* Enhance autogptq backend to support VL models
* update dependencies for autogptq
* remove redundant auto-gptq dependency
* Convert base64 to image_url for Qwen-VL model
* implemented model inference for qwen-vl
* remove user prompt from generated answer
* fixed write image error
* fixed use_triton issue when loading Qwen-VL model
---------
Co-authored-by: Binghua Wu <bingwu@estee.com>
* Streaming working
* Small fix for regression on CUDA and XPU
* use pip version of optimum[openvino]
* Update backend/python/transformers/transformers_server.py
Signed-off-by: Ettore Di Giacinto <mudler@users.noreply.github.com>
* Token streaming support
fix optimum[openvino] package in install.sh
* Token Streaming support
---------
Signed-off-by: Ettore Di Giacinto <mudler@users.noreply.github.com>
Co-authored-by: Ettore Di Giacinto <mudler@users.noreply.github.com>
* fixes#1775 and #1774
Add BitsAndBytes Quantization and fixes embedding on CUDA devices
* Manage 4bit and 8 bit quantization
Manage different BitsAndBytes options with the quantization: parameter in yaml
* fix compilation errors on non CUDA environment
* OpenVINO draft
First draft of OpenVINO integration in transformer backend
* first working implementation
* Streaming working
* Small fix for regression on CUDA and XPU
* use pip version of optimum[openvino]
* Update backend/python/transformers/transformers_server.py
Signed-off-by: Ettore Di Giacinto <mudler@users.noreply.github.com>
---------
Signed-off-by: Ettore Di Giacinto <mudler@users.noreply.github.com>
Co-authored-by: Ettore Di Giacinto <mudler@users.noreply.github.com>
* Enhance autogptq backend to support VL models
* update dependencies for autogptq
* remove redundant auto-gptq dependency
* Convert base64 to image_url for Qwen-VL model
* implemented model inference for qwen-vl
* remove user prompt from generated answer
* fixed write image error
---------
Co-authored-by: Binghua Wu <bingwu@estee.com>
* test with gguf instead of ggml. Updates testPrompt to match? Adds debugging line to Dockerfile that I've found helpful recently.
* fix testPrompt slightly
* Sad Experiment: Test GH runner without metal?
* break apart CGO_LDFLAGS
* switch runner
* upstream llama.cpp disables Metal on Github CI!
* missed a dir from clean-tests
* CGO_LDFLAGS
* tmate failure + NO_ACCELERATE
* whisper.cpp has a metal fix
* do the exact opposite of the name of this branch, but keep it around for unrelated fixes?
* add back newlines
* add tmate to linux for testing
* update fixtures
* timeout for tmate
* fix: clean up Makefile dependencies to allow for parallel builds
* refactor: remove old unused backend from Makefile
* fix: finish removing legacy backend, update piper
* fix: I broke llama... I fixed llama
* feat: give the tests and builds a few threads
* fix: ensure libraries are replaced before build, add dropreplace target
* Fix image build workflows
* feat(elevenlabs): map elevenlabs API support to TTS
This allows elevenlabs Clients to work automatically with LocalAI by
supporting the elevenlabs API.
The elevenlabs server endpoint is implemented such as it is wired to the
TTS endpoints.
Fixes: https://github.com/mudler/LocalAI/issues/1809
* feat(openai/tts): compat layer with openai tts
Fixes: #1276
* fix: adapt tts CLI
* fixes#1775 and #1774
Add BitsAndBytes Quantization and fixes embedding on CUDA devices
* Manage 4bit and 8 bit quantization
Manage different BitsAndBytes options with the quantization: parameter in yaml
* fix compilation errors on non CUDA environment
* feat(intel): add diffusers support
* try to consume upstream container image
* Debug
* Manually install deps
* Map transformers/hf cache dir to modelpath if not specified
* fix(compel): update initialization, pass by all gRPC options
* fix: add dependencies, implement transformers for xpu
* base it from the oneapi image
* Add pillow
* set threads if specified when launching the API
* Skip conda install if intel
* defaults to non-intel
* ci: add to pipelines
* prepare compel only if enabled
* Skip conda install if intel
* fix cleanup
* Disable compel by default
* Install torch 2.1.0 with Intel
* Skip conda on some setups
* Detect python
* Quiet output
* Do not override system python with conda
* Prefer python3
* Fixups
* exllama2: do not install without conda (overrides pytorch version)
* exllama/exllama2: do not install if not using cuda
* Add missing dataset dependency
* Small fixups, symlink to python, add requirements
* Add neural_speed to the deps
* correctly handle model offloading
* fix: device_map == xpu
* go back at calling python, fixed at dockerfile level
* Exllama2 restricted to only nvidia gpus
* Tokenizer to xpu
* fix: use vllm AsyncLLMEngine to bring true stream
Current vLLM implementation uses the LLMEngine, which was designed for offline batch inference, which results in the streaming mode outputing all blobs at once at the end of the inference.
This PR reworks the gRPC server to use asyncio and gRPC.aio, in combination with vLLM's AsyncLLMEngine to bring true stream mode.
This PR also passes more parameters to vLLM during inference (presence_penalty, frequency_penalty, stop, ignore_eos, seed, ...).
* Remove unused import
This PR specifically introduces a `core` folder and moves the following packages over, without any other changes:
- `api/backend`
- `api/config`
- `api/options`
- `api/schema`
Once this is merged and we confirm there's no regressions, I can migrate over the remaining changes piece by piece to split up application startup, backend services, http, and mqtt as was the goal of the earlier PRs!
* Dockerfile changes to build for ROCm
* Adjust linker flags for ROCm
* Update conda env for diffusers and transformers to use ROCm pytorch
* Update transformers conda env for ROCm
* ci: build hipblas images
* fixup rebase
* use self-hosted
Signed-off-by: mudler <mudler@localai.io>
* specify LD_LIBRARY_PATH only when BUILD_TYPE=hipblas
---------
Signed-off-by: mudler <mudler@localai.io>
Co-authored-by: mudler <mudler@localai.io>
Infinite context loop might as well trigger an infinite loop of context
shifting if the model hallucinates and does not stop answering.
This has the unpleasant effect that the predicion never terminates,
which is the case especially on small models which tends to hallucinate.
Workarounds https://github.com/mudler/LocalAI/issues/1333 by removing
context-shifting.
See also upstream issue: https://github.com/ggerganov/llama.cpp/issues/3969
* feat(refactor): refactor config and input reading
* feat(tts): read config file for TTS
* examples(kubernetes): Add simple deployment example
* examples(kubernetes): Add simple deployment for intel arc
* docs(sycl): add sycl example
* feat(tts): do not always pick a first model
* fixups to run vall-e-x on container
* Correctly resolve backend
* cleanup backends
* switch image to ubuntu 22.04
* adapt commands for ubuntu
* transformers cleanup
* no contrib on ubuntu
* Change test model to gguf
* ci: disable bark tests (too cpu-intensive)
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* cleanup
* refinements
* use intel base image
* Makefile: Add docker targets
* Change test model
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* feat(transformers): support also text generation
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* embedded: set seed -1
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Certain backends as vall-e-x are not meant to be used as a library, so
we want to start the process in the same folder where the backend and
all the assets are fixes#1394
* feat(conda): share env between diffusers and bark
* Detect if env already exists
* share diffusers and petals
* tests: add petals
* Use smaller model for tests with petals
* test only model load on petals
* tests(petals): run only load model tests
* Revert "test only model load on petals"
This reverts commit 111cfa97f1.
* move transformers and sentencetransformers to common env
* Share also transformers-musicgen
* feat(img2vid): Initial support for img2vid
* doc(SD): fix SDXL Example
* Minor fixups for img2vid
* docs(img2img): fix example curl call
* feat(txt2vid): initial support
Signed-off-by: Ettore Di Giacinto <mudler@users.noreply.github.com>
* diffusers: be retro-compatible with CUDA settings
* docs(img2vid, txt2vid): examples
* Add notice on docs
---------
Signed-off-by: Ettore Di Giacinto <mudler@users.noreply.github.com>
* Use cuda in transformers if available
tensorflow probably needs a different check.
Signed-off-by: Erich Schubert <kno10@users.noreply.github.com>
* feat: expose CUDA at top level
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* tests: add to tests and create workflow for py extra backends
* doc: update note on how to use core images
---------
Signed-off-by: Erich Schubert <kno10@users.noreply.github.com>
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Erich Schubert <kno10@users.noreply.github.com>
* Update docs for new requirements.txt path
Signed-off-by: Marcus Köhler <khler.marcus@gmail.com>
* Fix typo (.PONY -> .PHONY) in python backend makefiles
Signed-off-by: Marcus Köhler <khler.marcus@gmail.com>
---------
Signed-off-by: Marcus Köhler <khler.marcus@gmail.com>
* Fix python header comments for some extra gRPC backends
When a Python script is to be executed directly via exec(3), either the platform knows how to execute
the file itself (i.e. special configuration is necessary) or the first line
contains a shebang (#!) specifying the interpreter to run it (similar to
shell scripts).
The shebang MUST be on the first line for the script to work on all platforms,
so any header comments need to be in the lines following it. Otherwise
executing these scripts as extra backends will yield an "exec format
error" message.
Changes:
* Move introductory comments below the shebang line
* Change header comment in transformers.py to refer to the correct
python module
Signed-off-by: Marcus Köhler <khler.marcus@gmail.com>
* Make header comment in ttsbark.py more specific
Signed-off-by: Marcus Köhler <khler.marcus@gmail.com>
---------
Signed-off-by: Marcus Köhler <khler.marcus@gmail.com>
* Update huggingface.py
Switch SentenceTransformer for AutoModel in order to set trust_remote_code needed to use the encode method with embeddings models like jinai-v2
Signed-off-by: Lucas Hänke de Cansino <lhc@next-boss.eu>
* feat(transformers): split in separate backend
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
---------
Signed-off-by: Lucas Hänke de Cansino <lhc@next-boss.eu>
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
Co-authored-by: Lucas Hänke de Cansino <lhc@next-boss.eu>
* refactor: rename llama-stable to llama-ggml
* Makefile: get sources in sources/
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fixup path
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fixup sources
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fixups sd
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* update SD
* fixup
* fixup: create piper libdir also when not built
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix make target on linux test
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* refactor: move backends into the backends directory
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* refactor: move main close to implementation for every backend
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* feat(llama.cpp): support lora with scale
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* feat(llama.cpp): support yarn
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* wip
* wip
* Make it functional
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* wip
* Small fixups
* do not inject space on role encoding, encode img at beginning of messages
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* Add examples/config defaults
* Add include dir of current source dir
* cleanup
* fixes
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fixups
* Revert "fixups"
This reverts commit f1a4731cca.
* fixes
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* Current state of the branch.
* Now gRPC is build only when the BUILD_GRPC_FOR_BACKEND_LLAMA variable is defined.
* Now the local compilation of gRPC is executed on BUILD_GRPC_FOR_BACKEND_LLAMA.
* Revised the Makefile.
* Removed replace directives in go.mod.
---------
Signed-off-by: Diego <38375572+diego-minguzzi@users.noreply.github.com>
Co-authored-by: lunamidori5 <118759930+lunamidori5@users.noreply.github.com>
Co-authored-by: Ettore Di Giacinto <mudler@users.noreply.github.com>
* Fix backend/cpp/llama CMakeList.txt on OSX - detect OSX and use homebrew libraries
* sneak a logging fix in too for gallery debugging
* additional logging
* wip: llama.cpp c++ gRPC server
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* make it work, attach it to the build process
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* update deps
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* fix: add protobuf dep
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* try fix protobuf on cmake
* cmake: workarounds
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* add packages
* cmake: use fixed version of grpc
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* cmake(grpc): install locally
* install grpc
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* install required deps for grpc on debian bullseye
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* debug
* debug
* Fixups
* no need to install cmake manually
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* ci: fixup macOS
* use brew whenever possible
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* macOS fixups
* debug
* fix container build
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
* workaround
* try mac
https://stackoverflow.com/questions/23905661/on-mac-g-clang-fails-to-search-usr-local-include-and-usr-local-lib-by-def
* Disable temp. arm64 docker image builds
---------
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
