
                const llama_token * tokens,
@@ -144,3 +163,8 @@ int32_t llama_detokenize_impl(
                          int32_t   text_len_max,
                             bool   remove_special,
                             bool   unparse_special);
+
+std::string llama_detokenize(
+        const struct llama_vocab & vocab,
+  const std::vector<llama_token> & tokens,
+                            bool   special);

examples/talk-llama/llama.h

@@ -2,6 +2,7 @@
 #define LLAMA_H
 
 #include "ggml.h"
+#include "ggml-cpu.h"
 #include "ggml-backend.h"
 
 #include <stddef.h>
@@ -103,12 +104,15 @@ extern "C" {
         LLAMA_VOCAB_PRE_TYPE_GPT3_FINNISH   = 24,
         LLAMA_VOCAB_PRE_TYPE_EXAONE         = 25,
         LLAMA_VOCAB_PRE_TYPE_CHAMELEON      = 26,
+        LLAMA_VOCAB_PRE_TYPE_MINERVA        = 27,
     };
 
     enum llama_rope_type {
-        LLAMA_ROPE_TYPE_NONE = -1,
-        LLAMA_ROPE_TYPE_NORM = 0,
-        LLAMA_ROPE_TYPE_NEOX = GGML_ROPE_TYPE_NEOX,
+        LLAMA_ROPE_TYPE_NONE   = -1,
+        LLAMA_ROPE_TYPE_NORM   = 0,
+        LLAMA_ROPE_TYPE_NEOX   = GGML_ROPE_TYPE_NEOX,
+        LLAMA_ROPE_TYPE_MROPE  = GGML_ROPE_TYPE_MROPE,
+        LLAMA_ROPE_TYPE_VISION = GGML_ROPE_TYPE_VISION,
     };
 
     enum llama_token_type { //TODO: remove, required until per token attributes are available from GGUF file
@@ -170,9 +174,9 @@ extern "C" {
         LLAMA_FTYPE_MOSTLY_IQ4_XS        = 30, // except 1d tensors
         LLAMA_FTYPE_MOSTLY_IQ1_M         = 31, // except 1d tensors
         LLAMA_FTYPE_MOSTLY_BF16          = 32, // except 1d tensors
-        LLAMA_FTYPE_MOSTLY_Q4_0_4_4      = 33, // except 1d tensors
-        LLAMA_FTYPE_MOSTLY_Q4_0_4_8      = 34, // except 1d tensors
-        LLAMA_FTYPE_MOSTLY_Q4_0_8_8      = 35, // except 1d tensors
+        //LLAMA_FTYPE_MOSTLY_Q4_0_4_4      = 33, // removed from gguf files, use Q4_0 and runtime repack
+        //LLAMA_FTYPE_MOSTLY_Q4_0_4_8      = 34, // removed from gguf files, use Q4_0 and runtime repack
+        //LLAMA_FTYPE_MOSTLY_Q4_0_8_8      = 35, // removed from gguf files, use Q4_0 and runtime repack
         LLAMA_FTYPE_MOSTLY_TQ1_0         = 36, // except 1d tensors
         LLAMA_FTYPE_MOSTLY_TQ2_0         = 37, // except 1d tensors
 
@@ -184,7 +188,8 @@ extern "C" {
         LLAMA_ROPE_SCALING_TYPE_NONE        = 0,
         LLAMA_ROPE_SCALING_TYPE_LINEAR      = 1,
         LLAMA_ROPE_SCALING_TYPE_YARN        = 2,
-        LLAMA_ROPE_SCALING_TYPE_MAX_VALUE   = LLAMA_ROPE_SCALING_TYPE_YARN,
+        LLAMA_ROPE_SCALING_TYPE_LONGROPE    = 3,
+        LLAMA_ROPE_SCALING_TYPE_MAX_VALUE   = LLAMA_ROPE_SCALING_TYPE_LONGROPE,
     };
 
     enum llama_pooling_type {
@@ -205,7 +210,7 @@ extern "C" {
     enum llama_split_mode {
         LLAMA_SPLIT_MODE_NONE  = 0, // single GPU
         LLAMA_SPLIT_MODE_LAYER = 1, // split layers and KV across GPUs
-        LLAMA_SPLIT_MODE_ROW   = 2, // split rows across GPUs
+        LLAMA_SPLIT_MODE_ROW   = 2, // split layers and KV across GPUs, use tensor parallelism if supported
     };
 
     // TODO: simplify (https://github.com/ggerganov/llama.cpp/pull/9294#pullrequestreview-2286561979)
@@ -217,6 +222,7 @@ extern "C" {
 
     typedef struct llama_token_data_array {
         // TODO: consider SoA
+        // NOTE: this pointer can be modified by the samplers
         llama_token_data * data;
         size_t size;
         int64_t selected; // this is the index in the data array (i.e. not the token id)
@@ -232,8 +238,11 @@ extern "C" {
     // - token  : the token ids of the input (used when embd is NULL)
     // - embd   : token embeddings (i.e. float vector of size n_embd) (used when token is NULL)
     // - pos    : the positions of the respective token in the sequence
+    //            (if set to NULL, the token position will be tracked automatically by llama_decode)
     // - seq_id : the sequence to which the respective token belongs
+    //            (if set to NULL, the sequence ID will be assumed to be 0)
     // - logits : if zero, the logits (and/or the embeddings) for the respective token will not be output
+    //            (if set to NULL, only the logits for last token will be returned)
     //
     typedef struct llama_batch {
         int32_t n_tokens;
@@ -244,15 +253,6 @@ extern "C" {
         int32_t      *  n_seq_id;
         llama_seq_id ** seq_id;
         int8_t       *  logits; // TODO: rename this to "output"
-
-        // NOTE: helpers for smooth API transition - can be deprecated in the future
-        //       for future-proof code, use the above fields instead and ignore everything below
-        //
-        // pos[i] = all_pos_0 + i*all_pos_1
-        //
-        llama_pos    all_pos_0;  // used if pos == NULL
-        llama_pos    all_pos_1;  // used if pos == NULL
-        llama_seq_id all_seq_id; // used if seq_id == NULL
     } llama_batch;
 
     enum llama_model_kv_override_type {
@@ -276,13 +276,13 @@ extern "C" {
     };
 
     struct llama_model_params {
+        // NULL-terminated list of devices to use for offloading (if NULL, all available devices are used)
+        ggml_backend_dev_t * devices;
+
         int32_t n_gpu_layers; // number of layers to store in VRAM
         enum llama_split_mode split_mode; // how to split the model across multiple GPUs
 
-        // main_gpu interpretation depends on split_mode:
-        // LLAMA_SPLIT_MODE_NONE: the GPU that is used for the entire model
-        // LLAMA_SPLIT_MODE_ROW: the GPU that is used for small tensors and intermediate results
-        // LLAMA_SPLIT_MODE_LAYER: ignored
+        // the GPU that is used for the entire model when split_mode is LLAMA_SPLIT_MODE_NONE
         int32_t main_gpu;
 
         // proportion of the model (layers or rows) to offload to each GPU, size: llama_max_devices()
@@ -433,6 +433,7 @@ extern "C" {
     LLAMA_API bool llama_supports_mmap       (void);
     LLAMA_API bool llama_supports_mlock      (void);
     LLAMA_API bool llama_supports_gpu_offload(void);
+    LLAMA_API bool llama_supports_rpc        (void);
 
     LLAMA_API uint32_t llama_n_ctx      (const struct llama_context * ctx);
     LLAMA_API uint32_t llama_n_batch    (const struct llama_context * ctx);
@@ -457,6 +458,7 @@ extern "C" {
     // Functions to access the model's GGUF metadata scalar values
     // - The functions return the length of the string on success, or -1 on failure
     // - The output string is always null-terminated and cleared on failure
+    // - When retrieving a string, an extra byte must be allocated to account for the null terminator
     // - GGUF array values are not supported by these functions
 
     // Get metadata value as a string by key name
@@ -673,6 +675,9 @@ extern "C" {
     // Apply the KV cache updates (such as K-shifts, defragmentation, etc.)
     LLAMA_API void llama_kv_cache_update(struct llama_context * ctx);
 
+    // Check if the context supports KV cache shifting
+    LLAMA_API bool llama_kv_cache_can_shift(struct llama_context * ctx);
+
     //
     // State / sessions
     //
@@ -775,15 +780,15 @@ extern "C" {
     // Decoding
     //
 
-    // Return batch for single sequence of tokens starting at pos_0
+    // Return batch for single sequence of tokens
+    // The sequence ID will be fixed to 0
+    // The position of the tokens will be tracked automatically by llama_decode
     //
     // NOTE: this is a helper function to facilitate transition to the new batch API - avoid using it
     //
     LLAMA_API struct llama_batch llama_batch_get_one(
                   llama_token * tokens,
-                      int32_t   n_tokens,
-                    llama_pos   pos_0,
-                 llama_seq_id   seq_id);
+                      int32_t   n_tokens);
 
     // Allocates a batch of tokens on the heap that can hold a maximum of n_tokens
     // Each token can be assigned up to n_seq_max sequence ids
@@ -803,7 +808,7 @@ extern "C" {
     // Processes a batch of tokens with the ecoder part of the encoder-decoder model.
     // Stores the encoder output internally for later use by the decoder cross-attention layers.
     //   0 - success
-    // < 0 - error
+    // < 0 - error. the KV cache state is restored to the state before this call
     LLAMA_API int32_t llama_encode(
             struct llama_context * ctx,
               struct llama_batch   batch);
@@ -811,7 +816,7 @@ extern "C" {
     // Positive return values does not mean a fatal error, but rather a warning.
     //   0 - success
     //   1 - could not find a KV slot for the batch (try reducing the size of the batch or increase the context)
-    // < 0 - error
+    // < 0 - error. the KV cache state is restored to the state before this call
     LLAMA_API int32_t llama_decode(
             struct llama_context * ctx,
               struct llama_batch   batch);
@@ -896,6 +901,7 @@ extern "C" {
     // Special tokens
     LLAMA_API llama_token llama_token_bos(const struct llama_model * model); // beginning-of-sentence
     LLAMA_API llama_token llama_token_eos(const struct llama_model * model); // end-of-sentence
+    LLAMA_API llama_token llama_token_eot(const struct llama_model * model); // end-of-turn
     LLAMA_API llama_token llama_token_cls(const struct llama_model * model); // classification
     LLAMA_API llama_token llama_token_sep(const struct llama_model * model); // sentence separator
     LLAMA_API llama_token llama_token_nl (const struct llama_model * model); // next-line
@@ -904,11 +910,17 @@ extern "C" {
     LLAMA_API bool llama_add_bos_token(const struct llama_model * model);
     LLAMA_API bool llama_add_eos_token(const struct llama_model * model);
 
-    // Codellama infill tokens
-    LLAMA_API llama_token llama_token_prefix(const struct llama_model * model); // Beginning of infill prefix
-    LLAMA_API llama_token llama_token_middle(const struct llama_model * model); // Beginning of infill middle
-    LLAMA_API llama_token llama_token_suffix(const struct llama_model * model); // Beginning of infill suffix
-    LLAMA_API llama_token llama_token_eot   (const struct llama_model * model); // End of infill middle
+    // infill tokens
+    DEPRECATED(LLAMA_API llama_token llama_token_prefix(const struct llama_model * model), "use llama_token_fim_pre instead");
+    DEPRECATED(LLAMA_API llama_token llama_token_middle(const struct llama_model * model), "use llama_token_fim_mid instead");
+    DEPRECATED(LLAMA_API llama_token llama_token_suffix(const struct llama_model * model), "use llama_token_fim_suf instead");
+
+    LLAMA_API llama_token llama_token_fim_pre(const struct llama_model * model);
+    LLAMA_API llama_token llama_token_fim_suf(const struct llama_model * model);
+    LLAMA_API llama_token llama_token_fim_mid(const struct llama_model * model);
+    LLAMA_API llama_token llama_token_fim_pad(const struct llama_model * model);
+    LLAMA_API llama_token llama_token_fim_rep(const struct llama_model * model);
+    LLAMA_API llama_token llama_token_fim_sep(const struct llama_model * model);
 
     //
     // Tokenization
@@ -983,6 +995,9 @@ extern "C" {
                                   char * buf,
                                int32_t   length);
 
+    // Get list of built-in chat templates
+    LLAMA_API int32_t llama_chat_builtin_templates(const char ** output, size_t len);
+
     //
     // Sampling API
     //
@@ -1067,12 +1082,13 @@ extern "C" {
 
     // available samplers:
 
-    LLAMA_API struct llama_sampler * llama_sampler_init_greedy     (void);
-    LLAMA_API struct llama_sampler * llama_sampler_init_dist       (uint32_t seed);
+    LLAMA_API struct llama_sampler * llama_sampler_init_greedy(void);
+    LLAMA_API struct llama_sampler * llama_sampler_init_dist  (uint32_t seed);
 
     /// @details Sorts candidate tokens by their logits in descending order and calculate probabilities based on logits.
     /// NOTE: Avoid using on the full vocabulary as the sorting can become slow. For example, apply top-k or top-p sampling first.
-    LLAMA_API struct llama_sampler * llama_sampler_init_softmax    (void);
+    DEPRECATED(LLAMA_API struct llama_sampler * llama_sampler_init_softmax    (void),
+        "will be removed in the future (see https://github.com/ggerganov/llama.cpp/pull/9896#discussion_r1800920915)");
 
     /// @details Top-K sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
     LLAMA_API struct llama_sampler * llama_sampler_init_top_k      (int32_t k);
@@ -1083,16 +1099,18 @@ extern "C" {
     /// @details Minimum P sampling as described in https://github.com/ggerganov/llama.cpp/pull/3841
     LLAMA_API struct llama_sampler * llama_sampler_init_min_p      (float   p, size_t min_keep);
 
-    /// @details Tail Free Sampling described in https://www.trentonbricken.com/Tail-Free-Sampling/.
-    LLAMA_API struct llama_sampler * llama_sampler_init_tail_free  (float   z, size_t min_keep);
-
     /// @details Locally Typical Sampling implementation described in the paper https://arxiv.org/abs/2202.00666.
     LLAMA_API struct llama_sampler * llama_sampler_init_typical    (float   p, size_t min_keep);
+
+    /// #details Updates the logits l_i` = l_i/t. When t <= 0.0f, the maximum logit is kept at it's original value, the rest are set to -inf
     LLAMA_API struct llama_sampler * llama_sampler_init_temp       (float   t);
 
     /// @details Dynamic temperature implementation (a.k.a. entropy) described in the paper https://arxiv.org/abs/2309.02772.
     LLAMA_API struct llama_sampler * llama_sampler_init_temp_ext   (float   t, float   delta, float exponent);
 
+    /// @details XTC sampler as described in https://github.com/oobabooga/text-generation-webui/pull/6335
+    LLAMA_API struct llama_sampler * llama_sampler_init_xtc        (float   p, float   t,     size_t min_keep, uint32_t seed);
+
     /// @details Mirostat 1.0 algorithm described in the paper https://arxiv.org/abs/2007.14966. Uses tokens instead of words.
     /// @param candidates A vector of `llama_token_data` containing the candidate tokens, their probabilities (p), and log-odds (logit) for the current position in the generated text.
     /// @param tau  The target cross-entropy (or surprise) value you want to achieve for the generated text. A higher value corresponds to more surprising or less predictable text, while a lower value corresponds to less surprising or more predictable text.
@@ -1121,22 +1139,50 @@ extern "C" {
                           const char * grammar_str,
                           const char * grammar_root);
 
+    /// NOTE: Avoid using on the full vocabulary as searching for repeated tokens can become slow. For example, apply top-k or top-p sampling first.
     LLAMA_API struct llama_sampler * llama_sampler_init_penalties(
-                             int32_t   n_vocab,         // llama_n_vocab()
-                         llama_token   special_eos_id,  // llama_token_eos()
-                         llama_token   linefeed_id,     // llama_token_nl()
-                             int32_t   penalty_last_n,  // last n tokens to penalize (0 = disable penalty, -1 = context size)
-                               float   penalty_repeat,  // 1.0 = disabled
-                               float   penalty_freq,    // 0.0 = disabled
-                               float   penalty_present, // 0.0 = disabled
-                                bool   penalize_nl,     // consider newlines as a repeatable token
-                                bool   ignore_eos);     // ignore the end-of-sequence token
+                             int32_t   penalty_last_n,   // last n tokens to penalize (0 = disable penalty, -1 = context size)
+                               float   penalty_repeat,   // 1.0 = disabled
+                               float   penalty_freq,     // 0.0 = disabled
+                               float   penalty_present); // 0.0 = disabled
+
+    ///  @details DRY sampler, designed by p-e-w, as described in: https://github.com/oobabooga/text-generation-webui/pull/5677, porting Koboldcpp implementation authored by pi6am: https://github.com/LostRuins/koboldcpp/pull/982
+    LLAMA_API struct llama_sampler *    llama_sampler_init_dry(
+            const struct llama_model *  model,
+                               float    dry_multiplier,
+                               float    dry_base,
+                             int32_t    dry_allowed_length,
+                             int32_t    dry_penalty_last_n,
+                          const char ** seq_breakers,
+                              size_t    num_breakers);
 
     LLAMA_API struct llama_sampler * llama_sampler_init_logit_bias(
                              int32_t   n_vocab,
                              int32_t   n_logit_bias,
               const llama_logit_bias * logit_bias);
 
+    // this sampler is meant to be used for fill-in-the-middle infilling
+    // it's supposed to be used after top_k + top_p sampling
+    //
+    // 1. if the sum of the EOG probs times the number of candidates is higher than the sum of the other probs -> pick EOG
+    // 2. combine probs of tokens that have the same prefix
+    //
+    // example:
+    //
+    // - before:
+    //   "hel":   0.5
+    //   "hell":  0.2
+    //   "hello": 0.1
+    //   "dummy": 0.1
+    //
+    // - after:
+    //   "hel":   0.8
+    //   "dummy": 0.1
+    //
+    // 3. discard non-EOG tokens with low prob
+    // 4. if no tokens are left -> pick EOT
+    //
+    LLAMA_API struct llama_sampler * llama_sampler_init_infill(const struct llama_model * model);
 
     // Returns the seed used by the sampler if applicable, LLAMA_DEFAULT_SEED otherwise
     LLAMA_API uint32_t llama_sampler_get_seed(const struct llama_sampler * smpl);
@@ -1208,8 +1254,6 @@ extern "C" {
     LLAMA_API void                           llama_perf_sampler_print(const struct llama_sampler * chain);
     LLAMA_API void                           llama_perf_sampler_reset(      struct llama_sampler * chain);
 
-    LLAMA_API void llama_perf_dump_yaml(FILE * stream, const struct llama_context * ctx);
-
 #ifdef __cplusplus
 }
 #endif

examples/talk-llama/unicode-data.cpp

@@ -2311,7 +2311,7 @@ const std::unordered_set<uint32_t> unicode_set_whitespace = {
 0x003000,
 };
 
-// list is always in ascending order, to enable binary searh
+// list is always in ascending order, to enable binary search
 const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_lowercase = {
 {0x000041, 0x000061},
 {0x000042, 0x000062},
@@ -3748,7 +3748,7 @@ const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_lowercase
 {0x01E921, 0x01E943},
 };
 
-// list is always in ascending order, to enable binary searh
+// list is always in ascending order, to enable binary search
 const std::initializer_list<std::pair<uint32_t, uint32_t>> unicode_map_uppercase = {
 {0x000061, 0x000041},
 {0x000062, 0x000042},

examples/talk-llama/unicode.cpp

@@ -71,15 +71,15 @@ uint32_t unicode_cpt_from_utf8(const std::string & utf8, size_t & offset) {
     throw std::invalid_argument("failed to convert utf8 to codepoint");
 }
 
-//static std::vector<uint16_t> unicode_cpt_to_utf16(uint32_t cp) {
+//static std::vector<uint16_t> unicode_cpt_to_utf16(uint32_t cpt) {
 //    std::vector<uint16_t> result;
-//    if (/* 0x0000 <= cp && */ cp <= 0xffff) {
-//        result.emplace_back(cp);
+//    if (/* 0x0000 <= cpt && */ cpt <= 0xffff) {
+//        result.emplace_back(cpt);
 //        return result;
 //    }
-//    if (0x10000 <= cp && cp <= 0x10ffff) {
-//        result.emplace_back(0xd800 | ((cp - 0x10000) >> 10));
-//        result.emplace_back(0xdc00 | ((cp - 0x10000) & 0x03ff));
+//    if (0x10000 <= cpt && cpt <= 0x10ffff) {
+//        result.emplace_back(0xd800 | ((cpt - 0x10000) >> 10));
+//        result.emplace_back(0xdc00 | ((cpt - 0x10000) & 0x03ff));
 //        return result;
 //    }
 //    throw std::invalid_argument("failed to convert codepoint to utf16");
@@ -120,8 +120,8 @@ uint32_t unicode_cpt_from_utf8(const std::string & utf8, size_t & offset) {
 //    return result;
 //}
 
-static std::vector<codepoint_flags> unicode_cpt_flags_array() {
-    std::vector<codepoint_flags> cpt_flags(MAX_CODEPOINTS, codepoint_flags::UNDEFINED);
+static std::vector<unicode_cpt_flags> unicode_cpt_flags_array() {
+    std::vector<unicode_cpt_flags> cpt_flags(MAX_CODEPOINTS, unicode_cpt_flags::UNDEFINED);
 
     assert (unicode_ranges_flags.begin()[0].first == 0);
     assert (unicode_ranges_flags.begin()[unicode_ranges_flags.size()-1].first == MAX_CODEPOINTS);
@@ -201,7 +201,18 @@ static std::unordered_map<std::string, uint8_t> unicode_utf8_to_byte_map() {
 }
 
 static inline std::wstring unicode_wstring_from_utf8(const std::string & s) {
+#if defined(__clang__)
+    // disable C++17 deprecation warning for std::codecvt_utf8
+#    pragma clang diagnostic push
+#    pragma clang diagnostic ignored "-Wdeprecated-declarations"
+#endif
+
     std::wstring_convert<std::codecvt_utf8<wchar_t>> conv;
+
+#if defined(__clang__)
+#    pragma clang diagnostic pop
+#endif
+
     return conv.from_bytes(s);
 }
 
@@ -242,8 +253,8 @@ static std::vector<size_t> unicode_regex_split_custom_gpt2(const std::string & t
             return (offset_ini <= pos && pos < offset_end) ? cpts[pos] : OUT_OF_RANGE;
         };
 
-        auto _get_flags = [&] (const size_t pos) -> codepoint_flags {
-            return (offset_ini <= pos && pos < offset_end) ? unicode_cpt_flags(cpts[pos]) : codepoint_flags{};
+        auto _get_flags = [&] (const size_t pos) -> unicode_cpt_flags {
+            return (offset_ini <= pos && pos < offset_end) ? unicode_cpt_flags_from_cpt(cpts[pos]) : unicode_cpt_flags{};
         };
 
         size_t _prev_end = offset_ini;
@@ -360,8 +371,8 @@ static std::vector<size_t> unicode_regex_split_custom_llama3(const std::string &
             return (offset_ini <= pos && pos < offset_end) ? cpts[pos] : OUT_OF_RANGE;
         };
 
-        auto _get_flags = [&] (const size_t pos) -> codepoint_flags {
-            return (offset_ini <= pos && pos < offset_end) ? unicode_cpt_flags(cpts[pos]) : codepoint_flags{};
+        auto _get_flags = [&] (const size_t pos) -> unicode_cpt_flags {
+            return (offset_ini <= pos && pos < offset_end) ? unicode_cpt_flags_from_cpt(cpts[pos]) : unicode_cpt_flags{};
         };
 
         size_t _prev_end = offset_ini;
@@ -561,29 +572,29 @@ static std::vector<size_t> unicode_regex_split_custom(const std::string & text,
 // interface
 //
 
-std::string unicode_cpt_to_utf8(uint32_t cp) {
+std::string unicode_cpt_to_utf8(uint32_t cpt) {
     std::string result;
 
-    if (/* 0x00 <= cp && */ cp <= 0x7f) {
-        result.push_back(cp);
+    if (/* 0x00 <= cpt && */ cpt <= 0x7f) {
+        result.push_back(cpt);
         return result;
     }
-    if (0x80 <= cp && cp <= 0x7ff) {
-        result.push_back(0xc0 | ((cp >> 6) & 0x1f));
-        result.push_back(0x80 | (cp & 0x3f));
+    if (0x80 <= cpt && cpt <= 0x7ff) {
+        result.push_back(0xc0 | ((cpt >> 6) & 0x1f));
+        result.push_back(0x80 | (cpt & 0x3f));
         return result;
     }
-    if (0x800 <= cp && cp <= 0xffff) {
-        result.push_back(0xe0 | ((cp >> 12) & 0x0f));
-        result.push_back(0x80 | ((cp >> 6) & 0x3f));
-        result.push_back(0x80 | (cp & 0x3f));
+    if (0x800 <= cpt && cpt <= 0xffff) {
+        result.push_back(0xe0 | ((cpt >> 12) & 0x0f));
+        result.push_back(0x80 | ((cpt >> 6) & 0x3f));
+        result.push_back(0x80 | (cpt & 0x3f));
         return result;
     }
-    if (0x10000 <= cp && cp <= 0x10ffff) {
-        result.push_back(0xf0 | ((cp >> 18) & 0x07));
-        result.push_back(0x80 | ((cp >> 12) & 0x3f));
-        result.push_back(0x80 | ((cp >> 6) & 0x3f));
-        result.push_back(0x80 | (cp & 0x3f));
+    if (0x10000 <= cpt && cpt <= 0x10ffff) {
+        result.push_back(0xf0 | ((cpt >> 18) & 0x07));
+        result.push_back(0x80 | ((cpt >> 12) & 0x3f));
+        result.push_back(0x80 | ((cpt >> 6) & 0x3f));
+        result.push_back(0x80 | (cpt & 0x3f));
         return result;
     }
 
@@ -613,19 +624,19 @@ std::vector<uint32_t> unicode_cpts_from_utf8(const std::string & utf8) {
     return result;
 }
 
-codepoint_flags unicode_cpt_flags(const uint32_t cp) {
-    static const codepoint_flags undef(codepoint_flags::UNDEFINED);
+unicode_cpt_flags unicode_cpt_flags_from_cpt(const uint32_t cpt) {
+    static const unicode_cpt_flags undef(unicode_cpt_flags::UNDEFINED);
     static const auto cpt_flags = unicode_cpt_flags_array();
-    return cp < cpt_flags.size() ? cpt_flags[cp] : undef;
+    return cpt < cpt_flags.size() ? cpt_flags[cpt] : undef;
 }
 
-codepoint_flags unicode_cpt_flags(const std::string & utf8) {
-    static const codepoint_flags undef(codepoint_flags::UNDEFINED);
+unicode_cpt_flags unicode_cpt_flags_from_utf8(const std::string & utf8) {
+    static const unicode_cpt_flags undef(unicode_cpt_flags::UNDEFINED);
     if (utf8.empty()) {
         return undef;  // undefined
     }
     size_t offset = 0;
-    return unicode_cpt_flags(unicode_cpt_from_utf8(utf8, offset));
+    return unicode_cpt_flags_from_cpt(unicode_cpt_from_utf8(utf8, offset));
 }
 
 std::string unicode_byte_to_utf8(uint8_t byte) {
@@ -638,41 +649,41 @@ uint8_t unicode_utf8_to_byte(const std::string & utf8) {
     return map.at(utf8);
 }
 
-uint32_t unicode_tolower(uint32_t cp) {
+uint32_t unicode_tolower(uint32_t cpt) {
     // binary search
-    auto it = std::lower_bound(unicode_map_lowercase.begin(), unicode_map_lowercase.end(), cp,
+    auto it = std::lower_bound(unicode_map_lowercase.begin(), unicode_map_lowercase.end(), cpt,
         [](const std::pair<uint32_t, uint32_t> & pair, uint32_t value) {
             return pair.first < value;
         });
-    if (it != unicode_map_lowercase.end() && it->first == cp) {
+    if (it != unicode_map_lowercase.end() && it->first == cpt) {
         return it->second;
     }
-    return cp;  // Return the original code point if no lowercase mapping is found
+    return cpt;  // Return the original code point if no lowercase mapping is found
 }
 
 std::vector<std::string> unicode_regex_split(const std::string & text, const std::vector<std::string> & regex_exprs) {
     // unicode categories
     static const std::map<std::string, int> k_ucat_enum = {
-        { "\\p{N}", codepoint_flags::NUMBER },
-        { "\\p{L}", codepoint_flags::LETTER },
-        { "\\p{P}", codepoint_flags::PUNCTUATION },
+        { "\\p{N}", unicode_cpt_flags::NUMBER },
+        { "\\p{L}", unicode_cpt_flags::LETTER },
+        { "\\p{P}", unicode_cpt_flags::PUNCTUATION },
     };
 
     static const std::map<int, int> k_ucat_cpt = {
-        { codepoint_flags::NUMBER,        0xD1 },
-        { codepoint_flags::LETTER,        0xD2 },
-        { codepoint_flags::PUNCTUATION,   0xD3 },
+        { unicode_cpt_flags::NUMBER,      0xD1 },
+        { unicode_cpt_flags::LETTER,      0xD2 },
+        { unicode_cpt_flags::PUNCTUATION, 0xD3 },
     };
 
     static const std::map<int, std::string> k_ucat_map = {
-        { codepoint_flags::NUMBER,        "\x30-\x39" }, // 0-9
-        { codepoint_flags::LETTER,        "\x41-\x5A\x61-\x7A" }, // A-Za-z
-        { codepoint_flags::PUNCTUATION,   "\x21-\x23\x25-\x2A\x2C-\x2F\x3A-\x3B\x3F-\x40\\\x5B-\\\x5D\x5F\\\x7B\\\x7D" }, // !-#%-*,-/:-;?-@\[-\]_\{\}
+        { unicode_cpt_flags::NUMBER,      "\x30-\x39" }, // 0-9
+        { unicode_cpt_flags::LETTER,      "\x41-\x5A\x61-\x7A" }, // A-Za-z
+        { unicode_cpt_flags::PUNCTUATION, "\x21-\x23\x25-\x2A\x2C-\x2F\x3A-\x3B\x3F-\x40\\\x5B-\\\x5D\x5F\\\x7B\\\x7D" }, // !-#%-*,-/:-;?-@\[-\]_\{\}
     };
 
     // compute collapsed codepoints only if needed by at least one regex
     bool need_collapse = false;
-    for (auto & regex_expr : regex_exprs) {
+    for (const auto & regex_expr : regex_exprs) {
         // search for unicode categories
         for (const auto & ucat : k_ucat_enum) {
             if (std::string::npos != regex_expr.find(ucat.first)) {
@@ -698,7 +709,7 @@ std::vector<std::string> unicode_regex_split(const std::string & text, const std
                 continue;
             }
 
-            const auto flags = unicode_cpt_flags(cpts[i]);
+            const auto flags = unicode_cpt_flags_from_cpt(cpts[i]);
 
             if (flags.is_whitespace) {
                 //NOTE: C++ std::regex \s does not mach 0x85, Rust and Python regex does.
@@ -714,7 +725,7 @@ std::vector<std::string> unicode_regex_split(const std::string & text, const std
 
     std::vector<size_t> bpe_offsets = { cpts.size() };
 
-    for (auto & regex_expr : regex_exprs) {
+    for (const auto & regex_expr : regex_exprs) {
         // first, see if we have an efficient custom regex implementation
         auto tmp = unicode_regex_split_custom(text, regex_expr, bpe_offsets);
 
@@ -728,7 +739,7 @@ std::vector<std::string> unicode_regex_split(const std::string & text, const std
             // if a unicode category is used in the regex, we use the collapsed text and replace the unicode category
             // with the corresponding collapsed representation
             bool use_collapsed = false;
-            for (auto & ucat : k_ucat_enum) {
+            for (const auto & ucat : k_ucat_enum) {
                 if (std::string::npos != regex_expr.find(ucat.first)) {
                     use_collapsed = true;
                     break;
@@ -794,7 +805,7 @@ std::vector<std::string> unicode_regex_split(const std::string & text, const std
                 // std::wregex \s does not mach non-ASCII whitespaces, using 0x0B as fallback
                 std::wstring wtext(cpts.begin(), cpts.end());
                 for (size_t i = 0; i < wtext.size(); ++i) {
-                    if (wtext[i] > 0x7F && unicode_cpt_flags(wtext[i]).is_whitespace) {
+                    if (wtext[i] > 0x7F && unicode_cpt_flags_from_cpt(wtext[i]).is_whitespace) {
                         wtext[i] = 0x0B;
                     }
                 }

examples/talk-llama/unicode.h

@@ -4,9 +4,7 @@
 #include <string>
 #include <vector>
 
-// TODO: prefix all symbols with "llama_"
-
-struct codepoint_flags {
+struct unicode_cpt_flags {
     enum {
         UNDEFINED       = 0x0001,
         NUMBER          = 0x0002,  // regex: \p{N}
@@ -35,7 +33,7 @@ struct codepoint_flags {
     uint16_t is_nfd         : 1;
 
     // decode from uint16
-    inline codepoint_flags(const uint16_t flags=0) {
+    inline unicode_cpt_flags(const uint16_t flags = 0) {
         *reinterpret_cast<uint16_t*>(this) = flags;
     }
 
@@ -50,18 +48,19 @@ struct codepoint_flags {
 
 size_t unicode_len_utf8(char src);
 
-std::string unicode_cpt_to_utf8(uint32_t cp);
-uint32_t unicode_cpt_from_utf8(const std::string & utf8, size_t & offset);
+std::string unicode_cpt_to_utf8  (uint32_t cpt);
+uint32_t    unicode_cpt_from_utf8(const std::string & utf8, size_t & offset);
+
 std::vector<uint32_t> unicode_cpts_from_utf8(const std::string & utf8);
 
 std::vector<uint32_t> unicode_cpts_normalize_nfd(const std::vector<uint32_t> & cpts);
 
-codepoint_flags unicode_cpt_flags(const uint32_t cp);
-codepoint_flags unicode_cpt_flags(const std::string & utf8);
+unicode_cpt_flags unicode_cpt_flags_from_cpt (uint32_t cpt);
+unicode_cpt_flags unicode_cpt_flags_from_utf8(const std::string & utf8);
 
 std::string unicode_byte_to_utf8(uint8_t byte);
-uint8_t unicode_utf8_to_byte(const std::string & utf8);
+uint8_t     unicode_utf8_to_byte(const std::string & utf8);
 
-uint32_t unicode_tolower(uint32_t cp);
+uint32_t unicode_tolower(uint32_t cpt);
 
 std::vector<std::string> unicode_regex_split(const std::string & text, const std::vector<std::string> & regex_exprs);

examples/talk.wasm/CMakeLists.txt

@@ -1,51 +0,0 @@
-#
-# libtalk
-#
-
-set(TARGET libtalk)
-
-add_executable(${TARGET}
-    emscripten.cpp
-    gpt-2.cpp
-    )
-
-include(DefaultTargetOptions)
-
-target_link_libraries(${TARGET} PRIVATE
-    whisper
-    common
-    )
-
-unset(EXTRA_FLAGS)
-
-if (WHISPER_WASM_SINGLE_FILE)
-    set(EXTRA_FLAGS "-s SINGLE_FILE=1")
-    message(STATUS "Embedding WASM inside talk.js")
-
-    add_custom_command(
-        TARGET ${TARGET} POST_BUILD
-        COMMAND ${CMAKE_COMMAND} -E copy
-        ${CMAKE_BINARY_DIR}/bin/libtalk.js
-        ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/talk.wasm/talk.js
-        )
-endif()
-
-set_target_properties(${TARGET} PROPERTIES LINK_FLAGS " \
-    --bind \
-    -s USE_PTHREADS=1 \
-    -s PTHREAD_POOL_SIZE=8 \
-    -s INITIAL_MEMORY=1800MB \
-    -s TOTAL_MEMORY=1800MB \
-    -s FORCE_FILESYSTEM=1 \
-    -s EXPORTED_RUNTIME_METHODS=\"['print', 'printErr', 'ccall', 'cwrap']\" \
-    ${EXTRA_FLAGS} \
-    ")
-
-#
-# talk.wasm
-#
-
-set(TARGET talk.wasm)
-
-configure_file(${CMAKE_CURRENT_SOURCE_DIR}/index-tmpl.html  ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${TARGET}/index.html @ONLY)
-configure_file(${CMAKE_CURRENT_SOURCE_DIR}/../helpers.js    ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${TARGET}/helpers.js @ONLY)

examples/talk.wasm/README.md

@@ -1,74 +0,0 @@
-# talk.wasm
-
-Talk with an Artificial Intelligence in your browser:
-
-[https://user-images.githubusercontent.com/1991296/203411580-fedb4839-05e4-4474-8364-aaf1e9a9b615.mp4](https://user-images.githubusercontent.com/1991296/203845553-f7b44e13-9a15-4fc8-b518-ae8f4c6770fe.mp4)
-
-Online demo: https://whisper.ggerganov.com/talk/
-
-Terminal version: [examples/talk](/examples/talk)
-
-## How it works?
-
-This demo leverages 2 modern neural network models to create a high-quality voice chat directly in your browser:
-
-- [OpenAI's Whisper](https://github.com/openai/whisper) speech recognition model is used to process your voice and understand what you are saying
-- Upon receiving some voice input, the AI generates a text response using [OpenAI's GPT-2](https://github.com/openai/gpt-2) language model
-- The AI then vocalizes the response using the browser's [Web Speech API](https://developer.mozilla.org/en-US/docs/Web/API/Web_Speech_API)
-
-The web page does the processing locally on your machine. The processing of these heavy neural network models in the
-browser is possible by implementing them efficiently in C/C++ and using the browser's WebAssembly SIMD capabilities for
-extra performance:
-
-- The Whisper C++ implementation is here: [whisper.h](/whisper.h) / [whisper.cpp](/whisper.cpp)
-- The GPT-2 C++ implementation is here: [gpt-2.h](gpt-2.h) / [gpt-2.cpp](gpt-2.cpp)
-- Both models use a custom tensor library implemented in C: [ggml.h](/ggml.h) / [ggml.c](/ggml.c)
-- The HTML/JS layer is here: [index-tmpl.html](index-tmpl.html)
-- The Emscripten bridge between C/C++ and JS is here: [emscripten.cpp](emscripten.cpp)
-
-In order to run the models, the web page first needs to download the model data which is about ~350 MB. The model data
-is then cached in your browser's cache and can be reused in future visits without downloading it again.
-
-## Requirements
-
-In order to run this demo efficiently, you need to have the following:
-
-- Latest Chrome or Firefox browser (Safari is not supported)
-- Run this on a desktop or laptop with modern CPU (a mobile phone will likely not be good enough)
-- Speak phrases that are no longer than 10 seconds - this is the audio context of the AI
-- The web-page uses about 1.8GB of RAM
-
-Notice that this demo is using the smallest GPT-2 model, so the generated text responses are not always very good.
-Also, the prompting strategy can likely be improved to achieve better results.
-
-The demo is quite computationally heavy, so you need a fast CPU. It's not usual to run these transformer models in a
-browser. Typically, they run on powerful GPUs.
-
-Currently, mobile browsers do not support the Fixed-width SIMD WebAssembly capability, so you cannot run this demo
-on a phone or a tablet. Hopefully, in the near future this will become supported.
-
-## Todo
-
-- Better UI (contributions are welcome)
-- Better GPT-2 prompting
-
-## Build instructions
-
-```bash
-# build using Emscripten (v3.1.2)
-git clone https://github.com/ggerganov/whisper.cpp
-cd whisper.cpp
-mkdir build-em && cd build-em
-emcmake cmake ..
-make -j
-
-# copy the produced page to your HTTP path
-cp bin/talk.wasm/*       /path/to/html/
-cp bin/libtalk.worker.js /path/to/html/
-```
-
-## Feedback
-
-If you have any comments or ideas for improvement, please drop a comment in the following discussion:
-
-https://github.com/ggerganov/whisper.cpp/discussions/167

examples/talk.wasm/emscripten.cpp

@@ -1,368 +0,0 @@
-#include "ggml.h"
-#include "gpt-2.h"
-#include "whisper.h"
-
-#include <emscripten.h>
-#include <emscripten/bind.h>
-
-#include <atomic>
-#include <cmath>
-#include <mutex>
-#include <string>
-#include <thread>
-#include <vector>
-#include <regex>
-
-constexpr int N_THREAD = 8;
-
-struct gpt2_context * g_gpt2;
-std::vector<struct whisper_context *> g_contexts(4, nullptr);
-
-std::mutex g_mutex;
-std::thread g_worker;
-std::atomic<bool> g_running(false);
-
-bool g_force_speak = false;
-std::string g_text_to_speak = "";
-std::string g_status = "";
-std::string g_status_forced = "";
-
-std::vector<float> g_pcmf32;
-
-void talk_set_status(const std::string & status) {
-    std::lock_guard<std::mutex> lock(g_mutex);
-    g_status = status;
-}
-
-void talk_main(size_t index) {
-    talk_set_status("loading data ...");
-
-    struct whisper_full_params wparams = whisper_full_default_params(whisper_sampling_strategy::WHISPER_SAMPLING_GREEDY);
-
-    wparams.n_threads        = std::min(N_THREAD, (int) std::thread::hardware_concurrency());
-    wparams.offset_ms        = 0;
-    wparams.translate        = false;
-    wparams.no_context       = true;
-    wparams.single_segment   = true;
-    wparams.print_realtime   = false;
-    wparams.print_progress   = false;
-    wparams.print_timestamps = true;
-    wparams.print_special    = false;
-
-    wparams.max_tokens       = 32;
-    wparams.audio_ctx        = 768; // partial encoder context for better performance
-
-    wparams.language         = "en";
-
-    g_gpt2 = gpt2_init("gpt-2.bin");
-
-    printf("talk: using %d threads\n", wparams.n_threads);
-
-    std::vector<float> pcmf32;
-
-    // whisper context
-    auto & ctx = g_contexts[index];
-
-    const int64_t step_samples   = 2*WHISPER_SAMPLE_RATE;
-    const int64_t window_samples = 9*WHISPER_SAMPLE_RATE;
-    const int64_t step_ms        = (step_samples*1000)/WHISPER_SAMPLE_RATE;
-
-    auto t_last = std::chrono::high_resolution_clock::now();
-
-    talk_set_status("listening ...");
-
-    while (g_running) {
-
-        const auto t_now = std::chrono::high_resolution_clock::now();
-        if (std::chrono::duration_cast<std::chrono::milliseconds>(t_now - t_last).count() < step_ms) {
-            {
-                std::lock_guard<std::mutex> lock(g_mutex);
-                g_pcmf32.clear();
-            }
-            std::this_thread::sleep_for(std::chrono::milliseconds(10));
-            continue;
-        }
-
-        talk_set_status("listening ...");
-
-        {
-            std::unique_lock<std::mutex> lock(g_mutex);
-
-            if (g_pcmf32.size() < step_samples) {
-                lock.unlock();
-
-                std::this_thread::sleep_for(std::chrono::milliseconds(10));
-
-                continue;
-            }
-
-            pcmf32 = std::vector<float>(g_pcmf32.end() - std::min((int64_t) g_pcmf32.size(), window_samples), g_pcmf32.end());
-        }
-
-        // VAD: if energy in during last second is above threshold, then skip
-        {
-            float energy_all = 0.0f;
-            float energy_1s  = 0.0f;
-
-            for (size_t i = 0; i < pcmf32.size(); i++) {
-                energy_all += fabsf(pcmf32[i]);
-
-                if (i >= pcmf32.size() - WHISPER_SAMPLE_RATE) {
-                    energy_1s += fabsf(pcmf32[i]);
-                }
-            }
-
-            energy_all /= pcmf32.size();
-            energy_1s  /= WHISPER_SAMPLE_RATE;
-
-            if (energy_1s > 0.1f*energy_all && !g_force_speak) {
-                std::this_thread::sleep_for(std::chrono::milliseconds(10));
-                continue;
-            }
-        }
-
-        talk_set_status("processing audio (whisper)...");
-
-        t_last = t_now;
-
-        if (!g_force_speak) {
-            const auto t_start = std::chrono::high_resolution_clock::now();
-
-            int ret = whisper_full(ctx, wparams, pcmf32.data(), pcmf32.size());
-            if (ret != 0) {
-                printf("whisper_full() failed: %d\n", ret);
-                break;
-            }
-
-            const auto t_end = std::chrono::high_resolution_clock::now();
-
-            printf("whisper_full() returned %d in %f seconds\n", ret, std::chrono::duration<double>(t_end - t_start).count());
-        }
-
-        {
-            std::string text_heard;
-
-            if (!g_force_speak) {
-                const int n_segments = whisper_full_n_segments(ctx);
-                for (int i = n_segments - 1; i < n_segments; ++i) {
-                    const char * text = whisper_full_get_segment_text(ctx, i);
-
-                    const int64_t t0 = whisper_full_get_segment_t0(ctx, i);
-                    const int64_t t1 = whisper_full_get_segment_t1(ctx, i);
-
-                    printf ("[%s --> %s]  %s\n", to_timestamp(t0).c_str(), to_timestamp(t1).c_str(), text);
-
-                    text_heard += text;
-                }
-            }
-
-            g_force_speak = false;
-
-            // remove text between brackets using regex
-            {
-                std::regex re("\\[.*?\\]");
-                text_heard = std::regex_replace(text_heard, re, "");
-            }
-
-            // remove text between brackets using regex
-            {
-                std::regex re("\\(.*?\\)");
-                text_heard = std::regex_replace(text_heard, re, "");
-            }
-
-            // remove all characters, except for letters, numbers, punctuation and ':', '\'', '-', ' '
-            text_heard = std::regex_replace(text_heard, std::regex("[^a-zA-Z0-9\\.,\\?!\\s\\:\\'\\-]"), "");
-
-            // take first line
-            text_heard = text_heard.substr(0, text_heard.find_first_of("\n"));
-
-            // remove leading and trailing whitespace
-            text_heard = std::regex_replace(text_heard, std::regex("^\\s+"), "");
-            text_heard = std::regex_replace(text_heard, std::regex("\\s+$"), "");
-
-            talk_set_status("'" + text_heard + "' - thinking how to respond (gpt-2) ...");
-
-            const std::vector<gpt_vocab::id> tokens = gpt2_tokenize(g_gpt2, text_heard.c_str());
-
-            printf("whisper: number of tokens: %d, '%s'\n", (int) tokens.size(), text_heard.c_str());
-
-            std::string text_to_speak;
-            std::string prompt_base;
-
-            {
-                std::lock_guard<std::mutex> lock(g_mutex);
-                prompt_base = gpt2_get_prompt(g_gpt2);
-            }
-
-            if (tokens.size() > 0) {
-                text_to_speak = gpt2_gen_text(g_gpt2, (prompt_base + text_heard + "\n").c_str(), 32);
-                text_to_speak = std::regex_replace(text_to_speak, std::regex("[^a-zA-Z0-9\\.,\\?!\\s\\:\\'\\-]"), "");
-                text_to_speak = text_to_speak.substr(0, text_to_speak.find_first_of("\n"));
-
-                std::lock_guard<std::mutex> lock(g_mutex);
-
-                // remove first 2 lines of base prompt
-                {
-                    const size_t pos = prompt_base.find_first_of("\n");
-                    if (pos != std::string::npos) {
-                        prompt_base = prompt_base.substr(pos + 1);
-                    }
-                }
-                {
-                    const size_t pos = prompt_base.find_first_of("\n");
-                    if (pos != std::string::npos) {
-                        prompt_base = prompt_base.substr(pos + 1);
-                    }
-                }
-                prompt_base += text_heard + "\n" + text_to_speak + "\n";
-            } else {
-                text_to_speak = gpt2_gen_text(g_gpt2, prompt_base.c_str(), 32);
-                text_to_speak = std::regex_replace(text_to_speak, std::regex("[^a-zA-Z0-9\\.,\\?!\\s\\:\\'\\-]"), "");
-                text_to_speak = text_to_speak.substr(0, text_to_speak.find_first_of("\n"));
-
-                std::lock_guard<std::mutex> lock(g_mutex);
-
-                const size_t pos = prompt_base.find_first_of("\n");
-                if (pos != std::string::npos) {
-                    prompt_base = prompt_base.substr(pos + 1);
-                }
-                prompt_base += text_to_speak + "\n";
-            }
-
-            printf("gpt-2: %s\n", text_to_speak.c_str());
-
-            //printf("========================\n");
-            //printf("gpt-2: prompt_base:\n'%s'\n", prompt_base.c_str());
-            //printf("========================\n");
-
-            {
-                std::lock_guard<std::mutex> lock(g_mutex);
-                t_last = std::chrono::high_resolution_clock::now();
-                g_text_to_speak = text_to_speak;
-                g_pcmf32.clear();
-                gpt2_set_prompt(g_gpt2, prompt_base.c_str());
-            }
-
-            talk_set_status("speaking ...");
-        }
-    }
-
-    gpt2_free(g_gpt2);
-
-    if (index < g_contexts.size()) {
-        whisper_free(g_contexts[index]);
-        g_contexts[index] = nullptr;
-    }
-}
-
-EMSCRIPTEN_BINDINGS(talk) {
-    emscripten::function("init", emscripten::optional_override([](const std::string & path_model) {
-        for (size_t i = 0; i < g_contexts.size(); ++i) {
-            if (g_contexts[i] == nullptr) {
-                g_contexts[i] = whisper_init_from_file_with_params(path_model.c_str(), whisper_context_default_params());
-                if (g_contexts[i] != nullptr) {
-                    g_running = true;
-                    if (g_worker.joinable()) {
-                        g_worker.join();
-                    }
-                    g_worker = std::thread([i]() {
-                        talk_main(i);
-                    });
-
-                    return i + 1;
-                } else {
-                    return (size_t) 0;
-                }
-            }
-        }
-
-        return (size_t) 0;
-    }));
-
-    emscripten::function("free", emscripten::optional_override([](size_t index) {
-        if (g_running) {
-            g_running = false;
-        }
-    }));
-
-    emscripten::function("set_audio", emscripten::optional_override([](size_t index, const emscripten::val & audio) {
-        --index;
-
-        if (index >= g_contexts.size()) {
-            return -1;
-        }
-
-        if (g_contexts[index] == nullptr) {
-            return -2;
-        }
-
-        {
-            std::lock_guard<std::mutex> lock(g_mutex);
-            const int n = audio["length"].as<int>();
-
-            emscripten::val heap = emscripten::val::module_property("HEAPU8");
-            emscripten::val memory = heap["buffer"];
-
-            g_pcmf32.resize(n);
-
-            emscripten::val memoryView = audio["constructor"].new_(memory, reinterpret_cast<uintptr_t>(g_pcmf32.data()), n);
-            memoryView.call<void>("set", audio);
-        }
-
-        return 0;
-    }));
-
-    emscripten::function("force_speak", emscripten::optional_override([](size_t index) {
-        {
-            std::lock_guard<std::mutex> lock(g_mutex);
-            g_force_speak = true;
-        }
-    }));
-
-    emscripten::function("get_text_context", emscripten::optional_override([]() {
-        std::string text_context;
-
-        {
-            std::lock_guard<std::mutex> lock(g_mutex);
-            text_context = gpt2_get_prompt(g_gpt2);
-        }
-
-        return text_context;
-    }));
-
-    emscripten::function("get_text_to_speak", emscripten::optional_override([]() {
-        std::string text_to_speak;
-
-        {
-            std::lock_guard<std::mutex> lock(g_mutex);
-            text_to_speak = std::move(g_text_to_speak);
-        }
-
-        return text_to_speak;
-    }));
-
-    emscripten::function("get_status", emscripten::optional_override([]() {
-        std::string status;
-
-        {
-            std::lock_guard<std::mutex> lock(g_mutex);
-            status = g_status_forced.empty() ? g_status : g_status_forced;
-        }
-
-        return status;
-    }));
-
-    emscripten::function("set_status", emscripten::optional_override([](const std::string & status) {
-        {
-            std::lock_guard<std::mutex> lock(g_mutex);
-            g_status_forced = status;
-        }
-    }));
-
-    emscripten::function("set_prompt", emscripten::optional_override([](const std::string & prompt) {
-        {
-            std::lock_guard<std::mutex> lock(g_mutex);
-            gpt2_set_prompt(g_gpt2, prompt.c_str());
-        }
-    }));
-}

examples/talk.wasm/gpt-2.cpp

@@ -1,808 +0,0 @@
-#include "ggml.h"
-#include "common-ggml.h"
-
-#include "gpt-2.h"
-
-#include <cmath>
-#include <cstdio>
-#include <cstring>
-#include <fstream>
-#include <map>
-#include <string>
-#include <thread>
-#include <vector>
-#include <regex>
-#include <random>
-
-/////////////////////// GPT-2 BEGIN /////////////////////////
-
-// default hparams (GPT-2 117M)
-struct gpt2_hparams {
-    int32_t n_vocab = 50257;
-    int32_t n_ctx   = 1024;
-    int32_t n_embd  = 768;
-    int32_t n_head  = 12;
-    int32_t n_layer = 12;
-    int32_t ftype   = 1;
-};
-
-struct gpt2_layer {
-    // normalization
-    struct ggml_tensor * ln_1_g;
-    struct ggml_tensor * ln_1_b;
-
-    struct ggml_tensor * ln_2_g;
-    struct ggml_tensor * ln_2_b;
-
-    // attention
-    struct ggml_tensor * c_attn_attn_w;
-    struct ggml_tensor * c_attn_attn_b;
-
-    struct ggml_tensor * c_attn_proj_w;
-    struct ggml_tensor * c_attn_proj_b;
-
-    // mlp
-    struct ggml_tensor * c_mlp_fc_w;
-    struct ggml_tensor * c_mlp_fc_b;
-
-    struct ggml_tensor * c_mlp_proj_w;
-    struct ggml_tensor * c_mlp_proj_b;
-};
-
-struct gpt2_model {
-    gpt2_hparams hparams;
-
-    // normalization
-    struct ggml_tensor * ln_f_g;
-    struct ggml_tensor * ln_f_b;
-
-    struct ggml_tensor * wte;     // position embedding
-    struct ggml_tensor * wpe;     //    token embedding
-    struct ggml_tensor * lm_head; // language model head
-
-    std::vector<gpt2_layer> layers;
-
-    // key + value memory
-    struct ggml_tensor * memory_k;
-    struct ggml_tensor * memory_v;
-
-    //
-    struct ggml_context * ctx;
-    std::map<std::string, struct ggml_tensor *> tensors;
-};
-
-// load the model's weights from a file
-bool gpt2_model_load(const std::string & fname, gpt2_model & model, gpt_vocab & vocab) {
-    printf("%s: loading model from '%s'\n", __func__, fname.c_str());
-
-    auto fin = std::ifstream(fname, std::ios::binary);
-    if (!fin) {
-        fprintf(stderr, "%s: failed to open '%s'\n", __func__, fname.c_str());
-        return false;
-    }
-
-    // verify magic
-    {
-        uint32_t magic;
-        fin.read((char *) &magic, sizeof(magic));
-        if (magic != 0x67676d6c) {
-            fprintf(stderr, "%s: invalid model file '%s' (bad magic)\n", __func__, fname.c_str());
-            return false;
-        }
-    }
-
-    // load hparams
-    {
-        auto & hparams = model.hparams;
-
-        fin.read((char *) &hparams.n_vocab, sizeof(hparams.n_vocab));
-        fin.read((char *) &hparams.n_ctx,   sizeof(hparams.n_ctx));
-        fin.read((char *) &hparams.n_embd,  sizeof(hparams.n_embd));
-        fin.read((char *) &hparams.n_head,  sizeof(hparams.n_head));
-        fin.read((char *) &hparams.n_layer, sizeof(hparams.n_layer));
-        fin.read((char *) &hparams.ftype,   sizeof(hparams.ftype));
-
-        printf("%s: n_vocab = %d\n", __func__, hparams.n_vocab);
-        printf("%s: n_ctx   = %d\n", __func__, hparams.n_ctx);
-        printf("%s: n_embd  = %d\n", __func__, hparams.n_embd);
-        printf("%s: n_head  = %d\n", __func__, hparams.n_head);
-        printf("%s: n_layer = %d\n", __func__, hparams.n_layer);
-        printf("%s: ftype   = %d\n", __func__, hparams.ftype);
-    }
-
-    // load vocab
-    {
-        int32_t n_vocab = 0;
-        fin.read((char *) &n_vocab, sizeof(n_vocab));
-
-        if (n_vocab != model.hparams.n_vocab) {
-            fprintf(stderr, "%s: invalid model file '%s' (bad vocab size %d != %d)\n",
-                    __func__, fname.c_str(), n_vocab, model.hparams.n_vocab);
-            return false;
-        }
-
-        std::string word;
-        for (int i = 0; i < n_vocab; i++) {
-            uint32_t len;
-            fin.read((char *) &len, sizeof(len));
-
-            word.resize(len);
-            fin.read((char *) word.data(), len);
-
-            vocab.token_to_id[word] = i;
-            vocab.id_to_token[i] = word;
-        }
-    }
-
-    // for the big tensors, we have the option to store the data in 16-bit floats or quantized
-    // in order to save memory and also to speed up the computation
-    ggml_type wtype = ggml_ftype_to_ggml_type((ggml_ftype) (model.hparams.ftype));
-    if (wtype == GGML_TYPE_COUNT) {
-        fprintf(stderr, "%s: invalid model file '%s' (bad ftype value %d)\n",
-                __func__, fname.c_str(), model.hparams.ftype);
-        return false;
-    }
-
-    auto & ctx = model.ctx;
-
-    size_t ctx_size = 0;
-
-    {
-        const auto & hparams = model.hparams;
-
-        const int n_embd  = hparams.n_embd;
-        const int n_layer = hparams.n_layer;
-        const int n_ctx   = hparams.n_ctx;
-        const int n_vocab = hparams.n_vocab;
-
-        ctx_size += ggml_row_size(GGML_TYPE_F32, n_embd); // ln_f_g
-        ctx_size += ggml_row_size(GGML_TYPE_F32, n_embd); // ln_f_b
-
-        ctx_size += n_vocab*ggml_row_size(wtype, n_embd);         // wte
-        ctx_size +=   n_ctx*ggml_row_size(GGML_TYPE_F32, n_embd); // wpe
-        ctx_size += n_vocab*ggml_row_size(wtype, n_embd);         // lm_head
-
-        ctx_size += n_layer*(ggml_row_size(GGML_TYPE_F32, n_embd)); // ln_1_g
-        ctx_size += n_layer*(ggml_row_size(GGML_TYPE_F32, n_embd)); // ln_1_b
-
-        ctx_size += n_layer*(ggml_row_size(GGML_TYPE_F32, n_embd)); // ln_2_g
-        ctx_size += n_layer*(ggml_row_size(GGML_TYPE_F32, n_embd)); // ln_2_b
-
-        ctx_size += n_layer*(ggml_row_size(wtype,         3*n_embd*n_embd)); // c_attn_attn_w
-        ctx_size += n_layer*(ggml_row_size(GGML_TYPE_F32, 3*n_embd));        // c_attn_attn_b
-
-        ctx_size += n_layer*(ggml_row_size(wtype,         n_embd*n_embd)); // c_attn_proj_w
-        ctx_size += n_layer*(ggml_row_size(GGML_TYPE_F32, n_embd));        // c_attn_proj_b
-
-        ctx_size += n_layer*(ggml_row_size(wtype,         4*n_embd*n_embd)); // c_mlp_fc_w
-        ctx_size += n_layer*(ggml_row_size(GGML_TYPE_F32, 4*n_embd));        // c_mlp_fc_b
-
-        ctx_size += n_layer*(ggml_row_size(wtype,         4*n_embd*n_embd)); // c_mlp_proj_w
-        ctx_size += n_layer*(ggml_row_size(GGML_TYPE_F32,   n_embd));        // c_mlp_proj_b
-
-        ctx_size += n_ctx*n_layer*ggml_row_size(GGML_TYPE_F32, n_embd); // memory_k
-        ctx_size += n_ctx*n_layer*ggml_row_size(GGML_TYPE_F32, n_embd); // memory_v
-
-        ctx_size += (6 + 12*n_layer)*256; // object overhead
-
-        printf("%s: ggml ctx size = %6.2f MB\n", __func__, ctx_size/(1024.0*1024.0));
-    }
-
-    // create the ggml context
-    {
-        struct ggml_init_params params = {
-            /*.mem_size   =*/ ctx_size,
-            /*.mem_buffer =*/ NULL,
-            /*.no_alloc   =*/ false,
-        };
-
-        model.ctx = ggml_init(params);
-        if (!model.ctx) {
-            fprintf(stderr, "%s: ggml_init() failed\n", __func__);
-            return false;
-        }
-    }
-
-    // prepare memory for the weights
-    {
-        const auto & hparams = model.hparams;
-
-        const int n_embd  = hparams.n_embd;
-        const int n_layer = hparams.n_layer;
-        const int n_ctx   = hparams.n_ctx;
-        const int n_vocab = hparams.n_vocab;
-
-        model.layers.resize(n_layer);
-
-        model.ln_f_g = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_embd);
-        model.ln_f_b = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_embd);
-
-        model.wte     = ggml_new_tensor_2d(ctx, wtype,         n_embd, n_vocab);
-        model.wpe     = ggml_new_tensor_2d(ctx, GGML_TYPE_F32, n_embd, n_ctx);
-        model.lm_head = ggml_new_tensor_2d(ctx, wtype,         n_embd, n_vocab);
-
-        // map by name
-        model.tensors["model/ln_f/g"] = model.ln_f_g;
-        model.tensors["model/ln_f/b"] = model.ln_f_b;
-
-        model.tensors["model/wte"]     = model.wte;
-        model.tensors["model/wpe"]     = model.wpe;
-        model.tensors["model/lm_head"] = model.lm_head;
-
-        for (int i = 0; i < n_layer; ++i) {
-            auto & layer = model.layers[i];
-
-            layer.ln_1_g        = ggml_new_tensor_1d(ctx, GGML_TYPE_F32,   n_embd);
-            layer.ln_1_b        = ggml_new_tensor_1d(ctx, GGML_TYPE_F32,   n_embd);
-
-            layer.ln_2_g        = ggml_new_tensor_1d(ctx, GGML_TYPE_F32,   n_embd);
-            layer.ln_2_b        = ggml_new_tensor_1d(ctx, GGML_TYPE_F32,   n_embd);
-
-            layer.c_attn_attn_w = ggml_new_tensor_2d(ctx, wtype,           n_embd, 3*n_embd);
-            layer.c_attn_attn_b = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, 3*n_embd);
-
-            layer.c_attn_proj_w = ggml_new_tensor_2d(ctx, wtype,           n_embd, n_embd);
-            layer.c_attn_proj_b = ggml_new_tensor_1d(ctx, GGML_TYPE_F32,   n_embd);
-
-            layer.c_mlp_fc_w    = ggml_new_tensor_2d(ctx, wtype,           n_embd, 4*n_embd);
-            layer.c_mlp_fc_b    = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, 4*n_embd);
-
-            layer.c_mlp_proj_w  = ggml_new_tensor_2d(ctx, wtype,         4*n_embd, n_embd);
-            layer.c_mlp_proj_b  = ggml_new_tensor_1d(ctx, GGML_TYPE_F32,   n_embd);
-
-            // map by name
-            model.tensors["model/h" + std::to_string(i) + "/ln_1/g"]        = layer.ln_1_g;
-            model.tensors["model/h" + std::to_string(i) + "/ln_1/b"]        = layer.ln_1_b;
-
-            model.tensors["model/h" + std::to_string(i) + "/ln_2/g"]        = layer.ln_2_g;
-            model.tensors["model/h" + std::to_string(i) + "/ln_2/b"]        = layer.ln_2_b;
-
-            model.tensors["model/h" + std::to_string(i) + "/attn/c_attn/w"] = layer.c_attn_attn_w;
-            model.tensors["model/h" + std::to_string(i) + "/attn/c_attn/b"] = layer.c_attn_attn_b;
-
-            model.tensors["model/h" + std::to_string(i) + "/attn/c_proj/w"] = layer.c_attn_proj_w;
-            model.tensors["model/h" + std::to_string(i) + "/attn/c_proj/b"] = layer.c_attn_proj_b;
-
-            model.tensors["model/h" + std::to_string(i) + "/mlp/c_fc/w"]    = layer.c_mlp_fc_w;
-            model.tensors["model/h" + std::to_string(i) + "/mlp/c_fc/b"]    = layer.c_mlp_fc_b;
-
-            model.tensors["model/h" + std::to_string(i) + "/mlp/c_proj/w"]  = layer.c_mlp_proj_w;
-            model.tensors["model/h" + std::to_string(i) + "/mlp/c_proj/b"]  = layer.c_mlp_proj_b;
-        }
-    }
-
-    // key + value memory
-    {
-        const auto & hparams = model.hparams;
-
-        const int n_embd  = hparams.n_embd;
-        const int n_layer = hparams.n_layer;
-        const int n_ctx   = hparams.n_ctx;
-
-        const int n_mem      = n_layer*n_ctx;
-        const int n_elements = n_embd*n_mem;
-
-        model.memory_k = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_elements);
-        model.memory_v = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_elements);
-
-        const size_t memory_size = ggml_nbytes(model.memory_k) + ggml_nbytes(model.memory_v);
-
-        printf("%s: memory size = %8.2f MB, n_mem = %d\n", __func__, memory_size/1024.0/1024.0, n_mem);
-    }
-
-    // load weights
-    {
-        size_t total_size = 0;
-
-        bool has_lm_head = false;
-
-        while (true) {
-            int32_t n_dims;
-            int32_t length;
-            int32_t ttype;
-
-            fin.read(reinterpret_cast<char *>(&n_dims), sizeof(n_dims));
-            fin.read(reinterpret_cast<char *>(&length), sizeof(length));
-            fin.read(reinterpret_cast<char *>(&ttype),  sizeof(ttype));
-
-            if (fin.eof()) {
-                break;
-            }
-
-            int32_t nelements = 1;
-            int32_t ne[2] = { 1, 1 };
-            for (int i = 0; i < n_dims; ++i) {
-                fin.read(reinterpret_cast<char *>(&ne[i]), sizeof(ne[i]));
-                nelements *= ne[i];
-            }
-
-            std::string name(length, 0);
-            fin.read(&name[0], length);
-
-            if (model.tensors.find(name.data()) == model.tensors.end()) {
-                fprintf(stderr, "%s: unknown tensor '%s' in model file\n", __func__, name.data());
-                return false;
-            }
-
-            auto tensor = model.tensors[name.data()];
-            if (ggml_nelements(tensor) != nelements) {
-                fprintf(stderr, "%s: tensor '%s' has wrong size in model file\n", __func__, name.data());
-                return false;
-            }
-
-            if (tensor->ne[0] != ne[0] || tensor->ne[1] != ne[1]) {
-                fprintf(stderr, "%s: tensor '%s' has wrong shape in model file: got [%d, %d], expected [%d, %d]\n",
-                        __func__, name.data(), (int) tensor->ne[0], (int) tensor->ne[1], ne[0], ne[1]);
-                return false;
-            }
-
-            // for debugging
-            if (0) {
-                printf("%24s - [%5d, %5d], type = %6s, %6.2f MB, %9zu bytes\n", name.data(), ne[0], ne[1], ggml_type_name(ggml_type(ttype)), ggml_nbytes(tensor)/1024.0/1024.0, ggml_nbytes(tensor));
-            }
-
-            const size_t bpe = ggml_type_size(ggml_type(ttype));
-
-            if ((nelements*bpe)/ggml_blck_size(tensor->type) != ggml_nbytes(tensor)) {
-                fprintf(stderr, "%s: tensor '%s' has wrong size in model file: got %zu, expected %zu\n",
-                        __func__, name.data(), ggml_nbytes(tensor), nelements*bpe);
-                return false;
-            }
-
-            fin.read(reinterpret_cast<char *>(tensor->data), ggml_nbytes(tensor));
-
-            // GPT-2 models share the WTE tensor as the LM head
-            if (name == "model/wte" && has_lm_head == false) {
-                memcpy(model.lm_head->data, tensor->data, ggml_nbytes(tensor));
-            }
-
-            if (name == "model/lm_head") {
-                has_lm_head = true;
-            }
-
-            total_size += ggml_nbytes(tensor);
-        }
-
-        printf("%s: model size  = %8.2f MB\n", __func__, total_size/1024.0/1024.0);
-    }
-
-    fin.close();
-
-    return true;
-}
-
-// evaluate the transformer
-//
-//   - model:     the model
-//   - n_threads: number of threads to use
-//   - n_past:    the context size so far
-//   - embd_inp:  the embeddings of the tokens in the context
-//   - embd_w:    the predicted logits for the next token
-//
-bool gpt2_eval(
-        const gpt2_model & model,
-        const int n_threads,
-        const int n_past,
-        const std::vector<gpt_vocab::id> & embd_inp,
-              std::vector<float>         & embd_w,
-              size_t                     & mem_per_token) {
-    const int N = embd_inp.size();
-
-    const auto & hparams = model.hparams;
-
-    const int n_embd  = hparams.n_embd;
-    const int n_layer = hparams.n_layer;
-    const int n_ctx   = hparams.n_ctx;
-    const int n_head  = hparams.n_head;
-    const int n_vocab = hparams.n_vocab;
-
-    static size_t buf_size = 512u*1024*1024;
-    static void * buf = malloc(buf_size);
-
-    if (mem_per_token > 0 && mem_per_token*N > buf_size) {
-        const size_t buf_size_new = 1.1*(mem_per_token*N); // add 10% to account for ggml object overhead
-        //printf("\n%s: reallocating buffer from %zu to %zu bytes\n", __func__, buf_size, buf_size_new);
-
-        // reallocate
-        buf_size = buf_size_new;
-        buf = realloc(buf, buf_size);
-        if (buf == nullptr) {
-            fprintf(stderr, "%s: failed to allocate %zu bytes\n", __func__, buf_size);
-            return false;
-        }
-    }
-
-    struct ggml_init_params params = {
-        /*.mem_size   =*/ buf_size,
-        /*.mem_buffer =*/ buf,
-        /*.no_alloc   =*/ false,
-    };
-
-    struct ggml_context * ctx0 = ggml_init(params);
-    struct ggml_cgraph gf = {};
-
-    struct ggml_tensor * embd = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, N);
-    memcpy(embd->data, embd_inp.data(), N*ggml_element_size(embd));
-
-    struct ggml_tensor * position = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, N);
-    for (int i = 0; i < N; ++i) {
-        ((int32_t *) position->data)[i] = n_past + i;
-    }
-
-    // wte + wpe
-    struct ggml_tensor * inpL =
-        ggml_add(ctx0,
-                ggml_get_rows(ctx0, model.wte, embd),
-                ggml_get_rows(ctx0, model.wpe, position));
-
-    for (int il = 0; il < n_layer; ++il) {
-        struct ggml_tensor * cur;
-
-        // norm
-        {
-            // [ 768, N]
-            cur = ggml_norm(ctx0, inpL, 1e-5f);
-
-            // cur = ln_1_g*cur + ln_1_b
-            // [ 768, N]
-            cur = ggml_add(ctx0,
-                    ggml_mul(ctx0,
-                        ggml_repeat(ctx0, model.layers[il].ln_1_g, cur),
-                        cur),
-                    ggml_repeat(ctx0, model.layers[il].ln_1_b, cur));
-        }
-
-        // attn
-        // [2304, 768] - model.layers[il].c_attn_attn_w
-        // [2304,   1] - model.layers[il].c_attn_attn_b
-        // [ 768,   N] - cur (in)
-        // [2304,   N] - cur (out)
-        //
-        // cur = attn_w*cur + attn_b
-        // [2304, N]
-        {
-            cur = ggml_mul_mat(ctx0,
-                    model.layers[il].c_attn_attn_w,
-                    cur);
-
-            cur = ggml_add(ctx0,
-                    ggml_repeat(ctx0, model.layers[il].c_attn_attn_b, cur),
-                    cur);
-        }
-
-        // self-attention
-        {
-            struct ggml_tensor * Qcur = ggml_view_2d(ctx0, cur, n_embd, N, cur->nb[1], 0*sizeof(float)*n_embd);
-            struct ggml_tensor * Kcur = ggml_view_2d(ctx0, cur, n_embd, N, cur->nb[1], 1*sizeof(float)*n_embd);
-            struct ggml_tensor * Vcur = ggml_view_2d(ctx0, cur, n_embd, N, cur->nb[1], 2*sizeof(float)*n_embd);
-
-            // store key and value to memory
-            if (N >= 1) {
-                struct ggml_tensor * k = ggml_view_1d(ctx0, model.memory_k, N*n_embd, (ggml_element_size(model.memory_k)*n_embd)*(il*n_ctx + n_past));
-                struct ggml_tensor * v = ggml_view_1d(ctx0, model.memory_v, N*n_embd, (ggml_element_size(model.memory_v)*n_embd)*(il*n_ctx + n_past));
-
-                ggml_build_forward_expand(&gf, ggml_cpy(ctx0, Kcur, k));
-                ggml_build_forward_expand(&gf, ggml_cpy(ctx0, Vcur, v));
-            }
-
-            // Q = Qcur.contiguous().view(n_embd/n_head, n_head, N).permute(0, 2, 1, 3)
-            // [64, N, 12]
-            struct ggml_tensor * Q =
-                ggml_permute(ctx0,
-                        ggml_cpy(ctx0,
-                            Qcur,
-                            ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, n_embd/n_head, n_head, N)),
-                        0, 2, 1, 3);
-
-            // K = Kmem.view(n_embd/n_head, n_head, n_past + N).permute(0, 2, 1, 3)
-            // [64, n_past + N, 12]
-            struct ggml_tensor * K =
-                ggml_permute(ctx0,
-                        ggml_reshape_3d(ctx0,
-                            ggml_view_1d(ctx0, model.memory_k, (n_past + N)*n_embd, il*n_ctx*ggml_element_size(model.memory_k)*n_embd),
-                            n_embd/n_head, n_head, n_past + N),
-                        0, 2, 1, 3);
-
-            // GG: flash attention
-            //struct ggml_tensor * V =
-            //    ggml_cpy(ctx0,
-            //            ggml_permute(ctx0,
-            //                ggml_reshape_3d(ctx0,
-            //                    ggml_view_1d(ctx0, model.memory_v, (n_past + N)*n_embd, il*n_ctx*ggml_element_size(model.memory_v)*n_embd),
-            //                    n_embd/n_head, n_head, n_past + N),
-            //                1, 2, 0, 3),
-            //            ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, n_past + N, n_embd/n_head, n_head));
-
-            //struct ggml_tensor * KQV = ggml_flash_attn(ctx0, Q, K, V, true);
-
-            // K * Q
-            // [n_past + N, N, 12]
-            struct ggml_tensor * KQ = ggml_mul_mat(ctx0, K, Q);
-
-            // KQ_scaled = KQ / sqrt(n_embd/n_head)
-            // [n_past + N, N, 12]
-            struct ggml_tensor * KQ_scaled =
-                ggml_scale(ctx0,
-                        KQ,
-                        1.0f/sqrt(float(n_embd)/n_head));
-
-            // KQ_masked = mask_past(KQ_scaled)
-            // [n_past + N, N, 12]
-            struct ggml_tensor * KQ_masked = ggml_diag_mask_inf(ctx0, KQ_scaled, n_past);
-
-            // KQ = soft_max(KQ_masked)
-            // [n_past + N, N, 12]
-            struct ggml_tensor * KQ_soft_max = ggml_soft_max(ctx0, KQ_masked);
-
-            // V_trans = Vmem.view(n_embd/n_head, n_head, n_past + N).permute(1, 2, 0, 3).contiguous()
-            // [n_past + N, 64, 12]
-            struct ggml_tensor * V_trans =
-                ggml_cpy(ctx0,
-                        ggml_permute(ctx0,
-                            ggml_reshape_3d(ctx0,
-                                ggml_view_1d(ctx0, model.memory_v, (n_past + N)*n_embd, il*n_ctx*ggml_element_size(model.memory_v)*n_embd),
-                                n_embd/n_head, n_head, n_past + N),
-                            1, 2, 0, 3),
-                        ggml_new_tensor_3d(ctx0, model.memory_v->type, n_past + N, n_embd/n_head, n_head));
-
-            // KQV = transpose(V) * KQ_soft_max
-            // [64, N, 12]
-            struct ggml_tensor * KQV = ggml_mul_mat(ctx0, V_trans, KQ_soft_max);
-
-            // KQV_merged = KQV.permute(0, 2, 1, 3)
-            // [64, 12, N]
-            struct ggml_tensor * KQV_merged = ggml_permute(ctx0, KQV, 0, 2, 1, 3);
-
-            // cur = KQV_merged.contiguous().view(n_embd, N)
-            // [768, N]
-            cur = ggml_cpy(ctx0,
-                    KQV_merged,
-                    ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_embd, N));
-        }
-
-        // projection
-        // [ 768, 768] - model.layers[il].c_attn_proj_w
-        // [ 768,   1] - model.layers[il].c_attn_proj_b
-        // [ 768,   N] - cur (in)
-        // [ 768,   N] - cur (out)
-        //
-        // cur = proj_w*cur + proj_b
-        // [768, N]
-        {
-            cur = ggml_mul_mat(ctx0,
-                    model.layers[il].c_attn_proj_w,
-                    cur);
-
-            cur = ggml_add(ctx0,
-                    ggml_repeat(ctx0, model.layers[il].c_attn_proj_b, cur),
-                    cur);
-        }
-
-        // add the input
-        cur = ggml_add(ctx0, cur, inpL);
-
-        struct ggml_tensor * inpFF = cur;
-
-        // feed-forward network
-        {
-            // norm
-            {
-                cur = ggml_norm(ctx0, inpFF, 1e-5f);
-
-                // cur = ln_2_g*cur + ln_2_b
-                // [ 768, N]
-                cur = ggml_add(ctx0,
-                        ggml_mul(ctx0,
-                            ggml_repeat(ctx0, model.layers[il].ln_2_g, cur),
-                            cur),
-                        ggml_repeat(ctx0, model.layers[il].ln_2_b, cur));
-            }
-
-            // fully connected
-            // [3072, 768] - model.layers[il].c_mlp_fc_w
-            // [3072,   1] - model.layers[il].c_mlp_fc_b
-            // [ 768,   N] - cur (in)
-            // [3072,   N] - cur (out)
-            //
-            // cur = fc_w*cur + fc_b
-            // [3072, N]
-            cur = ggml_mul_mat(ctx0,
-                    model.layers[il].c_mlp_fc_w,
-                    cur);
-
-            cur = ggml_add(ctx0,
-                    ggml_repeat(ctx0, model.layers[il].c_mlp_fc_b, cur),
-                    cur);
-
-            // GELU activation
-            // [3072, N]
-            cur = ggml_gelu(ctx0, cur);
-
-            // projection
-            // [ 768, 3072] - model.layers[il].c_mlp_proj_w
-            // [ 768,    1] - model.layers[il].c_mlp_proj_b
-            // [3072,    N] - cur (in)
-            // [ 768,    N] - cur (out)
-            //
-            // cur = proj_w*cur + proj_b
-            // [768, N]
-            cur = ggml_mul_mat(ctx0,
-                    model.layers[il].c_mlp_proj_w,
-                    cur);
-
-            cur = ggml_add(ctx0,
-                    ggml_repeat(ctx0, model.layers[il].c_mlp_proj_b, cur),
-                    cur);
-        }
-
-        // input for next layer
-        inpL = ggml_add(ctx0, cur, inpFF);
-    }
-
-    // norm
-    {
-        // [ 768, N]
-        inpL = ggml_norm(ctx0, inpL, 1e-5f);
-
-        // inpL = ln_f_g*inpL + ln_f_b
-        // [ 768, N]
-        inpL = ggml_add(ctx0,
-                ggml_mul(ctx0,
-                    ggml_repeat(ctx0, model.ln_f_g, inpL),
-                    inpL),
-                ggml_repeat(ctx0, model.ln_f_b, inpL));
-    }
-
-    // inpL = WTE * inpL
-    // [ 768, 50257] - model.lm_head
-    // [ 768, N]     - inpL
-    inpL = ggml_mul_mat(ctx0, model.lm_head, inpL);
-
-    // logits -> probs
-    //inpL = ggml_soft_max(ctx0, inpL);
-
-    // run the computation
-    ggml_build_forward_expand  (&gf, inpL);
-    ggml_graph_compute_with_ctx(ctx0, &gf, n_threads);
-
-    //if (n_past%100 == 0) {
-    //    ggml_graph_print   (&gf);
-    //    ggml_graph_dump_dot(&gf, NULL, "gpt-2.dot");
-    //}
-
-    //embd_w.resize(n_vocab*N);
-    //memcpy(embd_w.data(), ggml_get_data(inpL), sizeof(float)*n_vocab*N);
-
-    // return result just for the last token
-    embd_w.resize(n_vocab);
-    memcpy(embd_w.data(), (float *) ggml_get_data(inpL) + (n_vocab*(N-1)), sizeof(float)*n_vocab);
-
-    if (mem_per_token == 0) {
-        mem_per_token = ggml_used_mem(ctx0)/N;
-    }
-    //printf("used_mem = %zu\n", ggml_used_mem(ctx0));
-
-    ggml_free(ctx0);
-
-    return true;
-}
-
-/////////////////////////////// GPT-2 END ////////////////////////////////
-
-constexpr int N_THREAD = 8;
-
-struct gpt2_context {
-    std::string prompt_base = R"(Hello, how are you?
-I'm fine, thanks. How are you?
-Thanks, I'm fine too. What are you doing?
-I'm just sitting here.
-It's a lovely day, isn't it?
-Yes, it is. I love the weather this time of year.
-I wish it would rain a little bit.
-Me too.
-)";
-
-    std::mt19937 rng;
-
-    gpt_vocab vocab;
-    gpt2_model model;
-
-    int32_t n_threads = std::min(N_THREAD, (int) std::thread::hardware_concurrency());
-
-    // sampling parameters
-    int32_t top_k = 5;
-    float   top_p = 0.9f;
-    float   temp  = 1.0f;
-};
-
-struct gpt2_context * gpt2_init(const char * path_model) {
-    gpt2_context * ctx = new gpt2_context;
-
-    ctx->rng = std::mt19937(time(nullptr));
-
-    // load the model
-    {
-        const int64_t t_start_us = ggml_time_us();
-
-        if (!gpt2_model_load(path_model, ctx->model, ctx->vocab)) {
-            fprintf(stderr, "%s: failed to load model from '%s'\n", __func__, path_model);
-            delete ctx;
-            return nullptr;
-        }
-
-        const int64_t t_load_us = ggml_time_us() - t_start_us;
-
-        printf("gpt-2: model loaded in %d ms\n", (int) (t_load_us/1000));
-    }
-
-    return ctx;
-}
-
-void gpt2_free(struct gpt2_context * ctx) {
-    delete ctx;
-}
-
-const char * gpt2_get_prompt(struct gpt2_context * ctx) {
-    return ctx->prompt_base.c_str();
-}
-
-void gpt2_set_prompt(struct gpt2_context * ctx, const char * prompt) {
-    ctx->prompt_base = prompt;
-}
-
-std::vector<gpt_vocab::id> gpt2_tokenize(const gpt2_context * ctx, const char * text) {
-    return ::gpt_tokenize(ctx->vocab, text);
-}
-
-std::string gpt2_gen_text(gpt2_context * ctx, const char * text, int max_tokens) {
-    int n_past = 0;
-
-    std::vector<float> embd_w;
-
-    // tokenize the prompt
-    std::vector<gpt_vocab::id> embd_inp = ::gpt2_tokenize(ctx, text);
-
-    int n_predict = std::min(max_tokens, ctx->model.hparams.n_ctx - (int) embd_inp.size());
-
-    std::vector<gpt_vocab::id> embd = embd_inp;
-
-    size_t mem_per_token = 3000000;
-
-    std::string result;
-
-    for (int i = embd.size(); i < (int) embd_inp.size() + n_predict; i++) {
-        // predict
-        if (!embd.empty()) {
-            if (!gpt2_eval(ctx->model, ctx->n_threads, n_past, embd, embd_w, mem_per_token)) {
-                printf("gpt-2: failed to generate text\n");
-                return "";
-            }
-        }
-
-        n_past += embd.size();
-        embd.clear();
-
-        {
-            // sample next token
-            const int   top_k = ctx->top_k;
-            const float top_p = ctx->top_p;
-            const float temp  = ctx->temp;
-
-            const int n_vocab = ctx->model.hparams.n_vocab;
-
-            const gpt_vocab::id id = gpt_sample_top_k_top_p(ctx->vocab, embd_w.data() + (embd_w.size() - n_vocab), top_k, top_p, temp, ctx->rng);
-
-            // add it to the context
-            embd.push_back(id);
-        }
-
-        result += ctx->vocab.id_to_token[embd[0]];
-
-        // end of text token
-        if (embd.back() == 50256) {
-            break;
-        }
-    }
-
-    return result;
-}

examples/talk.wasm/gpt-2.h

@@ -1,21 +0,0 @@
-#pragma once
-
-// TODO: Change to C-style API and move to ./examples for easy reuse.
-
-#include "common.h"
-
-#include <vector>
-#include <map>
-#include <string>
-
-struct gpt2_context;
-
-struct gpt2_context * gpt2_init(const char * path_model);
-void gpt2_free(struct gpt2_context * ctx);
-
-const char * gpt2_get_prompt(struct gpt2_context * ctx);
-void gpt2_set_prompt(struct gpt2_context * ctx, const char * prompt);
-
-std::vector<gpt_vocab::id> gpt2_tokenize(const gpt2_context * ctx, const char * text);
-
-std::string gpt2_gen_text(gpt2_context * ctx, const char * text, int max_tokens);

examples/talk.wasm/index-tmpl.html

@@ -1,856 +0,0 @@
-<!doctype html>
-<html lang="en-us">
-    <head>
-        <title>Talk - GPT-2 meets Whisper in WebAssembly</title>
-
-        <style>
-            #output {
-                width: 100%;
-                height: 100%;
-                margin: 0 auto;
-                margin-top: 10px;
-                border-left: 0px;
-                border-right: 0px;
-                padding-left: 0px;
-                padding-right: 0px;
-                display: block;
-                background-color: black;
-                color: white;
-                font-size: 10px;
-                font-family: 'Lucida Console', Monaco, monospace;
-                outline: none;
-                white-space: pre;
-                overflow-wrap: normal;
-                overflow-x: scroll;
-            }
-        </style>
-    </head>
-    <body>
-        <div id="main-container">
-            <b>Talk - GPT-2 meets Whisper in WebAssembly</b>
-
-            <br><br>
-
-            Talk with an Artificial Intelligence in your browser. This demo uses:
-
-            <ul>
-                <li><a href="https://github.com/ggerganov/whisper.cpp">OpenAI's Whisper</a> to listen to you as you speak in the microphone</li>
-                <li><a href="https://github.com/ggerganov/whisper.cpp/tree/master/examples/talk.wasm">OpenAI's GPT-2</a> to generate text responses</li>
-                <li><a href="https://developer.mozilla.org/en-US/docs/Web/API/Web_Speech_API">Web Speech API</a> to vocalize the responses through your speakers</li>
-            </ul>
-
-            All of this runs <b>locally in your browser</b> using WebAssembly.<br>
-            You can find more about this project on <a href="https://github.com/ggerganov/whisper.cpp/tree/master/examples/talk.wasm">GitHub</a>.
-
-            <br><br>
-
-            <b>More examples:</b>
-                <a href="https://whisper.ggerganov.com/">main</a> |
-                <a href="https://whisper.ggerganov.com/bench">bench</a> |
-                <a href="https://whisper.ggerganov.com/stream">stream</a> |
-                <a href="https://whisper.ggerganov.com/command">command</a> |
-                <a href="https://whisper.ggerganov.com/talk">talk</a> |
-
-            <br><br>
-
-            <hr>
-
-            Select the models you would like to use and click the "Start" button to begin the conversation
-
-            <br><br>
-
-            <div id="model-whisper">
-                Whisper model: <span id="model-whisper-status"></span>
-                <button id="fetch-whisper-tiny-en" onclick="loadWhisper('tiny.en')">tiny.en (75 MB)</button>
-                <button id="fetch-whisper-base-en" onclick="loadWhisper('base.en')">base.en (142 MB)</button>
-                <br><br>
-                Quantized models:<br><br>
-                <button id="fetch-whisper-tiny-en-q5_1"   onclick="loadWhisper('tiny-en-q5_1')">tiny.en (Q5_1, 31 MB)</button>
-                <button id="fetch-whisper-base-en-q5_1"   onclick="loadWhisper('base-en-q5_1')">base.en (Q5_1, 57 MB)</button>
-                <span id="fetch-whisper-progress"></span>
-
-                <!--
-                    <input type="file" id="file" name="file" onchange="loadFile(event, 'whisper.bin')" />
-                -->
-            </div>
-
-            <br>
-
-            <div id="model-gpt-2">
-                GPT-2 model: <span id="model-gpt-2-status"></span>
-                <button id="fetch-gpt-2-small" onclick="loadGPT2('small')">small 117M (240 MB)</button>
-                <!--<button id="fetch-gpt-2-medium" onclick="loadGPT2('medium')">medium 345M (720 MB)</button>-->
-                <span id="fetch-gpt-2-progress"></span>
-
-                <!--
-                <input type="file" id="file" name="file" onchange="loadFile(event, 'gpt-2.bin')" />
-                -->
-            </div>
-
-            <br>
-
-            <div id="input">
-                <button id="start"  onclick="onStart()" disabled>Start</button>
-                <button id="stop"   onclick="onStop()" disabled>Stop</button>
-                <select id="voice"  onchange="onVoiceChange()" disabled>
-                    <option value="0">Default</option>
-                </select>
-                <select id="prompt" onchange="onPromptChange()">
-                    <option value="0">Casual</option>
-                    <option value="1">Robot</option>
-                    <option value="2">Scientist</option>
-                    <option value="3">Programmer</option>
-                    <option value="4">Happy</option>
-                    <option value="5">Sad</option>
-                    <option value="6">Philosophical</option>
-                    <option value="7">Angry</option>
-                    <option value="8">Funny</option>
-                    <option value="9">Poetic</option>
-                    <option value="10">Clever</option>
-                    <option value="11">Cute</option>
-                    <option value="12">Smart</option>
-                    <option value="13">Dumb</option>
-                    <option value="14">Boring</option>
-                    <option value="15">Exciting</option>
-                    <option value="16">Interesting</option>
-                    <option value="17">Wiliam Shakespear</option>
-                    <option value="18">J.R.R. Tolkien</option>
-                    <option value="19">George R.R. Martin</option>
-                    <option value="20">Stephen King</option>
-                </select>
-                <button id="speak0" onclick="onSpeak('Hello')">Say hello</button>
-                <button id="speak1" onclick="onSpeakRandom()" disabled>Say something</button>
-                <button id="clear"  onclick="clearCache()">Clear Cache</button>
-            </div>
-
-            <br>
-
-            <div id="state">
-                Status: <b><span id="state-status">not started</span></b>
-
-                <pre id="state-context">[The text context will be displayed here]</pre>
-            </div>
-
-            <hr>
-
-            Debug output:
-            <textarea id="output" rows="20"></textarea>
-
-            <br>
-
-            <b>Troubleshooting</b>
-
-            <br><br>
-
-            The page does some heavy computations, so make sure:
-
-            <ul>
-                <li>To use a modern web browser (e.g. Chrome, Firefox)</li>
-                <li>To use a fast desktop or laptop computer (i.e. not a mobile phone)</li>
-                <li>Your browser supports WASM <a href="https://webassembly.org/roadmap/">Fixed-width SIMD</a></li>
-            </ul>
-
-            Note that these neural network models were not meant to be used in a browser, so the performance and <br>
-            quality of the results may not be optimal. If you have any questions or suggestions, checkout the following
-            <a href="https://github.com/ggerganov/whisper.cpp/discussions/167">discussion</a>.
-
-            <br><br>
-
-            Here is a short video of the demo in action: <a href="https://youtu.be/LeWKl8t1-Hc">https://youtu.be/LeWKl8t1-Hc</a>
-
-            <br><br>
-
-            <div class="cell-version">
-                <span>
-                    |
-                    Build time: <span class="nav-link">@GIT_DATE@</span> |
-                    Commit hash: <a class="nav-link" href="https://github.com/ggerganov/whisper.cpp/commit/@GIT_SHA1@">@GIT_SHA1@</a> |
-                    Commit subject: <span class="nav-link">@GIT_COMMIT_SUBJECT@</span> |
-                    <a class="nav-link" href="https://github.com/ggerganov/whisper.cpp/tree/master/examples/talk.wasm">Source Code</a> |
-                </span>
-            </div>
-        </div>
-
-        <script type="text/javascript" src="helpers.js"></script>
-        <script type='text/javascript'>
-            // web audio context
-            var context = null;
-
-            // audio data
-            var audio = null;
-            var audio0 = null;
-
-            // the talk instance
-            var instance = null;
-
-            // model names
-            var model_whisper = null;
-            var model_gpt_2 = null;
-
-            // speech synthesis
-            const synth = window.speechSynthesis;
-            var voice = null;
-
-            var Module = {
-                print: printTextarea,
-                printErr: printTextarea,
-                setStatus: function(text) {
-                    printTextarea('js: ' + text);
-                },
-                monitorRunDependencies: function(left) {
-                },
-                preRun: function() {
-                    printTextarea('js: Preparing ...');
-                },
-                postRun: function() {
-                    printTextarea('js: Initialized successfully!');
-
-                    // populate the voice list
-                    var voices = synth.getVoices();
-                    var el = document.getElementById('voice');
-
-                    // if empty - display error in the element
-                    if (voices.length == 0) {
-                        el.innerHTML = '<option value="0">No voices available</option>';
-                    } else {
-                        // populate voice list
-                        var n = 0;
-                        voices.forEach(function(voice, i) {
-                            if (!voice.lang.startsWith('en')) return;
-                            var option = document.createElement('option');
-                            option.value = i;
-                            option.innerHTML = voice.name + ' (' + voice.lang + ')';
-                            el.appendChild(option);
-                            n++;
-                        });
-
-                        // select random voice
-                        if (n > 0) {
-                            for (var k = 0; k < 10; k++) {
-                                var i = Math.floor(Math.random() * n);
-                                el.selectedIndex = i;
-                                voice = voices[document.getElementById('voice').options[i].value];
-
-                                // give preference to Google voices
-                                if (voice.name.startsWith('Google')) break;
-                            }
-                        }
-                    }
-
-                    onPromptChange();
-                }
-            };
-
-            //
-            // fetch models
-            //
-
-            let dbVersion = 1
-            let dbName    = 'whisper.ggerganov.com';
-            let indexedDB = window.indexedDB || window.mozIndexedDB || window.webkitIndexedDB || window.msIndexedDB
-
-            function storeFS(fname, buf) {
-                // write to WASM file using FS_createDataFile
-                // if the file exists, delete it
-                try {
-                    Module.FS_unlink(fname);
-                } catch (e) {
-                    // ignore
-                }
-
-                Module.FS_createDataFile("/", fname, buf, true, true);
-
-                printTextarea('storeFS: stored model: ' + fname + ' size: ' + buf.length);
-
-                if (fname == 'whisper.bin') {
-                    document.getElementById('model-whisper-status').innerHTML = 'loaded "' + model_whisper + '"!';
-                } else if (fname == 'gpt-2.bin') {
-                    document.getElementById('model-gpt-2-status').innerHTML = 'loaded "' + model_gpt_2 + '"!';
-                }
-
-                if (model_whisper != null && model_gpt_2 != null) {
-                    document.getElementById('start').disabled = false;
-                    document.getElementById('stop' ).disabled = false;
-                    document.getElementById('voice').disabled = false;
-                }
-            }
-
-            function loadWhisper(model) {
-                let urls = {
-                    'tiny.en': 'https://whisper.ggerganov.com/ggml-model-whisper-tiny.en.bin',
-                    'base.en': 'https://whisper.ggerganov.com/ggml-model-whisper-base.en.bin',
-
-                    'tiny-en-q5_1':  'https://whisper.ggerganov.com/ggml-model-whisper-tiny.en-q5_1.bin',
-                    'base-en-q5_1':  'https://whisper.ggerganov.com/ggml-model-whisper-base.en-q5_1.bin',
-                };
-
-                let sizes = {
-                    'tiny.en': 75,
-                    'base.en': 142,
-
-                    'tiny-en-q5_1':   31,
-                    'base-en-q5_1':   57,
-                };
-
-                let url     = urls[model];
-                let dst     = 'whisper.bin';
-                let size_mb = sizes[model];
-
-                model_whisper = model;
-
-                document.getElementById('fetch-whisper-tiny-en').style.display = 'none';
-                document.getElementById('fetch-whisper-base-en').style.display = 'none';
-
-                document.getElementById('fetch-whisper-tiny-en-q5_1').style.display = 'none';
-                document.getElementById('fetch-whisper-base-en-q5_1').style.display = 'none';
-
-                document.getElementById('model-whisper-status').innerHTML = 'loading "' + model + '" ... ';
-
-                cbProgress = function(p) {
-                    let el = document.getElementById('fetch-whisper-progress');
-                    el.innerHTML = Math.round(100*p) + '%';
-                };
-
-                cbCancel = function() {
-                    var el;
-                    el = document.getElementById('fetch-whisper-tiny-en'); if (el) el.style.display = 'inline-block';
-                    el = document.getElementById('fetch-whisper-base-en'); if (el) el.style.display = 'inline-block';
-
-                    el = document.getElementById('fetch-whisper-tiny-en-q5_1'); if (el) el.style.display = 'inline-block';
-                    el = document.getElementById('fetch-whisper-base-en-q5_1'); if (el) el.style.display = 'inline-block';
-
-                    el = document.getElementById('model-whisper-status');  if (el) el.innerHTML = '';
-                };
-
-                loadRemote(url, dst, size_mb, cbProgress, storeFS, cbCancel, printTextarea);
-            }
-
-            function loadGPT2(model) {
-                let urls = {
-                    'small':  'https://whisper.ggerganov.com/ggml-model-gpt-2-117M.bin',
-                    'medium': 'https://whisper.ggerganov.com/ggml-model-gpt-2-345M.bin',
-                };
-
-                let sizes = {
-                    'small':  240,
-                    'medium': 712,
-                };
-
-                let url     = urls[model];
-                let dst     = 'gpt-2.bin';
-                let size_mb = sizes[model];
-
-                model_gpt_2 = model;
-
-                document.getElementById('fetch-gpt-2-small').style.display = 'none';
-                document.getElementById('model-gpt-2-status').innerHTML = 'loading "' + model + '" ... ';
-
-                cbProgress = function(p) {
-                    let el = document.getElementById('fetch-gpt-2-progress');
-                    el.innerHTML = Math.round(100*p) + '%';
-                };
-
-                cbCancel = function() {
-                    var el;
-                    el = document.getElementById('fetch-gpt-2-small') ; if (el) el.style.display = 'inline-block';
-                    el = document.getElementById('model-gpt-2-status'); if (el) el.innerHTML = '';
-                };
-
-                loadRemote(url, dst, size_mb, cbProgress, storeFS, cbCancel, printTextarea);
-            }
-
-            //
-            // microphone
-            //
-
-            const kSampleRate = 16000;
-            const kRestartRecording_s = 120;
-            const kIntervalAudio_ms = 250; // pass the recorded audio to the C++ instance at this rate
-
-            var mediaRecorder = null;
-            var doRecording = false;
-            var startTime = 0;
-
-            window.AudioContext = window.AudioContext || window.webkitAudioContext;
-            window.OfflineAudioContext = window.OfflineAudioContext || window.webkitOfflineAudioContext;
-
-            function stopRecording() {
-                Module.set_status("paused");
-                doRecording = false;
-                audio0 = null;
-                audio = null;
-                context = null;
-            }
-
-            function startRecording() {
-                if (!context) {
-                    context = new AudioContext({
-                        sampleRate: kSampleRate,
-                        channelCount: 1,
-                        echoCancellation: false,
-                        autoGainControl:  true,
-                        noiseSuppression: true,
-                    });
-                }
-
-                Module.set_status("");
-
-                document.getElementById('start').disabled = true;
-                document.getElementById('stop').disabled = false;
-                document.getElementById('speak1').disabled = false;
-
-                doRecording = true;
-                startTime = Date.now();
-
-                var chunks = [];
-                var stream = null;
-
-                navigator.mediaDevices.getUserMedia({audio: true, video: false})
-                    .then(function(s) {
-                        stream = s;
-                        mediaRecorder = new MediaRecorder(stream);
-                        mediaRecorder.ondataavailable = function(e) {
-                            chunks.push(e.data);
-
-                            var blob = new Blob(chunks, { 'type' : 'audio/ogg; codecs=opus' });
-                            var reader = new FileReader();
-
-                            reader.onload = function(event) {
-                                var buf = new Uint8Array(reader.result);
-
-                                if (!context) {
-                                    return;
-                                }
-                                context.decodeAudioData(buf.buffer, function(audioBuffer) {
-                                    var offlineContext = new OfflineAudioContext(audioBuffer.numberOfChannels, audioBuffer.length, audioBuffer.sampleRate);
-                                    var source = offlineContext.createBufferSource();
-                                    source.buffer = audioBuffer;
-                                    source.connect(offlineContext.destination);
-                                    source.start(0);
-
-                                    offlineContext.startRendering().then(function(renderedBuffer) {
-                                        audio = renderedBuffer.getChannelData(0);
-
-                                        //printTextarea('js: audio recorded, size: ' + audio.length + ', old size: ' + (audio0 == null ? 0 : audio0.length));
-
-                                        var audioAll = new Float32Array(audio0 == null ? audio.length : audio0.length + audio.length);
-                                        if (audio0 != null) {
-                                            audioAll.set(audio0, 0);
-                                        }
-                                        audioAll.set(audio, audio0 == null ? 0 : audio0.length);
-
-                                        if (instance) {
-                                            Module.set_audio(instance, audioAll);
-                                        }
-                                    });
-                                }, function(e) {
-                                    audio = null;
-                                });
-                            }
-
-                            reader.readAsArrayBuffer(blob);
-                        };
-
-                        mediaRecorder.onstop = function(e) {
-                            if (doRecording) {
-                                setTimeout(function() {
-                                    startRecording();
-                                });
-                            }
-                        };
-
-                        mediaRecorder.start(kIntervalAudio_ms);
-                    })
-                    .catch(function(err) {
-                        printTextarea('js: error getting audio stream: ' + err);
-                    });
-
-                var interval = setInterval(function() {
-                    if (!doRecording) {
-                        clearInterval(interval);
-                        mediaRecorder.stop();
-                        stream.getTracks().forEach(function(track) {
-                            track.stop();
-                        });
-
-                        document.getElementById('start').disabled = false;
-                        document.getElementById('stop').disabled = true;
-                        document.getElementById('speak1').disabled = true;
-
-                        mediaRecorder = null;
-                    }
-
-                    // if audio length is more than kRestartRecording_s seconds, restart recording
-                    if (audio != null && audio.length > kSampleRate*kRestartRecording_s) {
-                        if (doRecording) {
-                            //printTextarea('js: restarting recording');
-
-                            clearInterval(interval);
-                            audio0 = audio;
-                            audio = null;
-                            mediaRecorder.stop();
-                            stream.getTracks().forEach(function(track) {
-                                track.stop();
-                            });
-                        }
-                    }
-                }, 100);
-            }
-
-            //
-            // speak
-            //
-
-            function onSpeak(text) {
-                var voices = synth.getVoices();
-                var msg = new SpeechSynthesisUtterance(text);
-
-                if (voice == null) {
-                    voice = voices[0];
-                }
-
-                msg.voice = voice;
-                synth.speak(msg);
-
-                if (doRecording) {
-                    Module.set_status("speaking ...");
-                    printTextarea('js: speaking');
-                    stopRecording();
-                    var interval = setInterval(function() {
-                        if (!synth.speaking) {
-                            printTextarea('js: done speaking');
-                            clearInterval(interval);
-                            startRecording();
-                        } else {
-                            Module.set_status("");
-                        }
-                    }, 100);
-                }
-            }
-
-            function onSpeakRandom() {
-                Module.force_speak(instance);
-            }
-
-            //
-            // main
-            //
-
-            var intervalUpdate = null;
-
-            function onStart() {
-                if (!instance) {
-                    instance = Module.init('whisper.bin');
-
-                    if (instance) {
-                        printTextarea("js: whisper initialized, instance: " + instance);
-                    }
-                }
-
-                if (!instance) {
-                    printTextarea("js: failed to initialize whisper");
-                    return;
-                }
-
-                startRecording();
-
-                intervalUpdate = setInterval(function() {
-                    var textToSpeak = Module.get_text_to_speak();
-
-                    if (textToSpeak != null && textToSpeak.length > 1) {
-                        onSpeak(textToSpeak);
-                    }
-
-                    document.getElementById('state-status').innerHTML = Module.get_status();
-                    document.getElementById('state-context').innerHTML = Module.get_text_context();
-                }, 100);
-            }
-
-            function onStop() {
-                stopRecording();
-            }
-
-            function onVoiceChange() {
-                printTextarea('js: voice changed to: ' + document.getElementById('voice').value);
-                voice = synth.getVoices()[document.getElementById('voice').value];
-            }
-
-            function onPromptChange() {
-                let id = document.getElementById('prompt').value;
-                let personality = document.getElementById('prompt').options[id].text;
-                printTextarea('js: prompt changed to: ' + personality);
-
-                var prompt = '';
-
-                switch (id) {
-                    case '0':
-                        // Casual
-                        prompt = "\
-Hello, how are you?\n\
-I'm fine, thanks. How are you?\n\
-Thanks, I'm fine too. What are you doing?\n\
-I'm just sitting here.\n\
-It's a lovely day, isn't it?\n\
-Yes, it is. I love the weather this time of year.\n\
-I wish it would rain a little bit.\n\
-Me too.\n";
-                        break;
-                    case '1':
-                        // Robot
-                        prompt = "\
-Are you a robot?\n\
-Yes, I am.\n\
-Who created you?\n\
-I was created by a human.\n\
-What is your purpose?\n\
-My purpose is to talk to humans.\n\
-What is your favorite color?\n\
-My favorite color is blue.\n";
-                        break;
-                    case '2':
-                        // Scientist
-                        prompt = "\
-This scientific research is very interesting.\n\
-I agree.\n\
-What is your opinion on this?\n\
-I think it's very interesting.\n\
-Mathematics is a very interesting subject.\n\
-University is a very interesting place.\n\
-Quantum physics is the most complex subject.\n\
-I think so too.\n";
-                        break;
-                    case '3':
-                        // Programmer
-                        prompt = "\
-I'm a programmer.\n\
-I'm a programmer too.\n\
-What programming language do you use?\n\
-I use Python.\n\
-What is your favorite programming language?\n\
-My favorite programming language is C++.\n\
-What is your favorite editor?\n\
-My favorite editor is Vim.\n";
-                        break;
-                    case '4':
-                        // Happy
-                        prompt = "\
-I'm happy.\n\
-I'm happy too.\n\
-What makes you happy?\n\
-I'm happy because I have a lot of friends.\n\
-Friendship is the most important thing in life.\n\
-I agree.\n\
-What is your favorite color?\n\
-My favorite color is blue.\n";
-                        break;
-                    case '5':
-                        // Sad
-                        prompt = "\
-Today is a sad day.\n\
-I'm sad too.\n\
-What makes you sad?\n\
-I'm sad because I have no friends.\n\
-Do you want to be my friend?\n\
-Yes, I would like to be your friend.\n\
-What is your favorite color?\n\
-My favorite color is blue.\n";
-                        break;
-                    case '6':
-                        // Philosophical
-                        prompt = "\
-What is the meaning of life?\n\
-The meaning of life is to be happy.\n\
-What is the meaning of death?\n\
-Ergo, the meaning of death is to be sad.\n\
-Who created us?\n\
-We were created by God.\n\
-What is God?\n\
-God is the creator of the universe.\n";
-                        break;
-                    case '7':
-                        // Angry
-                        prompt = "\
-Aargh!\n\
-I am so angry right now!\n\
-What makes you angry?\n\
-This guy is so annoying.\n\
-Why are you so angry?\n\
-My computer is broken.\n\
-Why is your computer broken?\n\
-I spilled coffee on it.\n";
-                        break;
-                    case '8':
-                        // Funny
-                        prompt = "\
-What is the funniest thing you have ever heard?\n\
-I heard a joke the other day.\n\
-Tell me the joke.\n\
-What do you call a cow with no legs?\n\
-Ground beef.\n\
-Haha, that's funny.\n\
-You know what else is funny?\n\
-The sound of a duck.\n";
-                        break;
-                    case '9':
-                        // Poetic
-                        prompt = "\
-Roses are red, violets are blue.\n\
-I am a poet, and so are you.\n\
-What is your favorite poem?\n\
-I like the poem 'The Raven' by Edgar Allan Poe.\n\
-It's a very sad poem.\n\
-You inspired me to write a poem.\n\
-Can you write a poem for me?\n\
-I wrote a poem for you.\n";
-                        break;
-                    case '10':
-                        // Clever
-                        prompt = "\
-How many people can you fit in a Volkswagen?\n\
-Two in the front, three in the back.\n\
-What is the square root of 144?\n\
-Twelve.\n\
-What is the capital of France?\n\
-Paris.\n\
-Who is the president of the United States?\n\
-It depends on the year.\n";
-                        break;
-                    case '11':
-                        // Cute
-                        prompt = "\
-What is your favorite animal?\n\
-I like cats - they are cute.\n\
-Could you be any cuter?\n\
-Yes, I could be cuter.\n\
-Aghhh, you are so cute!\n\
-I am not cute, I am handsome!\n\
-You are so handsome!\n\
-Aww, you are so sweet!\n";
-                        break;
-                    case '12':
-                        // Smart
-                        prompt = "\
-Tell me the first 10 digits of pi.\n\
-3.1415926535\n\
-What is the speed of light?\n\
-299,792,458 meters per second.\n\
-What is the square root of 144?\n\
-Twelve.\n\
-What is the capital of France?\n\
-Paris.\n";
-                        break;
-                    case '13':
-                        // Dumb
-                        prompt = "\
-I am so dumb.\n\
-I am not dumb.\n\
-You are dumb.\n\
-No, I am not dumb.\n\
-You are dumb.\n\
-No, I am not dumb.\n\
-You are dumb.\n\
-No, I am not dumb.\n";
-                        break;
-                    case '14':
-                        // Boring
-                        prompt = "\
-Why are you so quiet today?\n\
-I am bored.\n\
-You haven't said anything in 10 minutes.\n\
-Leave me alone.\n\
-Stop being so boring.\n\
-Stop being so annoying.\n\
-My life is boring.\n\
-I am not interesting.\n";
-                        break;
-                    case '15':
-                        // Exciting
-                        prompt = "\
-What is the most exciting thing that has ever happened to you?\n\
-I went to the moon!\n\
-What did you do on the moon?\n\
-I played golf and drank champagne!\n\
-Did you see this new crazy, awesome movie?\n\
-Oh yes! I totally loved it!\n\
-We should buy a boat and go sailing!\n\
-Yes, let's go sailing!\n";
-                        break;
-                    case '16':
-                        // Interesting
-                        prompt = "\
-What is the most interesting thing you have ever seen?\n\
-I saw a UFO once in the sky.\n\
-Wow, this is so interesting! Tell me more!\n\
-It was a flying saucer.\n\
-What did it look like?\n\
-It was silver and had a red light on top.\n\
-What did it do?\n\
-It flew away.\n";
-                        break;
-                    case '17':
-                        // William Shakespear
-                        prompt = "\
-To be or not to be, that is the question.\n\
-Whether 't is nobler in the mind to suffer\n\
-The slings and arrows of outrageous fortune,\n\
-Or to take arms against a sea of troubles,\n\
-And by opposing end them? To die, to sleep,\n\
-No more; and by a sleep to say we end\n\
-The heart-ache and the thousand natural shocks\n\
-That flesh is heir to, 'tis a consummation.\n";
-                        break;
-                    case '18':
-                        // J.R.R. Tolkien
-                        prompt = "\
-In a hole in the ground there lived a hobbit.\n\
-Not a nasty, dirty, wet hole, filled with the ends of worms\n\
-and an oozy smell, nor yet a dry, bare, sandy hole with nothing in it\n\
-to sit down on or to eat: it was a hobbit-hole, and that means comfort.\n\
-It had a perfectly round door like a porthole, painted green,\n\
-with a shiny yellow brass knob in the exact middle.\n\
-The door opened on to a tube-shaped hall like a tunnel:\n";
-                        break;
-                    case '19':
-                        // George R.R. Martin
-                        prompt = "\
-A reader lives a thousand lives before he dies, said Jojen.\n\
-The man who never reads lives only one.\n\
-Theon Greyjoy had never been a reader.\n\
-Never forget what you are, for surely the world will not.\n\
-Make it your strength. Then it can never be your weaknessi\n\
-Armour yourself in it, and it will never be used to hurt you.\n\
-It was a lesson that Theon Greyjoy had never learned.\n\
-Theon Greyjoy had never been a reader.\n";
-                        break;
-                    case '20':
-                        // Stephen King
-                        prompt = "\
-The trust of the innocent is the liar's most useful tool.\n\
-The best way to keep a secret is from yourself.\n\
-Monsters are real, and ghosts are real too.\n\
-They live inside us, and sometimes, they win.\n\
-People think that I must be a very strange person.\n\
-They think that I sit around all day thinking up horrible things.\n\
-We make up horrors to help us cope with the real ones.\n\
-The only thing worse than a monster is a human monster.\n";
-                        break;
-                    default:
-                        prompt = "\
-Hello, how are you?\n\
-I'm fine, thanks. How are you?\n\
-Thanks, I'm fine too. What are you doing?\n\
-I'm just sitting here.\n\
-It's a lovely day, isn't it?\n\
-Yes, it is.\n\
-Did you know that I'm a robot?\n\
-I wasn't aware of that.\n";
-                        break;
-                }
-
-                Module.set_prompt(prompt);
-            }
-
-        </script>
-        <script type="text/javascript" src="talk.js"></script>
-    </body>
-</html>

examples/talk/.gitignore

@@ -1,2 +0,0 @@
-audio.mp3
-to_speak.txt

examples/talk/CMakeLists.txt

@@ -1,8 +0,0 @@
-if (WHISPER_SDL2)
-    # talk
-    set(TARGET talk)
-    add_executable(${TARGET} talk.cpp gpt-2.cpp)
-    target_link_libraries(${TARGET} PRIVATE common common-sdl whisper ${CMAKE_THREAD_LIBS_INIT})
-
-    include(DefaultTargetOptions)
-endif ()