fix(ci): install latest git

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
2026-02-03 11:13:31 -05:00 · 2024-10-24 14:55:24 +02:00
371 changed files with 11396 additions and 10408 deletions
--- a/Requests/model
+++ b/Requests/model
@@ -1,11 +0,0 @@
-meta {
-  name: model delete
-  type: http
-  seq: 7
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models/galleries
-  body: none
-  auth: none
-}
--- a/Requests/transcription/gb1.ogg
+++ b/Requests/transcription/gb1.ogg
--- a/Requests/transcription/transcribe.bru
+++ b/Requests/transcription/transcribe.bru
@@ -1,16 +0,0 @@
-meta {
-  name: transcribe
-  type: http
-  seq: 1
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/v1/audio/transcriptions
-  body: multipartForm
-  auth: none
-}
-
-body:multipart-form {
-  file: @file(transcription/gb1.ogg)
-  model: whisper-1
-}
--- a/.env
+++ b/.env
@@ -82,15 +82,6 @@
 # Enable to allow p2p mode
 # LOCALAI_P2P=true

-# Enable to use federated mode
-# LOCALAI_FEDERATED=true
-
-# Enable to start federation server
-# FEDERATED_SERVER=true
-
-# Define to use federation token
-# TOKEN=""
-
 ### Watchdog settings
 ###
 # Enables watchdog to kill backends that are inactive for too much time
--- a/.gitattributes
+++ b/.gitattributes
@@ -1,2 +1 @@
 *.sh text eol=lf
-backend/cpp/llama/*.hpp linguist-vendored
--- a/.github/labeler.yml
+++ b/.github/labeler.yml
@@ -1,15 +1,6 @@
 enhancements:
 - head-branch: ['^feature', 'feature']

-dependencies:
- any:
-  - changed-files:
-    - any-glob-to-any-file: 'Makefile'
-  - changed-files:
-    - any-glob-to-any-file: '*.mod'
-  - changed-files:
-    - any-glob-to-any-file: '*.sum'
-
 kind/documentation:
 - any:
  - changed-files:
--- a/.github/workflows/bump_deps.yaml
+++ b/.github/workflows/bump_deps.yaml
@@ -12,14 +12,23 @@ jobs:
          - repository: "ggerganov/llama.cpp"
            variable: "CPPLLAMA_VERSION"
            branch: "master"
+          - repository: "go-skynet/go-ggml-transformers.cpp"
+            variable: "GOGGMLTRANSFORMERS_VERSION"
+            branch: "master"
+          - repository: "donomii/go-rwkv.cpp"
+            variable: "RWKV_VERSION"
+            branch: "main"
          - repository: "ggerganov/whisper.cpp"
            variable: "WHISPER_CPP_VERSION"
            branch: "master"
-          - repository: "PABannier/bark.cpp"
-            variable: "BARKCPP_VERSION"
+          - repository: "go-skynet/go-bert.cpp"
+            variable: "BERT_VERSION"
+            branch: "master"
+          - repository: "go-skynet/bloomz.cpp"
+            variable: "BLOOMZ_VERSION"
            branch: "main"
-          - repository: "leejet/stable-diffusion.cpp"
-            variable: "STABLEDIFFUSION_GGML_VERSION"
+          - repository: "mudler/go-ggllm.cpp"
+            variable: "GOGGLLM_VERSION"
            branch: "master"
          - repository: "mudler/go-stable-diffusion"
            variable: "STABLEDIFFUSION_VERSION"
--- a/.github/workflows/checksum_checker.yaml
+++ b/.github/workflows/checksum_checker.yaml
@@ -23,7 +23,7 @@ jobs:
          sudo pip install --upgrade pip
          pip install huggingface_hub
      - name: 'Setup yq'
-        uses: dcarbone/install-yq-action@v1.3.1
+        uses: dcarbone/install-yq-action@v1.1.1
        with:
          version: 'v4.44.2'
          download-compressed: true
--- a/.github/workflows/deploy-explorer.yaml
+++ b/.github/workflows/deploy-explorer.yaml
@@ -33,7 +33,7 @@ jobs:
        run: |
          CGO_ENABLED=0 make build-api
      - name: rm
-        uses: appleboy/ssh-action@v1.2.0
+        uses: appleboy/ssh-action@v1.1.0
        with:
            host: ${{ secrets.EXPLORER_SSH_HOST }}
            username: ${{ secrets.EXPLORER_SSH_USERNAME }}
@@ -53,7 +53,7 @@ jobs:
            rm: true
            target: ./local-ai
      - name: restarting
-        uses: appleboy/ssh-action@v1.2.0
+        uses: appleboy/ssh-action@v1.1.0
        with:
            host: ${{ secrets.EXPLORER_SSH_HOST }}
            username: ${{ secrets.EXPLORER_SSH_USERNAME }}
--- a/.github/workflows/generate_intel_image.yaml
+++ b/.github/workflows/generate_intel_image.yaml
@@ -15,7 +15,7 @@ jobs:
    strategy:
      matrix:
        include:
-          - base-image: intel/oneapi-basekit:2025.0.0-0-devel-ubuntu22.04
+          - base-image: intel/oneapi-basekit:2024.2.0-devel-ubuntu22.04
            runs-on: 'ubuntu-latest'
            platforms: 'linux/amd64'
    runs-on: ${{matrix.runs-on}}
--- a/.github/workflows/image.yml
+++ b/.github/workflows/image.yml
@@ -280,7 +280,6 @@ jobs:
      makeflags: ${{ matrix.makeflags }}
      latest-image: ${{ matrix.latest-image }}
      latest-image-aio: ${{ matrix.latest-image-aio }}
-      skip-drivers: ${{ matrix.skip-drivers }}
    secrets:
      dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
      dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
@@ -302,7 +301,6 @@ jobs:
            latest-image: 'latest-cpu'
            latest-image-aio: 'latest-aio-cpu'
            makeflags: "--jobs=4 --output-sync=target"
-            skip-drivers: 'false'
          - build-type: 'cublas'
            cuda-major-version: "11"
            cuda-minor-version: "7"
@@ -314,7 +312,6 @@ jobs:
            base-image: "ubuntu:22.04"
            runs-on: 'arc-runner-set'
            makeflags: "--jobs=4 --output-sync=target"
-            skip-drivers: 'false'
          - build-type: 'cublas'
            cuda-major-version: "12"
            cuda-minor-version: "0"
@@ -326,7 +323,6 @@ jobs:
            base-image: "ubuntu:22.04"
            runs-on: 'arc-runner-set'
            makeflags: "--jobs=4 --output-sync=target"
-            skip-drivers: 'false'
          - build-type: 'cublas'
            cuda-major-version: "11"
            cuda-minor-version: "7"
@@ -338,7 +334,6 @@ jobs:
            runs-on: 'arc-runner-set'
            base-image: "ubuntu:22.04"
            makeflags: "--jobs=4 --output-sync=target"
-            skip-drivers: 'false'
          - build-type: 'cublas'
            cuda-major-version: "12"
            cuda-minor-version: "0"
@@ -349,7 +344,6 @@ jobs:
            image-type: 'core'
            runs-on: 'arc-runner-set'
            base-image: "ubuntu:22.04"
-            skip-drivers: 'false'
            makeflags: "--jobs=4 --output-sync=target"
          - build-type: 'vulkan'
            platforms: 'linux/amd64'
@@ -360,45 +354,4 @@ jobs:
            image-type: 'core'
            runs-on: 'arc-runner-set'
            base-image: "ubuntu:22.04"
-            skip-drivers: 'false'
            makeflags: "--jobs=4 --output-sync=target"
-#  parallel-builds:
-#    uses: ./.github/workflows/image_build.yml
-#    with:
-#      tag-latest: ${{ matrix.tag-latest }}
-#      tag-suffix: ${{ matrix.tag-suffix }}
-#      ffmpeg: ${{ matrix.ffmpeg }}
-#      image-type: ${{ matrix.image-type }}
-#      build-type: ${{ matrix.build-type }}
-#      cuda-major-version: ${{ matrix.cuda-major-version }}
-#      cuda-minor-version: ${{ matrix.cuda-minor-version }}
-#      platforms: ${{ matrix.platforms }}
-#      runs-on: ${{ matrix.runs-on }}
-#      aio: ${{ matrix.aio }}
-#      base-image: ${{ matrix.base-image }}
-#      grpc-base-image: ${{ matrix.grpc-base-image }}
-#      makeflags: ${{ matrix.makeflags }}
-#      latest-image: ${{ matrix.latest-image }}
-#      latest-image-aio: ${{ matrix.latest-image-aio }}
-#      skip-drivers: ${{ matrix.skip-drivers }}
-#    secrets:
-#      dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
-#      dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
-#      quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
-#      quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
-#    strategy:
-#      matrix:
-#        include:
-#          - build-type: 'cublas'
-#            cuda-major-version: "12"
-#            cuda-minor-version: "0"
-#            platforms: 'linux/arm64'
-#            tag-latest: 'false'
-#            tag-suffix: '-nvidia-l4t-arm64-core'
-#            latest-image: 'latest-nvidia-l4t-arm64-core'
-#            ffmpeg: 'true'
-#            image-type: 'core'
-#            base-image: "nvcr.io/nvidia/l4t-jetpack:r36.4.0"
-#            runs-on: 'self-hosted'
-#            makeflags: "--jobs=4 --output-sync=target"
-#            skip-drivers: 'true'
--- a/.github/workflows/image_build.yml
+++ b/.github/workflows/image_build.yml
@@ -49,10 +49,6 @@ on:
        description: 'FFMPEG'
        default: ''
        type: string
-      skip-drivers:
-        description: 'Skip drivers by default'
-        default: 'false'
-        type: string
      image-type:
        description: 'Image type'
        default: ''
@@ -238,7 +234,6 @@ jobs:
            GRPC_MAKEFLAGS=--jobs=4 --output-sync=target
            GRPC_VERSION=v1.65.0
            MAKEFLAGS=${{ inputs.makeflags }}
-            SKIP_DRIVERS=${{ inputs.skip-drivers }}
          context: .
          file: ./Dockerfile
          cache-from: type=gha
@@ -267,7 +262,6 @@ jobs:
            GRPC_MAKEFLAGS=--jobs=4 --output-sync=target
            GRPC_VERSION=v1.65.0
            MAKEFLAGS=${{ inputs.makeflags }}
-            SKIP_DRIVERS=${{ inputs.skip-drivers }}
          context: .
          file: ./Dockerfile
          cache-from: type=gha
--- a/.github/workflows/secscan.yaml
+++ b/.github/workflows/secscan.yaml
@@ -18,7 +18,7 @@ jobs:
        if: ${{ github.actor != 'dependabot[bot]' }}
      - name: Run Gosec Security Scanner
        if: ${{ github.actor != 'dependabot[bot]' }}
-        uses: securego/gosec@v2.22.0
+        uses: securego/gosec@v2.21.4
        with:
          # we let the report trigger content trigger a failure using the GitHub Security features.
          args: '-no-fail -fmt sarif -out results.sarif ./...'
--- a/.github/workflows/test-extra.yml
+++ b/.github/workflows/test-extra.yml
@@ -105,6 +105,14 @@ jobs:
  tests-parler-tts:
    runs-on: ubuntu-latest
    steps:
+      - name: Force Install GIT latest
+        run: |
+          sudo apt-get update \
+          && sudo apt-get install -y software-properties-common \
+          && sudo apt-get update \
+          && sudo add-apt-repository -y ppa:git-core/ppa \
+          && sudo apt-get update \
+          && sudo apt-get install -y git
      - name: Clone
        uses: actions/checkout@v4
        with:
@@ -123,13 +131,6 @@ jobs:
        run: |
           make --jobs=5 --output-sync=target -C backend/python/parler-tts
           make --jobs=5 --output-sync=target -C backend/python/parler-tts test
-      - name: Setup tmate session if tests fail
-        if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.19
-        with:
-          detached: true
-          connect-timeout-seconds: 180
-          limit-access-to-actor: true

  tests-openvoice:
    runs-on: ubuntu-latest
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -224,7 +224,7 @@ jobs:
      - name: Dependencies
        run: |
          brew install protobuf grpc make protoc-gen-go protoc-gen-go-grpc libomp llvm
-          pip install --user --no-cache-dir grpcio-tools
+          pip install --user --no-cache-dir grpcio-tools==1.64.1
      - name: Test
        run: |
          export C_INCLUDE_PATH=/usr/local/include
--- a/.gitignore
+++ b/.gitignore
@@ -2,7 +2,6 @@
 /sources/
 __pycache__/
 *.a
-*.o
 get-sources
 prepare-sources
 /backend/cpp/llama/grpc-server
@@ -13,6 +12,7 @@ prepare-sources

 go-ggml-transformers
 go-gpt2
+go-rwkv
 whisper.cpp
 /bloomz
 go-bert
--- a/12
+++ b/12
@@ -85,8 +85,7 @@ WORKDIR /build
 # The requirements-extras target is for any builds with IMAGE_TYPE=extras. It should not be placed in this target unless every IMAGE_TYPE=extras build will use it
 FROM requirements-core AS requirements-extras

-# Install uv as a system package
-RUN curl -LsSf https://astral.sh/uv/install.sh | UV_INSTALL_DIR=/usr/bin sh
+RUN curl -LsSf https://astral.sh/uv/install.sh | sh
 ENV PATH="/root/.cargo/bin:${PATH}"

 RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
@@ -115,13 +114,12 @@ FROM requirements-${IMAGE_TYPE} AS requirements-drivers
 ARG BUILD_TYPE
 ARG CUDA_MAJOR_VERSION=12
 ARG CUDA_MINOR_VERSION=0
-ARG SKIP_DRIVERS=false

 ENV BUILD_TYPE=${BUILD_TYPE}

 # Vulkan requirements
 RUN <<EOT bash
-    if [ "${BUILD_TYPE}" = "vulkan" ] && [ "${SKIP_DRIVERS}" = "false" ]; then
+    if [ "${BUILD_TYPE}" = "vulkan" ]; then
        apt-get update && \
        apt-get install -y  --no-install-recommends \
            software-properties-common pciutils wget gpg-agent && \
@@ -137,7 +135,7 @@ EOT

 # CuBLAS requirements
 RUN <<EOT bash
-    if [ "${BUILD_TYPE}" = "cublas" ] && [ "${SKIP_DRIVERS}" = "false" ]; then
+    if [ "${BUILD_TYPE}" = "cublas" ]; then
        apt-get update && \
        apt-get install -y  --no-install-recommends \
            software-properties-common pciutils
@@ -163,7 +161,7 @@ RUN <<EOT bash
 EOT

 # If we are building with clblas support, we need the libraries for the builds
-RUN if [ "${BUILD_TYPE}" = "clblas" ] && [ "${SKIP_DRIVERS}" = "false" ]; then \
+RUN if [ "${BUILD_TYPE}" = "clblas" ]; then \
        apt-get update && \
        apt-get install -y --no-install-recommends \
            libclblast-dev && \
@@ -171,7 +169,7 @@ RUN if [ "${BUILD_TYPE}" = "clblas" ] && [ "${SKIP_DRIVERS}" = "false" ]; then \
        rm -rf /var/lib/apt/lists/* \
    ; fi

-RUN if [ "${BUILD_TYPE}" = "hipblas" ] && [ "${SKIP_DRIVERS}" = "false" ]; then \
+RUN if [ "${BUILD_TYPE}" = "hipblas" ]; then \
        apt-get update && \
        apt-get install -y --no-install-recommends \
            hipblas-dev \
--- a/188
+++ b/188
@@ -8,15 +8,23 @@ DETECT_LIBS?=true
 # llama.cpp versions
 GOLLAMA_REPO?=https://github.com/go-skynet/go-llama.cpp
 GOLLAMA_VERSION?=2b57a8ae43e4699d3dc5d1496a1ccd42922993be
-CPPLLAMA_VERSION?=504af20ee4eae72080a56d59d744f6774f7901ce
+CPPLLAMA_VERSION?=0a1c750c80147687df267114c81956757cc14382
+
+# go-rwkv version
+RWKV_REPO?=https://github.com/donomii/go-rwkv.cpp
+RWKV_VERSION?=661e7ae26d442f5cfebd2a0881b44e8c55949ec6

 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggerganov/whisper.cpp
-WHISPER_CPP_VERSION?=6266a9f9e56a5b925e9892acf650f3eb1245814d
+WHISPER_CPP_VERSION?=0fbaac9c891055796456df7b9122a70c220f9ca1
+
+# bert.cpp version
+BERT_REPO?=https://github.com/go-skynet/go-bert.cpp
+BERT_VERSION?=710044b124545415f555e4260d16b146c725a6e4

 # go-piper version
 PIPER_REPO?=https://github.com/mudler/go-piper
-PIPER_VERSION?=e10ca041a885d4a8f3871d52924b47792d5e5aa0
+PIPER_VERSION?=9d0100873a7dbb0824dfea40e8cec70a1b110759

 # stablediffusion version
 STABLEDIFFUSION_REPO?=https://github.com/mudler/go-stable-diffusion
@@ -26,18 +34,6 @@ STABLEDIFFUSION_VERSION?=4a3cd6aeae6f66ee57eae9a0075f8c58c3a6a38f
 TINYDREAM_REPO?=https://github.com/M0Rf30/go-tiny-dream
 TINYDREAM_VERSION?=c04fa463ace9d9a6464313aa5f9cd0f953b6c057

-# bark.cpp
-BARKCPP_REPO?=https://github.com/PABannier/bark.cpp.git
-BARKCPP_VERSION?=v1.0.0
-
-# stablediffusion.cpp (ggml)
-STABLEDIFFUSION_GGML_REPO?=https://github.com/leejet/stable-diffusion.cpp
-STABLEDIFFUSION_GGML_VERSION?=dcf91f9e0f2cbf9da472ee2a556751ed4bab2d2a
-
-ONNX_VERSION?=1.20.0
-ONNX_ARCH?=x64
-ONNX_OS?=linux
-
 export BUILD_TYPE?=
 export STABLE_BUILD_TYPE?=$(BUILD_TYPE)
 export CMAKE_ARGS?=
@@ -49,7 +45,6 @@ CGO_LDFLAGS_WHISPER+=-lggml
 CUDA_LIBPATH?=/usr/local/cuda/lib64/
 GO_TAGS?=
 BUILD_ID?=
-NATIVE?=false

 TEST_DIR=/tmp/test

@@ -88,25 +83,7 @@ ifndef UNAME_S
 UNAME_S := $(shell uname -s)
 endif

-# IF native is false, we add -DGGML_NATIVE=OFF to CMAKE_ARGS
-ifeq ($(NATIVE),false)
-	CMAKE_ARGS+=-DGGML_NATIVE=OFF
-endif
-
-# Detect if we are running on arm64
-ifneq (,$(findstring aarch64,$(shell uname -m)))
-	ONNX_ARCH=aarch64
-endif
-
 ifeq ($(OS),Darwin)
-	ONNX_OS=osx
-	ifneq (,$(findstring aarch64,$(shell uname -m)))
-		ONNX_ARCH=arm64
-	else ifneq (,$(findstring arm64,$(shell uname -m)))
-		ONNX_ARCH=arm64
-	else
-		ONNX_ARCH=x86_64
-	endif

 	ifeq ($(OSX_SIGNING_IDENTITY),)
 		OSX_SIGNING_IDENTITY := $(shell security find-identity -v -p codesigning | grep '"' | head -n 1 | sed -E 's/.*"(.*)"/\1/')
@@ -161,10 +138,10 @@ ifeq ($(BUILD_TYPE),hipblas)
 	export CC=$(ROCM_HOME)/llvm/bin/clang
 	# llama-ggml has no hipblas support, so override it here.
 	export STABLE_BUILD_TYPE=
-	export GGML_HIP=1
+	export GGML_HIPBLAS=1
 	GPU_TARGETS ?= gfx900,gfx906,gfx908,gfx940,gfx941,gfx942,gfx90a,gfx1030,gfx1031,gfx1100,gfx1101
 	AMDGPU_TARGETS ?= "$(GPU_TARGETS)"
-	CMAKE_ARGS+=-DGGML_HIP=ON -DAMDGPU_TARGETS="$(AMDGPU_TARGETS)" -DGPU_TARGETS="$(GPU_TARGETS)"
+	CMAKE_ARGS+=-DGGML_HIPBLAS=ON -DAMDGPU_TARGETS="$(AMDGPU_TARGETS)" -DGPU_TARGETS="$(GPU_TARGETS)"
 	CGO_LDFLAGS += -O3 --rtlib=compiler-rt -unwindlib=libgcc -lhipblas -lrocblas --hip-link -L${ROCM_HOME}/lib/llvm/lib
 endif

@@ -202,23 +179,16 @@ ifeq ($(findstring tts,$(GO_TAGS)),tts)
 endif

 ALL_GRPC_BACKENDS=backend-assets/grpc/huggingface
+ALL_GRPC_BACKENDS+=backend-assets/grpc/bert-embeddings
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-avx
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-avx2
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-fallback
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-ggml
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-grpc
 ALL_GRPC_BACKENDS+=backend-assets/util/llama-cpp-rpc-server
+ALL_GRPC_BACKENDS+=backend-assets/grpc/rwkv
 ALL_GRPC_BACKENDS+=backend-assets/grpc/whisper
-
-ifeq ($(ONNX_OS),linux)
-ifeq ($(ONNX_ARCH),x64)
-	ALL_GRPC_BACKENDS+=backend-assets/grpc/bark-cpp
-	ALL_GRPC_BACKENDS+=backend-assets/grpc/stablediffusion-ggml
-endif
-endif
-
 ALL_GRPC_BACKENDS+=backend-assets/grpc/local-store
-ALL_GRPC_BACKENDS+=backend-assets/grpc/silero-vad
 ALL_GRPC_BACKENDS+=$(OPTIONAL_GRPC)
 # Use filter-out to remove the specified backends
 ALL_GRPC_BACKENDS := $(filter-out $(SKIP_GRPC_BACKEND),$(ALL_GRPC_BACKENDS))
@@ -239,6 +209,19 @@ endif

 all: help

+## BERT embeddings
+sources/go-bert.cpp:
+	mkdir -p sources/go-bert.cpp
+	cd sources/go-bert.cpp && \
+	git init && \
+	git remote add origin $(BERT_REPO) && \
+	git fetch origin && \
+	git checkout $(BERT_VERSION) && \
+	git submodule update --init --recursive --depth 1 --single-branch
+
+sources/go-bert.cpp/libgobert.a: sources/go-bert.cpp
+	$(MAKE) -C sources/go-bert.cpp libgobert.a
+
 ## go-llama.cpp
 sources/go-llama.cpp:
 	mkdir -p sources/go-llama.cpp
@@ -252,23 +235,6 @@ sources/go-llama.cpp:
 sources/go-llama.cpp/libbinding.a: sources/go-llama.cpp
 	$(MAKE) -C sources/go-llama.cpp BUILD_TYPE=$(STABLE_BUILD_TYPE) libbinding.a

-## bark.cpp
-sources/bark.cpp:
-	git clone --recursive $(BARKCPP_REPO) sources/bark.cpp && \
-	cd sources/bark.cpp && \
-	git checkout $(BARKCPP_VERSION) && \
-	git submodule update --init --recursive --depth 1 --single-branch
-
-sources/bark.cpp/build/libbark.a: sources/bark.cpp
-	cd sources/bark.cpp && \
-	mkdir -p build && \
-	cd build && \
-	cmake $(CMAKE_ARGS) .. && \
-	cmake --build . --config Release
-
-backend/go/bark/libbark.a: sources/bark.cpp/build/libbark.a
-	$(MAKE) -C backend/go/bark libbark.a
-
 ## go-piper
 sources/go-piper:
 	mkdir -p sources/go-piper
@@ -282,7 +248,21 @@ sources/go-piper:
 sources/go-piper/libpiper_binding.a: sources/go-piper
 	$(MAKE) -C sources/go-piper libpiper_binding.a example/main piper.o

-## stable diffusion (onnx)
+
+## RWKV
+sources/go-rwkv.cpp:
+	mkdir -p sources/go-rwkv.cpp
+	cd sources/go-rwkv.cpp && \
+	git init && \
+	git remote add origin $(RWKV_REPO) && \
+	git fetch origin && \
+	git checkout $(RWKV_VERSION) && \
+	git submodule update --init --recursive --depth 1 --single-branch
+
+sources/go-rwkv.cpp/librwkv.a: sources/go-rwkv.cpp
+	cd sources/go-rwkv.cpp && cd rwkv.cpp &&	cmake . -DRWKV_BUILD_SHARED_LIBRARY=OFF &&	cmake --build . && 	cp librwkv.a ..
+
+## stable diffusion
 sources/go-stable-diffusion:
 	mkdir -p sources/go-stable-diffusion
 	cd sources/go-stable-diffusion && \
@@ -295,38 +275,6 @@ sources/go-stable-diffusion:
 sources/go-stable-diffusion/libstablediffusion.a: sources/go-stable-diffusion
 	CPATH="$(CPATH):/usr/include/opencv4" $(MAKE) -C sources/go-stable-diffusion libstablediffusion.a

-## stablediffusion (ggml)
-sources/stablediffusion-ggml.cpp:
-	git clone --recursive $(STABLEDIFFUSION_GGML_REPO) sources/stablediffusion-ggml.cpp && \
-	cd sources/stablediffusion-ggml.cpp && \
-	git checkout $(STABLEDIFFUSION_GGML_VERSION) && \
-	git submodule update --init --recursive --depth 1 --single-branch
-
-backend/go/image/stablediffusion-ggml/libsd.a: sources/stablediffusion-ggml.cpp
-	$(MAKE) -C backend/go/image/stablediffusion-ggml build/libstable-diffusion.a
-	$(MAKE) -C backend/go/image/stablediffusion-ggml libsd.a
-
-backend-assets/grpc/stablediffusion-ggml: backend/go/image/stablediffusion-ggml/libsd.a backend-assets/grpc
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/backend/go/image/stablediffusion-ggml/ LIBRARY_PATH=$(CURDIR)/backend/go/image/stablediffusion-ggml/ \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/stablediffusion-ggml ./backend/go/image/stablediffusion-ggml/
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/stablediffusion-ggml
-endif
-
-sources/onnxruntime:
-	mkdir -p sources/onnxruntime
-	curl -L https://github.com/microsoft/onnxruntime/releases/download/v$(ONNX_VERSION)/onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz -o sources/onnxruntime/onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz
-	cd sources/onnxruntime && tar -xvf onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz && rm onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz
-	cd sources/onnxruntime && mv onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION)/* ./
-
-backend-assets/lib/libonnxruntime.so.1: backend-assets/lib sources/onnxruntime
-	cp -rfv sources/onnxruntime/lib/* backend-assets/lib/
-ifeq ($(OS),Darwin)
-	mv backend-assets/lib/libonnxruntime.$(ONNX_VERSION).dylib backend-assets/lib/libonnxruntime.dylib
-else
-	mv backend-assets/lib/libonnxruntime.so.$(ONNX_VERSION) backend-assets/lib/libonnxruntime.so.1
-endif
-
 ## tiny-dream
 sources/go-tiny-dream:
 	mkdir -p sources/go-tiny-dream
@@ -353,19 +301,23 @@ sources/whisper.cpp:
 sources/whisper.cpp/libwhisper.a: sources/whisper.cpp
 	cd sources/whisper.cpp && $(MAKE) libwhisper.a libggml.a

-get-sources: sources/go-llama.cpp sources/go-piper sources/stablediffusion-ggml.cpp sources/bark.cpp sources/whisper.cpp sources/go-stable-diffusion sources/go-tiny-dream backend/cpp/llama/llama.cpp
+get-sources: sources/go-llama.cpp sources/go-piper sources/go-rwkv.cpp sources/whisper.cpp sources/go-bert.cpp sources/go-stable-diffusion sources/go-tiny-dream backend/cpp/llama/llama.cpp

 replace:
+	$(GOCMD) mod edit -replace github.com/donomii/go-rwkv.cpp=$(CURDIR)/sources/go-rwkv.cpp
 	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp=$(CURDIR)/sources/whisper.cpp
 	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp/bindings/go=$(CURDIR)/sources/whisper.cpp/bindings/go
+	$(GOCMD) mod edit -replace github.com/go-skynet/go-bert.cpp=$(CURDIR)/sources/go-bert.cpp
 	$(GOCMD) mod edit -replace github.com/M0Rf30/go-tiny-dream=$(CURDIR)/sources/go-tiny-dream
 	$(GOCMD) mod edit -replace github.com/mudler/go-piper=$(CURDIR)/sources/go-piper
 	$(GOCMD) mod edit -replace github.com/mudler/go-stable-diffusion=$(CURDIR)/sources/go-stable-diffusion
 	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(CURDIR)/sources/go-llama.cpp

 dropreplace:
+	$(GOCMD) mod edit -dropreplace github.com/donomii/go-rwkv.cpp
 	$(GOCMD) mod edit -dropreplace github.com/ggerganov/whisper.cpp
 	$(GOCMD) mod edit -dropreplace github.com/ggerganov/whisper.cpp/bindings/go
+	$(GOCMD) mod edit -dropreplace github.com/go-skynet/go-bert.cpp
 	$(GOCMD) mod edit -dropreplace github.com/M0Rf30/go-tiny-dream
 	$(GOCMD) mod edit -dropreplace github.com/mudler/go-piper
 	$(GOCMD) mod edit -dropreplace github.com/mudler/go-stable-diffusion
@@ -378,8 +330,10 @@ prepare-sources: get-sources replace
 rebuild: ## Rebuilds the project
 	$(GOCMD) clean -cache
 	$(MAKE) -C sources/go-llama.cpp clean
+	$(MAKE) -C sources/go-rwkv.cpp clean
 	$(MAKE) -C sources/whisper.cpp clean
 	$(MAKE) -C sources/go-stable-diffusion clean
+	$(MAKE) -C sources/go-bert.cpp clean
 	$(MAKE) -C sources/go-piper clean
 	$(MAKE) -C sources/go-tiny-dream clean
 	$(MAKE) build
@@ -394,9 +348,7 @@ clean: ## Remove build related file
 	rm -rf release/
 	rm -rf backend-assets/*
 	$(MAKE) -C backend/cpp/grpc clean
-	$(MAKE) -C backend/go/bark clean
 	$(MAKE) -C backend/cpp/llama clean
-	$(MAKE) -C backend/go/image/stablediffusion-ggml clean
 	rm -rf backend/cpp/llama-* || true
 	$(MAKE) dropreplace
 	$(MAKE) protogen-clean
@@ -487,6 +439,8 @@ test-models/testmodel.ggml:
 	wget -q https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin -O test-models/whisper-en
 	wget -q https://huggingface.co/mudler/all-MiniLM-L6-v2/resolve/main/ggml-model-q4_0.bin -O test-models/bert
 	wget -q https://cdn.openai.com/whisper/draft-20220913a/micro-machines.wav -O test-dir/audio.wav
+	wget -q https://huggingface.co/mudler/rwkv-4-raven-1.5B-ggml/resolve/main/RWKV-4-Raven-1B5-v11-Eng99%2525-Other1%2525-20230425-ctx4096_Q4_0.bin -O test-models/rwkv
+	wget -q https://raw.githubusercontent.com/saharNooby/rwkv.cpp/5eb8f09c146ea8124633ab041d9ea0b1f1db4459/rwkv/20B_tokenizer.json -O test-models/rwkv.tokenizer.json
 	cp tests/models_fixtures/* test-models

 prepare-test: grpcs
@@ -739,6 +693,13 @@ backend-assets/espeak-ng-data: sources/go-piper sources/go-piper/libpiper_bindin
 backend-assets/grpc: protogen-go replace
 	mkdir -p backend-assets/grpc

+backend-assets/grpc/bert-embeddings: sources/go-bert.cpp sources/go-bert.cpp/libgobert.a backend-assets/grpc
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-bert.cpp LIBRARY_PATH=$(CURDIR)/sources/go-bert.cpp \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/bert-embeddings ./backend/go/llm/bert/
+ifneq ($(UPX),)
+	$(UPX) backend-assets/grpc/bert-embeddings
+endif
+
 backend-assets/grpc/huggingface: backend-assets/grpc
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/huggingface ./backend/go/llm/langchain/
 ifneq ($(UPX),)
@@ -798,6 +759,10 @@ backend-assets/grpc/llama-cpp-fallback: backend-assets/grpc backend/cpp/llama/ll
 	$(info ${GREEN}I llama-cpp build info:fallback${RESET})
 	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off" $(MAKE) VARIANT="llama-fallback" build-llama-cpp-grpc-server
 	cp -rfv backend/cpp/llama-fallback/grpc-server backend-assets/grpc/llama-cpp-fallback
+# TODO: every binary should have its own folder instead, so can have different metal implementations
+ifeq ($(BUILD_TYPE),metal)
+	cp backend/cpp/llama-fallback/llama.cpp/build/bin/default.metallib backend-assets/grpc/
+endif

 backend-assets/grpc/llama-cpp-cuda: backend-assets/grpc backend/cpp/llama/llama.cpp
 	cp -rf backend/cpp/llama backend/cpp/llama-cuda
@@ -810,7 +775,7 @@ backend-assets/grpc/llama-cpp-hipblas: backend-assets/grpc backend/cpp/llama/lla
 	cp -rf backend/cpp/llama backend/cpp/llama-hipblas
 	$(MAKE) -C backend/cpp/llama-hipblas purge
 	$(info ${GREEN}I llama-cpp build info:hipblas${RESET})
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off" BUILD_TYPE="hipblas" $(MAKE) VARIANT="llama-hipblas" build-llama-cpp-grpc-server
+	BUILD_TYPE="hipblas" $(MAKE) VARIANT="llama-hipblas" build-llama-cpp-grpc-server
 	cp -rfv backend/cpp/llama-hipblas/grpc-server backend-assets/grpc/llama-cpp-hipblas

 backend-assets/grpc/llama-cpp-sycl_f16: backend-assets/grpc backend/cpp/llama/llama.cpp
@@ -845,13 +810,6 @@ ifneq ($(UPX),)
 	$(UPX) backend-assets/grpc/llama-ggml
 endif

-backend-assets/grpc/bark-cpp: backend/go/bark/libbark.a backend-assets/grpc
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/backend/go/bark/ LIBRARY_PATH=$(CURDIR)/backend/go/bark/ \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/bark-cpp ./backend/go/bark/
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/bark-cpp
-endif
-
 backend-assets/grpc/piper: sources/go-piper sources/go-piper/libpiper_binding.a backend-assets/grpc backend-assets/espeak-ng-data
 	CGO_CXXFLAGS="$(PIPER_CGO_CXXFLAGS)" CGO_LDFLAGS="$(PIPER_CGO_LDFLAGS)" LIBRARY_PATH=$(CURDIR)/sources/go-piper \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/piper ./backend/go/tts/
@@ -859,6 +817,13 @@ ifneq ($(UPX),)
 	$(UPX) backend-assets/grpc/piper
 endif

+backend-assets/grpc/rwkv: sources/go-rwkv.cpp sources/go-rwkv.cpp/librwkv.a backend-assets/grpc
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-rwkv.cpp LIBRARY_PATH=$(CURDIR)/sources/go-rwkv.cpp \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/rwkv ./backend/go/llm/rwkv
+ifneq ($(UPX),)
+	$(UPX) backend-assets/grpc/rwkv
+endif
+
 backend-assets/grpc/stablediffusion: sources/go-stable-diffusion sources/go-stable-diffusion/libstablediffusion.a backend-assets/grpc
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" CPATH="$(CPATH):$(CURDIR)/sources/go-stable-diffusion/:/usr/include/opencv4" LIBRARY_PATH=$(CURDIR)/sources/go-stable-diffusion/ \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/stablediffusion ./backend/go/image/stablediffusion
@@ -866,13 +831,6 @@ ifneq ($(UPX),)
 	$(UPX) backend-assets/grpc/stablediffusion
 endif

-backend-assets/grpc/silero-vad: backend-assets/grpc backend-assets/lib/libonnxruntime.so.1
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" CPATH="$(CPATH):$(CURDIR)/sources/onnxruntime/include/" LIBRARY_PATH=$(CURDIR)/backend-assets/lib \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/silero-vad ./backend/go/vad/silero
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/silero-vad
-endif
-
 backend-assets/grpc/tinydream: sources/go-tiny-dream sources/go-tiny-dream/libtinydream.a backend-assets/grpc
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" LIBRARY_PATH=$(CURDIR)/go-tiny-dream \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/tinydream ./backend/go/image/tinydream
@@ -933,7 +891,7 @@ docker-aio-all:

 docker-image-intel:
 	docker build \
-		--build-arg BASE_IMAGE=intel/oneapi-basekit:2025.0.0-0-devel-ubuntu22.04 \
+		--build-arg BASE_IMAGE=intel/oneapi-basekit:2024.2.0-devel-ubuntu22.04 \
 		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
 		--build-arg GO_TAGS="none" \
 		--build-arg MAKEFLAGS="$(DOCKER_MAKEFLAGS)" \
@@ -941,7 +899,7 @@ docker-image-intel:

 docker-image-intel-xpu:
 	docker build \
-		--build-arg BASE_IMAGE=intel/oneapi-basekit:2025.0.0-0-devel-ubuntu22.04 \
+		--build-arg BASE_IMAGE=intel/oneapi-basekit:2024.2.0-devel-ubuntu22.04 \
 		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
 		--build-arg GO_TAGS="none" \
 		--build-arg MAKEFLAGS="$(DOCKER_MAKEFLAGS)" \
--- a/README.md
+++ b/README.md
@@ -38,13 +38,9 @@
 </a>
 </p>

-<p align="center">
-<a href="https://trendshift.io/repositories/1484" target="_blank"><img src="https://trendshift.io/api/badge/repositories/1484" alt="go-skynet%2FLocalAI | Trendshift" style="width: 250px; height: 55px;" width="250" height="55"/></a>
-</p>
-
 > :bulb: Get help - [❓FAQ](https://localai.io/faq/) [💭Discussions](https://github.com/go-skynet/LocalAI/discussions) [:speech_balloon: Discord](https://discord.gg/uJAeKSAGDy) [:book: Documentation website](https://localai.io/)
 >
-> [💻 Quickstart](https://localai.io/basics/getting_started/) [🖼️ Models](https://models.localai.io/) [🚀 Roadmap](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap) [🥽 Demo](https://demo.localai.io) [🌍 Explorer](https://explorer.localai.io) [🛫 Examples](https://github.com/mudler/LocalAI-examples) 
+> [💻 Quickstart](https://localai.io/basics/getting_started/) [🖼️ Models](https://models.localai.io/) [🚀 Roadmap](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap) [🥽 Demo](https://demo.localai.io) [🌍 Explorer](https://explorer.localai.io) [🛫 Examples](https://github.com/go-skynet/LocalAI/tree/master/examples/) 

 [![tests](https://github.com/go-skynet/LocalAI/actions/workflows/test.yml/badge.svg)](https://github.com/go-skynet/LocalAI/actions/workflows/test.yml)[![Build and Release](https://github.com/go-skynet/LocalAI/actions/workflows/release.yaml/badge.svg)](https://github.com/go-skynet/LocalAI/actions/workflows/release.yaml)[![build container images](https://github.com/go-skynet/LocalAI/actions/workflows/image.yml/badge.svg)](https://github.com/go-skynet/LocalAI/actions/workflows/image.yml)[![Bump dependencies](https://github.com/go-skynet/LocalAI/actions/workflows/bump_deps.yaml/badge.svg)](https://github.com/go-skynet/LocalAI/actions/workflows/bump_deps.yaml)[![Artifact Hub](https://img.shields.io/endpoint?url=https://artifacthub.io/badge/repository/localai)](https://artifacthub.io/packages/search?repo=localai)

@@ -60,17 +56,14 @@ curl https://localai.io/install.sh | sh

 Or run with docker:
 ```bash
-# CPU only image:
-docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-cpu
-
-# Nvidia GPU:
-docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-gpu-nvidia-cuda-12
-
-# CPU and GPU image (bigger size):
-docker run -ti --name local-ai -p 8080:8080 localai/localai:latest
-
-# AIO images (it will pre-download a set of models ready for use, see https://localai.io/basics/container/)
 docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-aio-cpu
+# Alternative images:
+# - if you have an Nvidia GPU:
+# docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-aio-gpu-nvidia-cuda-12
+# - without preconfigured models
+# docker run -ti --name local-ai -p 8080:8080 localai/localai:latest
+# - without preconfigured models for Nvidia GPUs
+# docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-gpu-nvidia-cuda-12 
 ```

 To load models:
@@ -92,10 +85,6 @@ local-ai run oci://localai/phi-2:latest

 ## 📰 Latest project news

- Dec 2024: stablediffusion.cpp backend (ggml) added ( https://github.com/mudler/LocalAI/pull/4289 )
- Nov 2024: Bark.cpp backend added ( https://github.com/mudler/LocalAI/pull/4287 )
- Nov 2024: Voice activity detection models (**VAD**) added to the API: https://github.com/mudler/LocalAI/pull/4204
- Oct 2024: examples moved to [LocalAI-examples](https://github.com/mudler/LocalAI-examples)
 - Aug 2024:  🆕 FLUX-1, [P2P Explorer](https://explorer.localai.io)
 - July 2024: 🔥🔥 🆕 P2P Dashboard, LocalAI Federated mode and AI Swarms: https://github.com/mudler/LocalAI/pull/2723
 - June 2024: 🆕 You can browse now the model gallery without LocalAI! Check out https://models.localai.io
@@ -126,10 +115,10 @@ If you want to help and contribute, issues up for grabs: https://github.com/mudl

 ## 🚀 [Features](https://localai.io/features/)

- 📖 [Text generation with GPTs](https://localai.io/features/text-generation/) (`llama.cpp`, `transformers`, `vllm` ... [:book: and more](https://localai.io/model-compatibility/index.html#model-compatibility-table))
+- 📖 [Text generation with GPTs](https://localai.io/features/text-generation/) (`llama.cpp`, `gpt4all.cpp`, ... [:book: and more](https://localai.io/model-compatibility/index.html#model-compatibility-table))
 - 🗣 [Text to Audio](https://localai.io/features/text-to-audio/)
 - 🔈 [Audio to Text](https://localai.io/features/audio-to-text/) (Audio transcription with `whisper.cpp`)
- 🎨 [Image generation](https://localai.io/features/image-generation)
+- 🎨 [Image generation with stable diffusion](https://localai.io/features/image-generation)
 - 🔥 [OpenAI-alike tools API](https://localai.io/features/openai-functions/) 
 - 🧠 [Embeddings generation for vector databases](https://localai.io/features/embeddings/)
 - ✍️ [Constrained grammars](https://localai.io/features/constrained_grammars/)
@@ -137,7 +126,6 @@ If you want to help and contribute, issues up for grabs: https://github.com/mudl
 - 🥽 [Vision API](https://localai.io/features/gpt-vision/)
 - 📈 [Reranker API](https://localai.io/features/reranker/)
 - 🆕🖧 [P2P Inferencing](https://localai.io/features/distribute/)
- 🔊 Voice activity detection (Silero-VAD support)
 - 🌍 Integrated WebUI!

 ## 💻 Usage
@@ -160,7 +148,6 @@ Model galleries
 Other:
 - Helm chart https://github.com/go-skynet/helm-charts
 - VSCode extension https://github.com/badgooooor/localai-vscode-plugin
- Langchain: https://python.langchain.com/docs/integrations/providers/localai/
 - Terminal utility https://github.com/djcopley/ShellOracle
 - Local Smart assistant https://github.com/mudler/LocalAGI
 - Home Assistant https://github.com/sammcj/homeassistant-localai / https://github.com/drndos/hass-openai-custom-conversation / https://github.com/valentinfrlch/ha-gpt4vision
@@ -168,9 +155,6 @@ Other:
 - Slack bot https://github.com/mudler/LocalAGI/tree/main/examples/slack
 - Shell-Pilot(Interact with LLM using LocalAI models via pure shell scripts on your Linux or MacOS system) https://github.com/reid41/shell-pilot
 - Telegram bot https://github.com/mudler/LocalAI/tree/master/examples/telegram-bot
- Another Telegram Bot https://github.com/JackBekket/Hellper
- Auto-documentation https://github.com/JackBekket/Reflexia
- Github bot which answer on issues, with code and documentation as context https://github.com/JackBekket/GitHelper
 - Github Actions: https://github.com/marketplace/actions/start-localai
 - Examples: https://github.com/mudler/LocalAI/tree/master/examples/
  
@@ -245,6 +229,7 @@ LocalAI couldn't have been built without the help of great software already avai
 - https://github.com/antimatter15/alpaca.cpp
 - https://github.com/EdVince/Stable-Diffusion-NCNN
 - https://github.com/ggerganov/whisper.cpp
+- https://github.com/saharNooby/rwkv.cpp
 - https://github.com/rhasspy/piper

 ## 🤗 Contributors
--- a/aio/cpu/embeddings.yaml
+++ b/aio/cpu/embeddings.yaml
@@ -1,7 +1,7 @@
 name: text-embedding-ada-002
-embeddings: true
+backend: bert-embeddings
 parameters:
-  model: huggingface://hugging-quants/Llama-3.2-1B-Instruct-Q4_K_M-GGUF/llama-3.2-1b-instruct-q4_k_m.gguf
+  model: huggingface://mudler/all-MiniLM-L6-v2/ggml-model-q4_0.bin

 usage: |
    You can test this model with curl like this:
--- a/backend/backend.proto
+++ b/backend/backend.proto
@@ -28,8 +28,6 @@ service Backend {
  rpc Rerank(RerankRequest) returns (RerankResult) {}

  rpc GetMetrics(MetricsRequest) returns (MetricsResponse);
-
-  rpc VAD(VADRequest) returns (VADResponse) {}
 }

 // Define the empty request
@@ -159,7 +157,6 @@ message Reply {
  bytes message = 1;
  int32 tokens = 2;
  int32 prompt_tokens = 3;
-  bytes audio = 5;
 }

 message ModelOptions {
@@ -236,16 +233,6 @@ message ModelOptions {

  bool FlashAttention = 56;
  bool NoKVOffload = 57;
-
-  string ModelPath = 59;
-
-  repeated string LoraAdapters = 60;
-  repeated float LoraScales = 61;
-
-  repeated string Options = 62;
-
-  string CacheTypeKey = 63;
-  string CacheTypeValue = 64;
 }

 message Result {
@@ -301,19 +288,6 @@ message TTSRequest {
  optional string language = 5;
 }

-message VADRequest {
-  repeated float audio = 1;
-}
-
-message VADSegment {
-  float start = 1;
-  float end = 2;
-}
-
-message VADResponse {
-  repeated VADSegment segments = 1;
-}
-
 message SoundGenerationRequest {
  string text = 1;
  string model = 2;
--- a/backend/cpp/llama/Makefile
+++ b/backend/cpp/llama/Makefile
@@ -22,7 +22,7 @@ else ifeq ($(BUILD_TYPE),clblas)
 	CMAKE_ARGS+=-DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
 # If it's hipblas we do have also to set CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++ 
 else ifeq ($(BUILD_TYPE),hipblas)
-	CMAKE_ARGS+=-DGGML_HIP=ON
+	CMAKE_ARGS+=-DGGML_HIPBLAS=ON
 # If it's OSX, DO NOT embed the metal library - -DGGML_METAL_EMBED_LIBRARY=ON requires further investigation
 # But if it's OSX without metal, disable it here
 else ifeq ($(OS),Darwin)
@@ -30,7 +30,9 @@ else ifeq ($(OS),Darwin)
 		CMAKE_ARGS+=-DGGML_METAL=OFF
 	else
 		CMAKE_ARGS+=-DGGML_METAL=ON
-		CMAKE_ARGS+=-DGGML_METAL_EMBED_LIBRARY=ON
+# Until this is tested properly, we disable embedded metal file
+# as we already embed it as part of the LocalAI assets
+		CMAKE_ARGS+=-DGGML_METAL_EMBED_LIBRARY=OFF
 		TARGET+=--target ggml-metal
 	endif
 endif
--- a/backend/cpp/llama/grpc-server.cpp
+++ b/backend/cpp/llama/grpc-server.cpp
@@ -203,7 +203,7 @@ struct llama_client_slot
    std::string stopping_word;

    // sampling
-    struct common_params_sampling sparams;
+    struct common_sampler_params sparams;
    common_sampler *ctx_sampling = nullptr;

    int32_t ga_i = 0;   // group-attention state
@@ -428,7 +428,6 @@ struct llama_server_context
 {
    llama_model *model = nullptr;
    llama_context *ctx = nullptr;
-    const llama_vocab * vocab = nullptr;

    clip_ctx *clp_ctx = nullptr;

@@ -440,7 +439,6 @@ struct llama_server_context
    bool clean_kv_cache     = true;
    bool all_slots_are_idle = false;
    bool add_bos_token      = true;
-    bool has_eos_token      = true;

    int32_t n_ctx;  // total context for all clients / slots

@@ -494,8 +492,8 @@ struct llama_server_context
        }

        common_init_result common_init = common_init_from_params(params);
-        model = common_init.model.release();
-        ctx = common_init.context.release();
+        model = common_init.model;
+        ctx = common_init.context;
        if (model == nullptr)
        {
            LOG_ERR("unable to load model: %s", params.model.c_str());
@@ -504,7 +502,7 @@ struct llama_server_context

        if (multimodal) {
            const int n_embd_clip = clip_n_mmproj_embd(clp_ctx);
-            const int n_embd_llm  = llama_model_n_embd(model);
+            const int n_embd_llm  = llama_n_embd(model);
            if (n_embd_clip != n_embd_llm) {
                LOG("%s: embedding dim of the multimodal projector (%d) is not equal to that of LLaMA (%d). Make sure that you use the correct mmproj file.\n", __func__, n_embd_clip, n_embd_llm);
                llama_free(ctx);
@@ -513,15 +511,23 @@ struct llama_server_context
            }
        }

-        vocab = llama_model_get_vocab(model);
        n_ctx = llama_n_ctx(ctx);

-        add_bos_token = llama_vocab_get_add_bos(vocab);
-        has_eos_token = llama_vocab_eos(vocab) != LLAMA_TOKEN_NULL;
+        add_bos_token = llama_add_bos_token(model);

        return true;
    }

+    void validate_model_chat_template(server_params & sparams) {
+        llama_chat_message chat[] = {{"user", "test"}};
+        std::vector<char> buf(1);
+        int res = llama_chat_apply_template(model, nullptr, chat, 1, true, buf.data(), buf.size());
+        if (res < 0) {
+            LOG_ERR("The chat template comes with this model is not yet supported, falling back to chatml. This may cause the model to output suboptimal responses", __func__);
+            sparams.chat_template = "<|im_start|>"; // llama_chat_apply_template only checks if <|im_start|> exist in the template
+        }
+    }
+
    llama_client_slot* get_active_slot() {
        for (llama_client_slot& slot : slots) {
            // Check if the slot is currently processing
@@ -656,7 +662,7 @@ struct llama_server_context

    bool launch_slot_with_data(llama_client_slot* &slot, json data) {
        slot_params default_params;
-        common_params_sampling default_sparams;
+        common_sampler_params default_sparams;
 
        slot->params.stream             = json_value(data, "stream",            false);
        slot->params.cache_prompt       = json_value(data, "cache_prompt",      false);
@@ -664,6 +670,7 @@ struct llama_server_context
        slot->sparams.top_k             = json_value(data, "top_k",             default_sparams.top_k);
        slot->sparams.top_p             = json_value(data, "top_p",             default_sparams.top_p);
        slot->sparams.min_p             = json_value(data, "min_p",             default_sparams.min_p);
+        slot->sparams.tfs_z             = json_value(data, "tfs_z",             default_sparams.tfs_z);
        slot->sparams.typ_p             = json_value(data, "typical_p",         default_sparams.typ_p);
        slot->sparams.temp              = json_value(data, "temperature",       default_sparams.temp);
        slot->sparams.dynatemp_range    = json_value(data, "dynatemp_range",    default_sparams.dynatemp_range);
@@ -675,6 +682,7 @@ struct llama_server_context
        slot->sparams.mirostat          = json_value(data, "mirostat",          default_sparams.mirostat);
        slot->sparams.mirostat_tau      = json_value(data, "mirostat_tau",      default_sparams.mirostat_tau);
        slot->sparams.mirostat_eta      = json_value(data, "mirostat_eta",      default_sparams.mirostat_eta);
+        slot->sparams.penalize_nl       = json_value(data, "penalize_nl",       default_sparams.penalize_nl);
        slot->params.n_keep             = json_value(data, "n_keep",            slot->params.n_keep);
        slot->sparams.seed               = json_value(data, "seed",              default_sparams.seed);
        slot->sparams.grammar           = json_value(data, "grammar",           default_sparams.grammar);
@@ -719,8 +727,8 @@ struct llama_server_context
            slot->prompt = "";
        }

-        if (json_value(data, "ignore_eos", false) && has_eos_token) {
-                slot->sparams.logit_bias.push_back({llama_vocab_eos(vocab), -INFINITY});
+        if (json_value(data, "ignore_eos", false)) {
+                slot->sparams.logit_bias.push_back({llama_token_eos(model), -INFINITY});
        }
        /*
        slot->sparams.penalty_prompt_tokens.clear();
@@ -759,13 +767,13 @@ struct llama_server_context
            }
        }
      */
+
        slot->sparams.logit_bias.clear();

        const auto &logit_bias = data.find("logit_bias");
        if (logit_bias != data.end() && logit_bias->is_array())
        {
-            const llama_vocab * vocab = llama_model_get_vocab(model);
-            const int n_vocab = llama_vocab_n_tokens(vocab);
+            const int n_vocab = llama_n_vocab(model);
            for (const auto &el : *logit_bias)
            {
                if (el.is_array() && el.size() == 2)
@@ -794,7 +802,7 @@ struct llama_server_context
                    }
                    else if (el[0].is_string())
                    {
-                        auto toks = common_tokenize(vocab, el[0].get<std::string>(), false);
+                        auto toks = common_tokenize(model, el[0].get<std::string>(), false);
                        for (auto tok : toks)
                        {
                            slot->sparams.logit_bias.push_back({tok, bias});
@@ -1124,7 +1132,7 @@ struct llama_server_context
            slot.has_next_token = false;
        }

-        if (result.tok == llama_vocab_eos(vocab) || llama_vocab_is_eog(vocab, result.tok))
+        if (result.tok == llama_token_eos(model))
        {
            slot.stopped_eos = true;
            slot.has_next_token = false;
@@ -1198,6 +1206,7 @@ struct llama_server_context
            {"top_k",             slot.sparams.top_k},
            {"top_p",             slot.sparams.top_p},
            {"min_p",             slot.sparams.min_p},
+            {"tfs_z",             slot.sparams.tfs_z},
            {"typical_p",         slot.sparams.typ_p},
            {"repeat_last_n",     slot.sparams.penalty_last_n},
            {"repeat_penalty",    slot.sparams.penalty_repeat},
@@ -1206,12 +1215,13 @@ struct llama_server_context
            {"mirostat",          slot.sparams.mirostat},
            {"mirostat_tau",      slot.sparams.mirostat_tau},
            {"mirostat_eta",      slot.sparams.mirostat_eta},
+            {"penalize_nl",       slot.sparams.penalize_nl},
            {"stop",              slot.params.antiprompt},
            {"n_predict",         slot.params.n_predict},
            {"n_keep",            params.n_keep},
            {"ignore_eos",        slot.sparams.ignore_eos},
            {"stream",            slot.params.stream},
-             //      {"logit_bias",        slot.sparams.logit_bias},
+      //      {"logit_bias",        slot.sparams.logit_bias},
            {"n_probs",           slot.sparams.n_probs},
            {"min_keep",          slot.sparams.min_keep},
            {"grammar",           slot.sparams.grammar},
@@ -1319,7 +1329,7 @@ struct llama_server_context
        res.error = false;
        res.stop = true;

-        const int n_embd = llama_model_n_embd(model);
+        const int n_embd = llama_n_embd(model);
        if (!params.embedding)
        {
            LOG_WARNING("embedding disabled", {
@@ -1418,7 +1428,7 @@ struct llama_server_context
                    n_eval = n_batch;
                }

-                const int n_embd = llama_model_n_embd(model);
+                const int n_embd = llama_n_embd(model);
                float * embd = img.image_embedding + i * n_embd;
                llava_embd_batch llava_batch = llava_embd_batch(embd, n_eval, slot.n_past, 0);
                if (llama_decode(ctx, llava_batch.batch))
@@ -1699,11 +1709,11 @@ struct llama_server_context
                            suffix_tokens.erase(suffix_tokens.begin());
                        }

-                        prefix_tokens.insert(prefix_tokens.begin(), llama_vocab_fim_pre(vocab));
-                        prefix_tokens.insert(prefix_tokens.begin(), llama_vocab_bos(vocab)); // always add BOS
-                        prefix_tokens.insert(prefix_tokens.end(),   llama_vocab_fim_suf(vocab));
+                        prefix_tokens.insert(prefix_tokens.begin(), llama_token_prefix(model));
+                        prefix_tokens.insert(prefix_tokens.begin(), llama_token_bos(model)); // always add BOS
+                        prefix_tokens.insert(prefix_tokens.end(),   llama_token_suffix(model));
                        prefix_tokens.insert(prefix_tokens.end(),   suffix_tokens.begin(), suffix_tokens.end());
-                        prefix_tokens.push_back(llama_vocab_fim_mid(vocab));
+                        prefix_tokens.push_back(llama_token_middle(model));
                        prompt_tokens = prefix_tokens;
                    }
                    else
@@ -2095,6 +2105,7 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
    //     slot->params.n_predict        = json_value(data, "n_predict",         default_params.n_predict);
    //     slot->sparams.top_k           = json_value(data, "top_k",             default_sparams.top_k);
    //     slot->sparams.top_p           = json_value(data, "top_p",             default_sparams.top_p);
+    //     slot->sparams.tfs_z           = json_value(data, "tfs_z",             default_sparams.tfs_z);
    //     slot->sparams.typical_p       = json_value(data, "typical_p",         default_sparams.typical_p);
    //     slot->sparams.temp            = json_value(data, "temperature",       default_sparams.temp);
    //     slot->sparams.penalty_last_n  = json_value(data, "repeat_last_n",     default_sparams.penalty_last_n);
@@ -2104,6 +2115,7 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
    //     slot->sparams.mirostat        = json_value(data, "mirostat",          default_sparams.mirostat);
    //     slot->sparams.mirostat_tau    = json_value(data, "mirostat_tau",      default_sparams.mirostat_tau);
    //     slot->sparams.mirostat_eta    = json_value(data, "mirostat_eta",      default_sparams.mirostat_eta);
+    //     slot->sparams.penalize_nl     = json_value(data, "penalize_nl",       default_sparams.penalize_nl);
    //     slot->params.n_keep           = json_value(data, "n_keep",            slot->params.n_keep);
    //     slot->params.seed             = json_value(data, "seed",              default_params.seed);
    //     slot->sparams.grammar         = json_value(data, "grammar",           default_sparams.grammar);
@@ -2117,6 +2129,7 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
    data["n_predict"] = predict->tokens() == 0 ? -1 : predict->tokens();
    data["top_k"] = predict->topk();
    data["top_p"] = predict->topp();
+    data["tfs_z"] = predict->tailfreesamplingz();
    data["typical_p"] = predict->typicalp();
    data["temperature"] = predict->temperature();
    data["repeat_last_n"] = predict->repeat();
@@ -2126,6 +2139,7 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
    data["mirostat"] = predict->mirostat();
    data["mirostat_tau"] = predict->mirostattau();
    data["mirostat_eta"] = predict->mirostateta();
+    data["penalize_nl"] = predict->penalizenl();
    data["n_keep"] = predict->nkeep();
    data["seed"] = predict->seed();
    data["grammar"] = predict->grammar();
@@ -2162,6 +2176,7 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
 //     llama.params.n_predict = predict->tokens() == 0 ? -1 : predict->tokens();
 //     llama.params.sparams.top_k = predict->topk();
 //     llama.params.sparams.top_p = predict->topp();
+//     llama.params.sparams.tfs_z = predict->tailfreesamplingz();
 //     llama.params.sparams.typical_p = predict->typicalp();
 //     llama.params.sparams.penalty_last_n = predict->repeat();
 //     llama.params.sparams.temp = predict->temperature();
@@ -2171,6 +2186,7 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
 //     llama.params.sparams.mirostat = predict->mirostat();
 //     llama.params.sparams.mirostat_tau = predict->mirostattau();
 //     llama.params.sparams.mirostat_eta = predict->mirostateta();
+//     llama.params.sparams.penalize_nl = predict->penalizenl();
 //     llama.params.n_keep = predict->nkeep();
 //     llama.params.seed = predict->seed();
 //     llama.params.sparams.grammar = predict->grammar();
@@ -2217,35 +2233,6 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
 //     }
 // }

-const std::vector<ggml_type> kv_cache_types = {
-    GGML_TYPE_F32,
-    GGML_TYPE_F16,
-    GGML_TYPE_BF16,
-    GGML_TYPE_Q8_0,
-    GGML_TYPE_Q4_0,
-    GGML_TYPE_Q4_1,
-    GGML_TYPE_IQ4_NL,
-    GGML_TYPE_Q5_0,
-    GGML_TYPE_Q5_1,
-};
-
-static ggml_type kv_cache_type_from_str(const std::string & s) {
-    for (const auto & type : kv_cache_types) {
-        if (ggml_type_name(type) == s) {
-            return type;
-        }
-    }
-    throw std::runtime_error("Unsupported cache type: " + s);
-}
-
-static std::string get_all_kv_cache_types() {
-    std::ostringstream msg;
-    for (const auto & type : kv_cache_types) {
-        msg << ggml_type_name(type) << (&type == &kv_cache_types.back() ? "" : ", ");
-    }
-    return msg.str();
-}
-
 static void params_parse(const backend::ModelOptions* request,
                                common_params & params) {
   
@@ -2259,12 +2246,6 @@ static void params_parse(const backend::ModelOptions* request,
    }
    //  params.model_alias ??
    params.model_alias =  request->modelfile();
-    if (!request->cachetypekey().empty()) {
-        params.cache_type_k = kv_cache_type_from_str(request->cachetypekey());
-    }
-    if (!request->cachetypevalue().empty()) {
-        params.cache_type_v = kv_cache_type_from_str(request->cachetypevalue());
-    }
    params.n_ctx = request->contextsize();
    //params.memory_f16 = request->f16memory();
    params.cpuparams.n_threads = request->threads();
@@ -2323,7 +2304,6 @@ static void params_parse(const backend::ModelOptions* request,
    params.use_mmap = request->mmap();
    params.flash_attn = request->flashattention();
    params.no_kv_offload = request->nokvoffload();
-    params.ctx_shift = false; // We control context-shifting in any case (and we disable it as it could just lead to infinite loops)

    params.embedding = request->embeddings();

--- a/backend/cpp/llama/patches/01-llava.patch
+++ b/backend/cpp/llama/patches/01-llava.patch
@@ -1,13 +1,13 @@
 diff --git a/examples/llava/clip.cpp b/examples/llava/clip.cpp
-index 3cd0d2fa..6c5e811a 100644
+index 342042ff..224db9b5 100644
 --- a/examples/llava/clip.cpp
 +++ b/examples/llava/clip.cpp
-@@ -2608,7 +2608,7 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
-                 struct ggml_tensor * patches = ggml_graph_get_tensor(gf, "patches");
-                 int* patches_data = (int*)malloc(ggml_nbytes(patches));
-                 for (int i = 0; i < num_patches; i++) {
-                    patches_data[i] = i + 1;
-+                    patches_data[i] = i;
-                 }
-                 ggml_backend_tensor_set(patches, patches_data, 0, ggml_nbytes(patches));
-                 free(patches_data);
+@@ -2419,7 +2419,7 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
+             struct ggml_tensor * patches = ggml_graph_get_tensor(gf, "patches");
+             int* patches_data = (int*)malloc(ggml_nbytes(patches));
+             for (int i = 0; i < num_patches; i++) {
+-                patches_data[i] = i + 1;
+                patches_data[i] = i;
+             }
+             ggml_backend_tensor_set(patches, patches_data, 0, ggml_nbytes(patches));
+             free(patches_data);
--- a/backend/go/bark/Makefile
+++ b/backend/go/bark/Makefile
@@ -1,25 +0,0 @@
-INCLUDE_PATH := $(abspath ./)
-LIBRARY_PATH := $(abspath ./)
-
-AR?=ar
-
-BUILD_TYPE?=
-# keep standard at C11 and C++11
-CXXFLAGS = -I. -I$(INCLUDE_PATH)/../../../sources/bark.cpp/examples -I$(INCLUDE_PATH)/../../../sources/bark.cpp/spm-headers -I$(INCLUDE_PATH)/../../../sources/bark.cpp -O3 -DNDEBUG -std=c++17 -fPIC
-LDFLAGS  = -L$(LIBRARY_PATH) -L$(LIBRARY_PATH)/../../../sources/bark.cpp/build/examples -lbark -lstdc++ -lm
-
-# warnings
-CXXFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function
-
-gobark.o:
-	$(CXX) $(CXXFLAGS) gobark.cpp -o gobark.o -c $(LDFLAGS)
-
-libbark.a: gobark.o
-	cp $(INCLUDE_PATH)/../../../sources/bark.cpp/build/libbark.a ./
-	$(AR) rcs libbark.a gobark.o
-	$(AR) rcs libbark.a $(LIBRARY_PATH)/../../../sources/bark.cpp/build/encodec.cpp/ggml/src/CMakeFiles/ggml.dir/ggml.c.o
-	$(AR) rcs libbark.a $(LIBRARY_PATH)/../../../sources/bark.cpp/build/encodec.cpp/ggml/src/CMakeFiles/ggml.dir/ggml-alloc.c.o
-	$(AR) rcs libbark.a $(LIBRARY_PATH)/../../../sources/bark.cpp/build/encodec.cpp/ggml/src/CMakeFiles/ggml.dir/ggml-backend.c.o
-
-clean:
-	rm -f gobark.o libbark.a
--- a/backend/go/bark/gobark.cpp
+++ b/backend/go/bark/gobark.cpp
@@ -1,85 +0,0 @@
-#include <iostream>
-#include <tuple>
-
-#include "bark.h"
-#include "gobark.h"
-#include "common.h"
-#include "ggml.h"
-
-struct bark_context *c;
-
-void bark_print_progress_callback(struct bark_context *bctx, enum bark_encoding_step step, int progress, void *user_data) {
-    if (step == bark_encoding_step::SEMANTIC) {
-        printf("\rGenerating semantic tokens... %d%%", progress);
-    } else if (step == bark_encoding_step::COARSE) {
-        printf("\rGenerating coarse tokens... %d%%", progress);
-    } else if (step == bark_encoding_step::FINE) {
-        printf("\rGenerating fine tokens... %d%%", progress);
-    }
-    fflush(stdout);
-}
-
-int load_model(char *model) {
-    // initialize bark context
-    struct bark_context_params ctx_params = bark_context_default_params();
-    bark_params params;
-
-    params.model_path = model;
-
-   // ctx_params.verbosity = verbosity;
-    ctx_params.progress_callback = bark_print_progress_callback;
-    ctx_params.progress_callback_user_data = nullptr;
-
-    struct bark_context *bctx = bark_load_model(params.model_path.c_str(), ctx_params, params.seed);
-    if (!bctx) {
-        fprintf(stderr, "%s: Could not load model\n", __func__);
-        return 1;
-    }
-
-    c = bctx;
-
-    return 0;
-}
-
-int tts(char *text,int  threads, char *dst ) {
-
-    ggml_time_init();
-    const int64_t t_main_start_us = ggml_time_us();
-
-    // generate audio
-    if (!bark_generate_audio(c, text, threads)) {
-        fprintf(stderr, "%s: An error occured. If the problem persists, feel free to open an issue to report it.\n", __func__);
-        return 1;
-    }
-
-    const float *audio_data = bark_get_audio_data(c);
-    if (audio_data == NULL) {
-        fprintf(stderr, "%s: Could not get audio data\n", __func__);
-        return 1;
-    }
-
-    const int audio_arr_size = bark_get_audio_data_size(c);
-
-    std::vector<float> audio_arr(audio_data, audio_data + audio_arr_size);
-
-    write_wav_on_disk(audio_arr, dst);
-
-    // report timing
-    {
-        const int64_t t_main_end_us = ggml_time_us();
-        const int64_t t_load_us = bark_get_load_time(c);
-        const int64_t t_eval_us = bark_get_eval_time(c);
-
-        printf("\n\n");
-        printf("%s:     load time = %8.2f ms\n", __func__, t_load_us / 1000.0f);
-        printf("%s:     eval time = %8.2f ms\n", __func__, t_eval_us / 1000.0f);
-        printf("%s:    total time = %8.2f ms\n", __func__, (t_main_end_us - t_main_start_us) / 1000.0f);
-    }
-    
-    return 0;
-}
-
-int unload() {
-    bark_free(c);
-}
-
--- a/backend/go/bark/gobark.go
+++ b/backend/go/bark/gobark.go
@@ -1,52 +0,0 @@
-package main
-
-// #cgo CXXFLAGS: -I${SRCDIR}/../../../sources/bark.cpp/ -I${SRCDIR}/../../../sources/bark.cpp/encodec.cpp -I${SRCDIR}/../../../sources/bark.cpp/examples -I${SRCDIR}/../../../sources/bark.cpp/spm-headers
-// #cgo LDFLAGS: -L${SRCDIR}/ -L${SRCDIR}/../../../sources/bark.cpp/build/examples -L${SRCDIR}/../../../sources/bark.cpp/build/encodec.cpp/ -lbark -lencodec -lcommon
-// #include <gobark.h>
-// #include <stdlib.h>
-import "C"
-
-import (
-	"fmt"
-	"unsafe"
-
-	"github.com/mudler/LocalAI/pkg/grpc/base"
-	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
-)
-
-type Bark struct {
-	base.SingleThread
-	threads int
-}
-
-func (sd *Bark) Load(opts *pb.ModelOptions) error {
-
-	sd.threads = int(opts.Threads)
-
-	modelFile := C.CString(opts.ModelFile)
-	defer C.free(unsafe.Pointer(modelFile))
-
-	ret := C.load_model(modelFile)
-	if ret != 0 {
-		return fmt.Errorf("inference failed")
-	}
-
-	return nil
-}
-
-func (sd *Bark) TTS(opts *pb.TTSRequest) error {
-	t := C.CString(opts.Text)
-	defer C.free(unsafe.Pointer(t))
-
-	dst := C.CString(opts.Dst)
-	defer C.free(unsafe.Pointer(dst))
-
-	threads := C.int(sd.threads)
-
-	ret := C.tts(t, threads, dst)
-	if ret != 0 {
-		return fmt.Errorf("inference failed")
-	}
-
-	return nil
-}
--- a/backend/go/bark/gobark.h
+++ b/backend/go/bark/gobark.h
@@ -1,8 +0,0 @@
-#ifdef __cplusplus
-extern "C" {
-#endif
-int load_model(char *model);
-int tts(char *text,int  threads, char *dst );
-#ifdef __cplusplus
-}
-#endif
--- a/backend/go/image/stablediffusion-ggml/Makefile
+++ b/backend/go/image/stablediffusion-ggml/Makefile
@@ -1,96 +0,0 @@
-INCLUDE_PATH := $(abspath ./)
-LIBRARY_PATH := $(abspath ./)
-
-AR?=ar
-CMAKE_ARGS?=
-BUILD_TYPE?=
-ONEAPI_VARS?=/opt/intel/oneapi/setvars.sh
-# keep standard at C11 and C++11
-CXXFLAGS = -I. -I$(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp/thirdparty -I$(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp/ggml/include -I$(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp -O3 -DNDEBUG -std=c++17 -fPIC
-
-# Disable Shared libs as we are linking on static gRPC and we can't mix shared and static
-CMAKE_ARGS+=-DBUILD_SHARED_LIBS=OFF
-
-# If build type is cublas, then we set -DGGML_CUDA=ON to CMAKE_ARGS automatically
-ifeq ($(BUILD_TYPE),cublas)
-	CMAKE_ARGS+=-DGGML_CUDA=ON
-# If build type is openblas then we set -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS
-# to CMAKE_ARGS automatically
-else ifeq ($(BUILD_TYPE),openblas)
-	CMAKE_ARGS+=-DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS
-# If build type is clblas (openCL) we set -DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
-else ifeq ($(BUILD_TYPE),clblas)
-	CMAKE_ARGS+=-DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
-# If it's hipblas we do have also to set CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++ 
-else ifeq ($(BUILD_TYPE),hipblas)
-	CMAKE_ARGS+=-DGGML_HIP=ON
-# If it's OSX, DO NOT embed the metal library - -DGGML_METAL_EMBED_LIBRARY=ON requires further investigation
-# But if it's OSX without metal, disable it here
-else ifeq ($(OS),Darwin)
-	ifneq ($(BUILD_TYPE),metal)
-		CMAKE_ARGS+=-DGGML_METAL=OFF
-	else
-		CMAKE_ARGS+=-DGGML_METAL=ON
-		CMAKE_ARGS+=-DGGML_METAL_EMBED_LIBRARY=ON
-		TARGET+=--target ggml-metal
-	endif
-endif
-
-# ifeq ($(BUILD_TYPE),sycl_f16)
-# 	CMAKE_ARGS+=-DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_SYCL_F16=ON -DSD_SYCL=ON -DGGML_SYCL_F16=ON
-# endif
-
-# ifeq ($(BUILD_TYPE),sycl_f32)
-# 	CMAKE_ARGS+=-DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DSD_SYCL=ON
-# endif
-
-# warnings
-CXXFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function
-
-# Find all .a archives in ARCHIVE_DIR
-# (ggml can have different backends cpu, cuda, etc., each backend generates a .a archive)
-GGML_ARCHIVE_DIR := build/ggml/src/
-ALL_ARCHIVES := $(shell find $(GGML_ARCHIVE_DIR) -type f -name '*.a')
-
-# Name of the single merged library
-COMBINED_LIB := libggmlall.a
-
-# Rule to merge all the .a files into one
-$(COMBINED_LIB): $(ALL_ARCHIVES)
-	@echo "Merging all .a into $(COMBINED_LIB)"
-	rm -f $@
-	mkdir -p merge-tmp
-	for a in $(ALL_ARCHIVES); do \
-		( cd merge-tmp && ar x ../$$a ); \
-	done
-	( cd merge-tmp && ar rcs ../$@ *.o )
-	# Ensure we have a proper index
-	ranlib $@
-	# Clean up
-	rm -rf merge-tmp
-
-build/libstable-diffusion.a:
-	@echo "Building SD with $(BUILD_TYPE) build type and $(CMAKE_ARGS)"
-ifneq (,$(findstring sycl,$(BUILD_TYPE)))
-	+bash -c "source $(ONEAPI_VARS); \
-	mkdir -p build && \
-	cd build && \
-	cmake $(CMAKE_ARGS) ../../../../../sources/stablediffusion-ggml.cpp && \
-	cmake --build . --config Release"
-else
-	mkdir -p build && \
-	cd build && \
-	cmake $(CMAKE_ARGS) ../../../../../sources/stablediffusion-ggml.cpp && \
-	cmake --build . --config Release
-endif
-	$(MAKE) $(COMBINED_LIB)
-
-gosd.o:
-	$(CXX) $(CXXFLAGS) gosd.cpp -o gosd.o -c
-
-libsd.a: gosd.o
-	cp $(INCLUDE_PATH)/build/libstable-diffusion.a ./libsd.a
-	$(AR) rcs libsd.a gosd.o
-
-clean:
-	rm -rf gosd.o libsd.a build $(COMBINED_LIB)
--- a/backend/go/image/stablediffusion-ggml/gosd.cpp
+++ b/backend/go/image/stablediffusion-ggml/gosd.cpp
@@ -1,228 +0,0 @@
-#include <stdio.h>
-#include <string.h>
-#include <time.h>
-#include <iostream>
-#include <random>
-#include <string>
-#include <vector>
-#include "gosd.h"
-
-// #include "preprocessing.hpp"
-#include "flux.hpp"
-#include "stable-diffusion.h"
-
-#define STB_IMAGE_IMPLEMENTATION
-#define STB_IMAGE_STATIC
-#include "stb_image.h"
-
-#define STB_IMAGE_WRITE_IMPLEMENTATION
-#define STB_IMAGE_WRITE_STATIC
-#include "stb_image_write.h"
-
-#define STB_IMAGE_RESIZE_IMPLEMENTATION
-#define STB_IMAGE_RESIZE_STATIC
-#include "stb_image_resize.h"
-
-// Names of the sampler method, same order as enum sample_method in stable-diffusion.h
-const char* sample_method_str[] = {
-    "euler_a",
-    "euler",
-    "heun",
-    "dpm2",
-    "dpm++2s_a",
-    "dpm++2m",
-    "dpm++2mv2",
-    "ipndm",
-    "ipndm_v",
-    "lcm",
-};
-
-// Names of the sigma schedule overrides, same order as sample_schedule in stable-diffusion.h
-const char* schedule_str[] = {
-    "default",
-    "discrete",
-    "karras",
-    "exponential",
-    "ays",
-    "gits",
-};
-
-sd_ctx_t* sd_c;
-
-sample_method_t sample_method;
-
-int load_model(char *model, char* options[], int threads, int diff) {
-    fprintf (stderr, "Loading model!\n");
-
-    char *stableDiffusionModel = "";
-    if (diff == 1 ) {
-        stableDiffusionModel = model;
-        model = "";
-    }
-
-    // decode options. Options are in form optname:optvale, or if booleans only optname.
-    char *clip_l_path  = "";
-    char *clip_g_path  = "";
-    char *t5xxl_path  = "";
-    char *vae_path  = "";
-    char *scheduler = "";
-    char *sampler = "";
-
-    // If options is not NULL, parse options
-    for (int i = 0; options[i] != NULL; i++) {
-        char *optname = strtok(options[i], ":");
-        char *optval = strtok(NULL, ":");
-        if (optval == NULL) {
-            optval = "true";
-        }
-
-        if (!strcmp(optname, "clip_l_path")) {
-            clip_l_path = optval;
-        }
-        if (!strcmp(optname, "clip_g_path")) {
-            clip_g_path = optval;
-        }
-        if (!strcmp(optname, "t5xxl_path")) {
-            t5xxl_path = optval;
-        }
-        if (!strcmp(optname, "vae_path")) {
-            vae_path = optval;
-        }
-        if (!strcmp(optname, "scheduler")) {
-            scheduler = optval;
-        }
-        if (!strcmp(optname, "sampler")) {
-            sampler = optval;
-        }
-    }
-
-    int sample_method_found = -1;
-    for (int m = 0; m < N_SAMPLE_METHODS; m++) {
-        if (!strcmp(sampler, sample_method_str[m])) {
-            sample_method_found = m;
-        }
-    }
-    if (sample_method_found == -1) {
-        fprintf(stderr, "Invalid sample method, default to EULER_A!\n");
-        sample_method_found = EULER_A;
-    }
-    sample_method = (sample_method_t)sample_method_found;
-
-    int schedule_found            = -1;
-    for (int d = 0; d < N_SCHEDULES; d++) {
-        if (!strcmp(scheduler, schedule_str[d])) {
-            schedule_found = d;
-                fprintf (stderr, "Found scheduler: %s\n", scheduler);
-
-        }
-    }
-
-    if (schedule_found == -1) {
-        fprintf (stderr, "Invalid scheduler! using DEFAULT\n");
-        schedule_found = DEFAULT;
-    }
-
-    schedule_t schedule = (schedule_t)schedule_found;
-    
-    fprintf (stderr, "Creating context\n");
-    sd_ctx_t* sd_ctx = new_sd_ctx(model,
-                                  clip_l_path,
-                                  clip_g_path,
-                                  t5xxl_path,
-                                  stableDiffusionModel,
-                                  vae_path,
-                                  "",
-                                  "",
-                                  "",
-                                  "",
-                                  "",
-                                  false,
-                                  false,
-                                  false,
-                                  threads,
-                                  SD_TYPE_COUNT,
-                                  STD_DEFAULT_RNG,
-                                  schedule,
-                                  false,
-                                  false,
-                                  false,
-                                  false);
-
-    if (sd_ctx == NULL) {
-        fprintf (stderr, "failed loading model (generic error)\n");
-        return 1;
-    }
-    fprintf (stderr, "Created context: OK\n");
-
-    sd_c = sd_ctx;
-
-    return 0;
-}
-
-int gen_image(char *text, char *negativeText, int width, int height, int steps, int seed , char *dst, float cfg_scale) {
-
-    sd_image_t* results;
-
-    std::vector<int> skip_layers = {7, 8, 9};
-
-    fprintf (stderr, "Generating image\n");
-
-    results = txt2img(sd_c,
-                            text,
-                            negativeText,
-                            -1, //clip_skip
-                            cfg_scale, // sfg_scale
-                            3.5f,
-                            width,
-                            height,
-                            sample_method, 
-                            steps,
-                            seed,
-                            1,
-                            NULL,
-                            0.9f,
-                            20.f,
-                            false,
-                            "",
-                            skip_layers.data(),
-                            skip_layers.size(),
-                            0,
-                            0.01,
-                            0.2);
-
-    if (results == NULL) {
-        fprintf (stderr, "NO results\n");
-        return 1;
-    }
-
-    if (results[0].data == NULL) {
-        fprintf (stderr, "Results with no data\n");
-        return 1;
-    }
-
-    fprintf (stderr, "Writing PNG\n");
-
-    fprintf (stderr, "DST: %s\n", dst);
-    fprintf (stderr, "Width: %d\n", results[0].width);
-    fprintf (stderr, "Height: %d\n", results[0].height);
-    fprintf (stderr, "Channel: %d\n", results[0].channel);
-    fprintf (stderr, "Data: %p\n", results[0].data);
-
-    stbi_write_png(dst, results[0].width, results[0].height, results[0].channel,
-                       results[0].data, 0, NULL);
-    fprintf (stderr, "Saved resulting image to '%s'\n", dst);
-
-    // TODO: free results. Why does it crash?
-
-    free(results[0].data);
-    results[0].data = NULL;
-    free(results);
-    fprintf (stderr, "gen_image is done", dst);
-
-    return 0;
-}
-
-int unload() {
-    free_sd_ctx(sd_c);
-}
-
--- a/backend/go/image/stablediffusion-ggml/gosd.go
+++ b/backend/go/image/stablediffusion-ggml/gosd.go
@@ -1,96 +0,0 @@
-package main
-
-// #cgo CXXFLAGS: -I${SRCDIR}/../../../../sources/stablediffusion-ggml.cpp/thirdparty -I${SRCDIR}/../../../../sources/stablediffusion-ggml.cpp -I${SRCDIR}/../../../../sources/stablediffusion-ggml.cpp/ggml/include
-// #cgo LDFLAGS: -L${SRCDIR}/ -lsd -lstdc++ -lm -lggmlall -lgomp
-// #include <gosd.h>
-// #include <stdlib.h>
-import "C"
-
-import (
-	"fmt"
-	"os"
-	"path/filepath"
-	"strings"
-	"unsafe"
-
-	"github.com/mudler/LocalAI/pkg/grpc/base"
-	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
-	"github.com/mudler/LocalAI/pkg/utils"
-)
-
-type SDGGML struct {
-	base.SingleThread
-	threads      int
-	sampleMethod string
-	cfgScale     float32
-}
-
-func (sd *SDGGML) Load(opts *pb.ModelOptions) error {
-
-	sd.threads = int(opts.Threads)
-
-	modelFile := C.CString(opts.ModelFile)
-	defer C.free(unsafe.Pointer(modelFile))
-
-	var options **C.char
-	// prepare the options array to pass to C
-
-	size := C.size_t(unsafe.Sizeof((*C.char)(nil)))
-	length := C.size_t(len(opts.Options))
-	options = (**C.char)(C.malloc(length * size))
-	view := (*[1 << 30]*C.char)(unsafe.Pointer(options))[0:len(opts.Options):len(opts.Options)]
-
-	var diffusionModel int
-
-	var oo []string
-	for _, op := range opts.Options {
-		if op == "diffusion_model" {
-			diffusionModel = 1
-			continue
-		}
-
-		// If it's an option path, we resolve absolute path from the model path
-		if strings.Contains(op, ":") && strings.Contains(op, "path") {
-			data := strings.Split(op, ":")
-			data[1] = filepath.Join(opts.ModelPath, data[1])
-			if err := utils.VerifyPath(data[1], opts.ModelPath); err == nil {
-				oo = append(oo, strings.Join(data, ":"))
-			}
-		} else {
-			oo = append(oo, op)
-		}
-	}
-
-	fmt.Fprintf(os.Stderr, "Options: %+v\n", oo)
-
-	for i, x := range oo {
-		view[i] = C.CString(x)
-	}
-
-	sd.cfgScale = opts.CFGScale
-
-	ret := C.load_model(modelFile, options, C.int(opts.Threads), C.int(diffusionModel))
-	if ret != 0 {
-		return fmt.Errorf("could not load model")
-	}
-
-	return nil
-}
-
-func (sd *SDGGML) GenerateImage(opts *pb.GenerateImageRequest) error {
-	t := C.CString(opts.PositivePrompt)
-	defer C.free(unsafe.Pointer(t))
-
-	dst := C.CString(opts.Dst)
-	defer C.free(unsafe.Pointer(dst))
-
-	negative := C.CString(opts.NegativePrompt)
-	defer C.free(unsafe.Pointer(negative))
-
-	ret := C.gen_image(t, negative, C.int(opts.Width), C.int(opts.Height), C.int(opts.Step), C.int(opts.Seed), dst, C.float(sd.cfgScale))
-	if ret != 0 {
-		return fmt.Errorf("inference failed")
-	}
-
-	return nil
-}
--- a/backend/go/image/stablediffusion-ggml/gosd.h
+++ b/backend/go/image/stablediffusion-ggml/gosd.h
@@ -1,8 +0,0 @@
-#ifdef __cplusplus
-extern "C" {
-#endif
-int load_model(char *model, char* options[], int threads, int diffusionModel);
-int gen_image(char *text, char *negativeText, int width, int height, int steps, int seed, char *dst, float cfg_scale);
-#ifdef __cplusplus
-}
-#endif
--- a/backend/go/image/stablediffusion-ggml/main.go
+++ b/backend/go/image/stablediffusion-ggml/main.go
@@ -1,20 +0,0 @@
-package main
-
-// Note: this is started internally by LocalAI and a server is allocated for each model
-import (
-	"flag"
-
-	grpc "github.com/mudler/LocalAI/pkg/grpc"
-)
-
-var (
-	addr = flag.String("addr", "localhost:50051", "the address to connect to")
-)
-
-func main() {
-	flag.Parse()
-
-	if err := grpc.StartServer(*addr, &SDGGML{}); err != nil {
-		panic(err)
-	}
-}
--- a/backend/go/llm/bert/bert.go
+++ b/backend/go/llm/bert/bert.go
@@ -0,0 +1,34 @@
+package main
+
+// This is a wrapper to statisfy the GRPC service interface
+// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
+import (
+	bert "github.com/go-skynet/go-bert.cpp"
+
+	"github.com/mudler/LocalAI/pkg/grpc/base"
+	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
+)
+
+type Embeddings struct {
+	base.SingleThread
+	bert *bert.Bert
+}
+
+func (llm *Embeddings) Load(opts *pb.ModelOptions) error {
+	model, err := bert.New(opts.ModelFile)
+	llm.bert = model
+	return err
+}
+
+func (llm *Embeddings) Embeddings(opts *pb.PredictOptions) ([]float32, error) {
+
+	if len(opts.EmbeddingTokens) > 0 {
+		tokens := []int{}
+		for _, t := range opts.EmbeddingTokens {
+			tokens = append(tokens, int(t))
+		}
+		return llm.bert.TokenEmbeddings(tokens, bert.SetThreads(int(opts.Threads)))
+	}
+
+	return llm.bert.Embeddings(opts.Embeddings, bert.SetThreads(int(opts.Threads)))
+}
--- a/backend/go/llm/bert/main.go
+++ b/backend/go/llm/bert/main.go
@@ -1,6 +1,7 @@
 package main

 // Note: this is started internally by LocalAI and a server is allocated for each model
+
 import (
 	"flag"

@@ -14,7 +15,7 @@ var (
 func main() {
 	flag.Parse()

-	if err := grpc.StartServer(*addr, &Bark{}); err != nil {
+	if err := grpc.StartServer(*addr, &Embeddings{}); err != nil {
 		panic(err)
 	}
 }
--- a/backend/go/vad/silero/main.go
+++ b/backend/go/vad/silero/main.go
@@ -15,7 +15,7 @@ var (
 func main() {
 	flag.Parse()

-	if err := grpc.StartServer(*addr, &VAD{}); err != nil {
+	if err := grpc.StartServer(*addr, &LLM{}); err != nil {
 		panic(err)
 	}
 }
--- a/backend/go/llm/rwkv/rwkv.go
+++ b/backend/go/llm/rwkv/rwkv.go
@@ -0,0 +1,95 @@
+package main
+
+// This is a wrapper to statisfy the GRPC service interface
+// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
+import (
+	"fmt"
+	"path/filepath"
+
+	"github.com/donomii/go-rwkv.cpp"
+	"github.com/mudler/LocalAI/pkg/grpc/base"
+	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
+)
+
+const tokenizerSuffix = ".tokenizer.json"
+
+type LLM struct {
+	base.SingleThread
+
+	rwkv *rwkv.RwkvState
+}
+
+func (llm *LLM) Load(opts *pb.ModelOptions) error {
+	tokenizerFile := opts.Tokenizer
+	if tokenizerFile == "" {
+		modelFile := filepath.Base(opts.ModelFile)
+		tokenizerFile = modelFile + tokenizerSuffix
+	}
+	modelPath := filepath.Dir(opts.ModelFile)
+	tokenizerPath := filepath.Join(modelPath, tokenizerFile)
+
+	model := rwkv.LoadFiles(opts.ModelFile, tokenizerPath, uint32(opts.GetThreads()))
+
+	if model == nil {
+		return fmt.Errorf("rwkv could not load model")
+	}
+	llm.rwkv = model
+	return nil
+}
+
+func (llm *LLM) Predict(opts *pb.PredictOptions) (string, error) {
+	stopWord := "\n"
+	if len(opts.StopPrompts) > 0 {
+		stopWord = opts.StopPrompts[0]
+	}
+
+	if err := llm.rwkv.ProcessInput(opts.Prompt); err != nil {
+		return "", err
+	}
+
+	response := llm.rwkv.GenerateResponse(int(opts.Tokens), stopWord, float32(opts.Temperature), float32(opts.TopP), nil)
+
+	return response, nil
+}
+
+func (llm *LLM) PredictStream(opts *pb.PredictOptions, results chan string) error {
+	go func() {
+
+		stopWord := "\n"
+		if len(opts.StopPrompts) > 0 {
+			stopWord = opts.StopPrompts[0]
+		}
+
+		if err := llm.rwkv.ProcessInput(opts.Prompt); err != nil {
+			fmt.Println("Error processing input: ", err)
+			return
+		}
+
+		llm.rwkv.GenerateResponse(int(opts.Tokens), stopWord, float32(opts.Temperature), float32(opts.TopP), func(s string) bool {
+			results <- s
+			return true
+		})
+		close(results)
+	}()
+
+	return nil
+}
+
+func (llm *LLM) TokenizeString(opts *pb.PredictOptions) (pb.TokenizationResponse, error) {
+	tokens, err := llm.rwkv.Tokenizer.Encode(opts.Prompt)
+	if err != nil {
+		return pb.TokenizationResponse{}, err
+	}
+
+	l := len(tokens)
+	i32Tokens := make([]int32, l)
+
+	for i, t := range tokens {
+		i32Tokens[i] = int32(t.ID)
+	}
+
+	return pb.TokenizationResponse{
+		Length: int32(l),
+		Tokens: i32Tokens,
+	}, nil
+}
--- a/backend/go/vad/silero/vad.go
+++ b/backend/go/vad/silero/vad.go
@@ -1,54 +0,0 @@
-package main
-
-// This is a wrapper to statisfy the GRPC service interface
-// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
-import (
-	"fmt"
-
-	"github.com/mudler/LocalAI/pkg/grpc/base"
-	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
-	"github.com/streamer45/silero-vad-go/speech"
-)
-
-type VAD struct {
-	base.SingleThread
-	detector *speech.Detector
-}
-
-func (vad *VAD) Load(opts *pb.ModelOptions) error {
-	v, err := speech.NewDetector(speech.DetectorConfig{
-		ModelPath:  opts.ModelFile,
-		SampleRate: 16000,
-		//WindowSize:           1024,
-		Threshold:            0.5,
-		MinSilenceDurationMs: 100,
-		SpeechPadMs:          30,
-	})
-	if err != nil {
-		return fmt.Errorf("create silero detector: %w", err)
-	}
-
-	vad.detector = v
-	return err
-}
-
-func (vad *VAD) VAD(req *pb.VADRequest) (pb.VADResponse, error) {
-	audio := req.Audio
-
-	segments, err := vad.detector.Detect(audio)
-	if err != nil {
-		return pb.VADResponse{}, fmt.Errorf("detect: %w", err)
-	}
-
-	vadSegments := []*pb.VADSegment{}
-	for _, s := range segments {
-		vadSegments = append(vadSegments, &pb.VADSegment{
-			Start: float32(s.SpeechStartAt),
-			End:   float32(s.SpeechEndAt),
-		})
-	}
-
-	return pb.VADResponse{
-		Segments: vadSegments,
-	}, nil
-}
--- a/backend/python/autogptq/requirements-intel.txt
+++ b/backend/python/autogptq/requirements-intel.txt
@@ -1,6 +1,5 @@
 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
-torch==2.3.1+cxx11.abi
-oneccl_bind_pt==2.3.100+xpu
+intel-extension-for-pytorch
+torch
 optimum[openvino]
-setuptools
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/autogptq/requirements.txt
+++ b/backend/python/autogptq/requirements.txt
@@ -1,6 +1,6 @@
 accelerate
 auto-gptq==0.7.1
-grpcio==1.69.0
+grpcio==1.67.0
 protobuf
 certifi
 transformers
--- a/backend/python/bark/requirements-intel.txt
+++ b/backend/python/bark/requirements-intel.txt
@@ -1,9 +1,8 @@
 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
-torch==2.3.1+cxx11.abi
-torchaudio==2.3.1+cxx11.abi
-oneccl_bind_pt==2.3.100+xpu
+intel-extension-for-pytorch
+torch
+torchaudio
 optimum[openvino]
-setuptools
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
 transformers
 accelerate
--- a/backend/python/bark/requirements.txt
+++ b/backend/python/bark/requirements.txt
@@ -1,4 +1,4 @@
 bark==0.1.5
-grpcio==1.69.0
+grpcio==1.67.0
 protobuf
 certifi
--- a/backend/python/common/libbackend.sh
+++ b/backend/python/common/libbackend.sh
@@ -17,9 +17,6 @@
 # LIMIT_TARGETS="cublas12"
 # source $(dirname $0)/../common/libbackend.sh
 #
-
-PYTHON_VERSION="3.10"
-
 function init() {
    # Name of the backend (directory name)
    BACKEND_NAME=${PWD##*/}
@@ -91,7 +88,7 @@ function getBuildProfile() {
 # always result in an activated virtual environment
 function ensureVenv() {
    if [ ! -d "${EDIR}/venv" ]; then
-        uv venv --python ${PYTHON_VERSION} ${EDIR}/venv
+        uv venv ${EDIR}/venv
        echo "virtualenv created"
    fi

--- a/backend/python/common/template/Makefile
+++ b/backend/python/common/template/Makefile
@@ -1,9 +1,8 @@
 .DEFAULT_GOAL := install

 .PHONY: install
-install:
+install: protogen
 	bash install.sh
-	$(MAKE) protogen

 .PHONY: protogen
 protogen: backend_pb2_grpc.py backend_pb2.py
@@ -13,7 +12,7 @@ protogen-clean:
 	$(RM) backend_pb2_grpc.py backend_pb2.py

 backend_pb2_grpc.py backend_pb2.py:
-	bash protogen.sh
+	python3 -m grpc_tools.protoc -I../.. --python_out=. --grpc_python_out=. backend.proto

 .PHONY: clean
 clean: protogen-clean
--- a/backend/python/common/template/protogen.sh
+++ b/backend/python/common/template/protogen.sh
@@ -1,6 +0,0 @@
-#!/bin/bash
-set -e
-
-source $(dirname $0)/../common/libbackend.sh
-
-python3 -m grpc_tools.protoc -I../.. --python_out=. --grpc_python_out=. backend.proto
--- a/backend/python/common/template/requirements-intel.txt
+++ b/backend/python/common/template/requirements-intel.txt
@@ -1,5 +1,4 @@
 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
-torch==2.3.1+cxx11.abi
-oneccl_bind_pt==2.3.100+xpu
+intel-extension-for-pytorch
+torch
 optimum[openvino]
--- a/backend/python/common/template/requirements.txt
+++ b/backend/python/common/template/requirements.txt
@@ -1,3 +1,2 @@
-grpcio==1.69.0
-protobuf
-grpcio-tools
+grpcio==1.67.0
+protobuf
--- a/backend/python/coqui/requirements-intel.txt
+++ b/backend/python/coqui/requirements-intel.txt
@@ -1,10 +1,9 @@
 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
-torch==2.3.1+cxx11.abi
-torchaudio==2.3.1+cxx11.abi
-oneccl_bind_pt==2.3.100+xpu
+intel-extension-for-pytorch
+torch
+torchaudio
 optimum[openvino]
-setuptools
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
 transformers
 accelerate
 coqui-tts
--- a/backend/python/coqui/requirements.txt
+++ b/backend/python/coqui/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.69.0
+grpcio==1.67.0
 protobuf
 certifi
 packaging==24.1
--- a/backend/python/diffusers/backend.py
+++ b/backend/python/diffusers/backend.py
@@ -17,7 +17,7 @@ import backend_pb2_grpc

 import grpc

-from diffusers import SanaPipeline, StableDiffusion3Pipeline, StableDiffusionXLPipeline, StableDiffusionDepth2ImgPipeline, DPMSolverMultistepScheduler, StableDiffusionPipeline, DiffusionPipeline, \
+from diffusers import StableDiffusion3Pipeline, StableDiffusionXLPipeline, StableDiffusionDepth2ImgPipeline, DPMSolverMultistepScheduler, StableDiffusionPipeline, DiffusionPipeline, \
    EulerAncestralDiscreteScheduler, FluxPipeline, FluxTransformer2DModel
 from diffusers import StableDiffusionImg2ImgPipeline, AutoPipelineForText2Image, ControlNetModel, StableVideoDiffusionPipeline
 from diffusers.pipelines.stable_diffusion import safety_checker
@@ -247,16 +247,11 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                        use_safetensors=True,
                        variant=variant)
            elif request.PipelineType == "FluxPipeline":
-                if fromSingleFile:
-                    self.pipe = FluxPipeline.from_single_file(modelFile,
-                                                              torch_dtype=torchType,
-                                                              use_safetensors=True)
-                else:
                    self.pipe = FluxPipeline.from_pretrained(
                        request.Model,
                        torch_dtype=torch.bfloat16)
-                if request.LowVRAM:
-                    self.pipe.enable_model_cpu_offload()
+                    if request.LowVRAM:
+                        self.pipe.enable_model_cpu_offload()
            elif request.PipelineType == "FluxTransformer2DModel":
                    dtype = torch.bfloat16
                    # specify from environment or default to "ChuckMcSneed/FLUX.1-dev"
@@ -275,13 +270,6 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):

                    if request.LowVRAM:
                        self.pipe.enable_model_cpu_offload()
-            elif request.PipelineType == "SanaPipeline":
-                self.pipe = SanaPipeline.from_pretrained(
-                    request.Model,
-                    variant="bf16",
-                    torch_dtype=torch.bfloat16)
-                self.pipe.vae.to(torch.bfloat16)
-                self.pipe.text_encoder.to(torch.bfloat16)

            if CLIPSKIP and request.CLIPSkip != 0:
                self.clip_skip = request.CLIPSkip
@@ -308,34 +296,22 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                self.pipe.controlnet = self.controlnet
            else:
                self.controlnet = None
-
-            if request.LoraAdapter and not os.path.isabs(request.LoraAdapter):
+            # Assume directory from request.ModelFile.
+            # Only if request.LoraAdapter it's not an absolute path
+            if request.LoraAdapter and request.ModelFile != "" and not os.path.isabs(request.LoraAdapter) and request.LoraAdapter:
+                # get base path of modelFile
+                modelFileBase = os.path.dirname(request.ModelFile)
                # modify LoraAdapter to be relative to modelFileBase
-                request.LoraAdapter = os.path.join(request.ModelPath, request.LoraAdapter)
-
+                request.LoraAdapter = os.path.join(modelFileBase, request.LoraAdapter)
            device = "cpu" if not request.CUDA else "cuda"
            self.device = device
            if request.LoraAdapter:
                # Check if its a local file and not a directory ( we load lora differently for a safetensor file )
                if os.path.exists(request.LoraAdapter) and not os.path.isdir(request.LoraAdapter):
+                    # self.load_lora_weights(request.LoraAdapter, 1, device, torchType)
                    self.pipe.load_lora_weights(request.LoraAdapter)
                else:
                    self.pipe.unet.load_attn_procs(request.LoraAdapter)
-            if len(request.LoraAdapters) > 0:
-                i = 0
-                adapters_name = []
-                adapters_weights = []
-                for adapter in request.LoraAdapters:
-                    if not os.path.isabs(adapter):
-                        adapter = os.path.join(request.ModelPath, adapter)
-                    self.pipe.load_lora_weights(adapter, adapter_name=f"adapter_{i}")
-                    adapters_name.append(f"adapter_{i}")
-                    i += 1
-
-                for adapters_weight in request.LoraScales:
-                    adapters_weights.append(adapters_weight)
-
-                self.pipe.set_adapters(adapters_name, adapter_weights=adapters_weights)

            if request.CUDA:
                self.pipe.to('cuda')
@@ -416,6 +392,8 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
        # create a dictionary of values for the parameters
        options = {
            "negative_prompt": request.negative_prompt,
+            "width": request.width,
+            "height": request.height,
            "num_inference_steps": steps,
        }

@@ -433,13 +411,13 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
        keys = options.keys()

        if request.EnableParameters != "":
-            keys = [key.strip() for key in request.EnableParameters.split(",")]
+            keys = request.EnableParameters.split(",")

        if request.EnableParameters == "none":
            keys = []

        # create a dictionary of parameters by using the keys from EnableParameters and the values from defaults
-        kwargs = {key: options.get(key) for key in keys if key in options}
+        kwargs = {key: options[key] for key in keys}

        # Set seed
        if request.seed > 0:
@@ -450,12 +428,6 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
        if self.PipelineType == "FluxPipeline":
            kwargs["max_sequence_length"] = 256

-        if request.width:
-            kwargs["width"] = request.width
-
-        if request.height:
-            kwargs["height"] = request.height
-
        if self.PipelineType == "FluxTransformer2DModel":
            kwargs["output_type"] = "pil"
            kwargs["generator"] = torch.Generator("cpu").manual_seed(0)
@@ -475,7 +447,6 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
            export_to_video(video_frames, request.dst)
            return backend_pb2.Result(message="Media generated successfully", success=True)

-        print(f"Generating image with {kwargs=}", file=sys.stderr)
        image = {}
        if COMPEL:
            conditioning, pooled = self.compel.build_conditioning_tensor(prompt)
--- a/backend/python/diffusers/requirements-intel.txt
+++ b/backend/python/diffusers/requirements-intel.txt
@@ -1,10 +1,9 @@
 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
-torch==2.3.1+cxx11.abi
-torchvision==0.18.1+cxx11.abi
-oneccl_bind_pt==2.3.100+xpu
+intel-extension-for-pytorch
+torch
+torchvision
 optimum[openvino]
-setuptools
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
 diffusers
 opencv-python
 transformers
--- a/backend/python/diffusers/requirements.txt
+++ b/backend/python/diffusers/requirements.txt
@@ -1,5 +1,5 @@
 setuptools
-grpcio==1.69.0
+grpcio==1.67.0
 pillow
 protobuf
 certifi
--- a/backend/python/exllama2/requirements.txt
+++ b/backend/python/exllama2/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.69.0
+grpcio==1.67.0
 protobuf
 certifi
 wheel
--- a/backend/python/mamba/requirements.txt
+++ b/backend/python/mamba/requirements.txt
@@ -1,3 +1,3 @@
-grpcio==1.69.0
+grpcio==1.67.0
 protobuf
 certifi
--- a/backend/python/openvoice/requirements-cpu.txt
+++ b/backend/python/openvoice/requirements-cpu.txt
@@ -1,7 +1,3 @@
 torch==2.4.1
 git+https://github.com/myshell-ai/MeloTTS.git
-git+https://github.com/myshell-ai/OpenVoice.git
-whisper-timestamped
-pydub==0.25.1
-wavmark==0.0.3
-eng_to_ipa==0.0.2
+git+https://github.com/myshell-ai/OpenVoice.git
--- a/backend/python/openvoice/requirements-cublas11.txt
+++ b/backend/python/openvoice/requirements-cublas11.txt
@@ -1,8 +1,4 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
 torch==2.4.1+cu118
 git+https://github.com/myshell-ai/MeloTTS.git
-git+https://github.com/myshell-ai/OpenVoice.git
-whisper-timestamped
-pydub==0.25.1
-wavmark==0.0.3
-eng_to_ipa==0.0.2
+git+https://github.com/myshell-ai/OpenVoice.git
--- a/backend/python/openvoice/requirements-cublas12.txt
+++ b/backend/python/openvoice/requirements-cublas12.txt
@@ -1,7 +1,3 @@
 torch==2.4.1
 git+https://github.com/myshell-ai/MeloTTS.git
-git+https://github.com/myshell-ai/OpenVoice.git
-whisper-timestamped
-pydub==0.25.1
-wavmark==0.0.3
-eng_to_ipa==0.0.2
+git+https://github.com/myshell-ai/OpenVoice.git
--- a/backend/python/openvoice/requirements-hipblas.txt
+++ b/backend/python/openvoice/requirements-hipblas.txt
@@ -1,8 +1,4 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
 torch==2.4.1+rocm6.0
 git+https://github.com/myshell-ai/MeloTTS.git
-git+https://github.com/myshell-ai/OpenVoice.git
-whisper-timestamped
-pydub==0.25.1
-wavmark==0.0.3
-eng_to_ipa==0.0.2
+git+https://github.com/myshell-ai/OpenVoice.git
--- a/backend/python/openvoice/requirements-intel.txt
+++ b/backend/python/openvoice/requirements-intel.txt
@@ -1,15 +1,14 @@
 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
-torch==2.3.1+cxx11.abi
-torchaudio==2.3.1+cxx11.abi
-oneccl_bind_pt==2.3.100+xpu
+intel-extension-for-pytorch
+torch
 optimum[openvino]
-grpcio==1.69.0
+grpcio==1.67.0
 protobuf
 librosa==0.9.1
 faster-whisper==0.9.0
 pydub==0.25.1
 wavmark==0.0.3
+numpy==1.22.0
 eng_to_ipa==0.0.2
 inflect==7.0.0
 unidecode==1.3.7
--- a/backend/python/openvoice/requirements.txt
+++ b/backend/python/openvoice/requirements.txt
@@ -1,17 +1,20 @@
-grpcio==1.69.0
+grpcio==1.67.0
 protobuf
 librosa
 faster-whisper
+pydub==0.25.1
+wavmark==0.0.3
+numpy==1.22.0
+eng_to_ipa==0.0.2
 inflect
 unidecode
+whisper-timestamped
 openai
 python-dotenv
 pypinyin
 cn2an==0.5.22
-numpy==1.22.0
 networkx==2.8.8
 jieba==0.42.1
-gradio==5.9.1
+gradio==3.48.0
 langid==1.1.6
 llvmlite==0.43.0
-setuptools
--- a/backend/python/parler-tts/Makefile
+++ b/backend/python/parler-tts/Makefile
@@ -12,10 +12,9 @@ export SKIP_CONDA=1
 endif

 .PHONY: parler-tts
-parler-tts:
+parler-tts: protogen
 	@echo "Installing $(CONDA_ENV_PATH)..."
 	bash install.sh $(CONDA_ENV_PATH)
-	$(MAKE) protogen

 .PHONY: run
 run: protogen
@@ -37,7 +36,7 @@ protogen-clean:
 	$(RM) backend_pb2_grpc.py backend_pb2.py

 backend_pb2_grpc.py backend_pb2.py:
-	bash protogen.sh
+	python3 -m grpc_tools.protoc -I../.. --python_out=. --grpc_python_out=. backend.proto

 .PHONY: clean
 clean: protogen-clean
--- a/backend/python/parler-tts/install.sh
+++ b/backend/python/parler-tts/install.sh
@@ -11,18 +11,16 @@ if [ "x${BUILD_PROFILE}" == "xintel" ]; then
    EXTRA_PIP_INSTALL_FLAGS+=" --upgrade --index-strategy=unsafe-first-match"
 fi

-
 installRequirements

-
 # https://github.com/descriptinc/audiotools/issues/101
 # incompatible protobuf versions.
-PYDIR=python3.10
-pyenv="${MY_DIR}/venv/lib/${PYDIR}/site-packages/google/protobuf/internal/"
+# PYDIR=python3.10
+# pyenv="${MY_DIR}/venv/lib/${PYDIR}/site-packages/google/protobuf/internal/"

-if [ ! -d ${pyenv} ]; then
-    echo "(parler-tts/install.sh): Error: ${pyenv} does not exist"
-    exit 1
-fi
+# if [ ! -d ${pyenv} ]; then
+#     echo "(parler-tts/install.sh): Error: ${pyenv} does not exist"
+#     exit 1
+# fi

-curl -L https://raw.githubusercontent.com/protocolbuffers/protobuf/main/python/google/protobuf/internal/builder.py -o ${pyenv}/builder.py
+# curl -L https://raw.githubusercontent.com/protocolbuffers/protobuf/main/python/google/protobuf/internal/builder.py -o ${pyenv}/builder.py
--- a/backend/python/parler-tts/protogen.sh
+++ b/backend/python/parler-tts/protogen.sh
@@ -1,6 +0,0 @@
-#!/bin/bash
-set -e
-
-source $(dirname $0)/../common/libbackend.sh
-
-python3 -m grpc_tools.protoc -I../.. --python_out=. --grpc_python_out=. backend.proto
--- a/backend/python/parler-tts/requirements-after.txt
+++ b/backend/python/parler-tts/requirements-after.txt
@@ -1,4 +1,4 @@
 git+https://github.com/huggingface/parler-tts.git@8e465f1b5fcd223478e07175cb40494d19ffbe17
 llvmlite==0.43.0
 numba==0.60.0
-grpcio-tools==1.42.0
+git+https://github.com/descriptinc/audiotools
--- a/backend/python/parler-tts/requirements-intel.txt
+++ b/backend/python/parler-tts/requirements-intel.txt
@@ -1,8 +1,8 @@
 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
-torch==2.3.1+cxx11.abi
-torchaudio==2.3.1+cxx11.abi
-oneccl_bind_pt==2.3.100+xpu
+intel-extension-for-pytorch
+torch
+torchaudio
 optimum[openvino]
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
 transformers
 accelerate
--- a/backend/python/parler-tts/requirements.txt
+++ b/backend/python/parler-tts/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.69.0
+grpcio==1.67.0
+protobuf
 certifi
-llvmlite==0.43.0
-setuptools
+llvmlite==0.43.0
--- a/backend/python/rerankers/requirements-intel.txt
+++ b/backend/python/rerankers/requirements-intel.txt
@@ -1,9 +1,8 @@
 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
+intel-extension-for-pytorch
 transformers
 accelerate
-torch==2.3.1+cxx11.abi
-oneccl_bind_pt==2.3.100+xpu
+torch
 rerankers[transformers]
 optimum[openvino]
-setuptools
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/rerankers/requirements.txt
+++ b/backend/python/rerankers/requirements.txt
@@ -1,3 +1,3 @@
-grpcio==1.69.0
+grpcio==1.67.0
 protobuf
 certifi
--- a/backend/python/sentencetransformers/requirements-cpu.txt
+++ b/backend/python/sentencetransformers/requirements-cpu.txt
@@ -2,5 +2,5 @@ torch==2.4.1
 accelerate
 transformers
 bitsandbytes
-sentence-transformers==3.3.1
+sentence-transformers==3.2.0
 transformers
--- a/backend/python/sentencetransformers/requirements-cublas11.txt
+++ b/backend/python/sentencetransformers/requirements-cublas11.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
 torch==2.4.1+cu118
 accelerate
-sentence-transformers==3.3.1
+sentence-transformers==3.2.0
 transformers
--- a/backend/python/sentencetransformers/requirements-cublas12.txt
+++ b/backend/python/sentencetransformers/requirements-cublas12.txt
@@ -1,4 +1,4 @@
 torch==2.4.1
 accelerate
-sentence-transformers==3.3.1
+sentence-transformers==3.2.0
 transformers
--- a/backend/python/sentencetransformers/requirements-hipblas.txt
+++ b/backend/python/sentencetransformers/requirements-hipblas.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
 torch==2.4.1+rocm6.0
 accelerate
-sentence-transformers==3.3.1
+sentence-transformers==3.2.0
 transformers
--- a/backend/python/sentencetransformers/requirements-intel.txt
+++ b/backend/python/sentencetransformers/requirements-intel.txt
@@ -1,9 +1,8 @@
 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
-torch==2.3.1+cxx11.abi
-oneccl_bind_pt==2.3.100+xpu
+intel-extension-for-pytorch
+torch
 optimum[openvino]
-setuptools
+setuptools==69.5.1 # https://github.com/mudler/LocalAI/issues/2406
 accelerate
-sentence-transformers==3.3.1
+sentence-transformers==3.2.0
 transformers
--- a/backend/python/sentencetransformers/requirements.txt
+++ b/backend/python/sentencetransformers/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.69.0
+grpcio==1.67.0
 protobuf
 certifi
 datasets
--- a/backend/python/transformers-musicgen/requirements-intel.txt
+++ b/backend/python/transformers-musicgen/requirements-intel.txt
@@ -1,8 +1,7 @@
 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
+intel-extension-for-pytorch
 transformers
-oneccl_bind_pt==2.3.100+xpu
 accelerate
-torch==2.3.1+cxx11.abi
+torch
 optimum[openvino]
-setuptools
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/transformers-musicgen/requirements.txt
+++ b/backend/python/transformers-musicgen/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.69.0
+grpcio==1.67.0
 protobuf
 scipy==1.14.0
 certifi
--- a/backend/python/transformers/requirements-intel.txt
+++ b/backend/python/transformers/requirements-intel.txt
@@ -1,7 +1,6 @@
 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
-torch==2.3.1+cxx11.abi
-oneccl_bind_pt==2.3.100+xpu
+intel-extension-for-pytorch
+torch
 optimum[openvino]
 intel-extension-for-transformers
 bitsandbytes
--- a/backend/python/transformers/requirements.txt
+++ b/backend/python/transformers/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.69.0
+grpcio==1.67.0
 protobuf
 certifi
-setuptools
+setuptools==69.5.1 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/vall-e-x/requirements-intel.txt
+++ b/backend/python/vall-e-x/requirements-intel.txt
@@ -1,7 +1,7 @@
 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
+intel-extension-for-pytorch
 accelerate
-torch==2.3.1+cxx11.abi
-torchaudio==2.3.1+cxx11.abi
+torch
+torchaudio
 optimum[openvino]
-oneccl_bind_pt==2.3.100+xpu
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/vall-e-x/requirements.txt
+++ b/backend/python/vall-e-x/requirements.txt
@@ -1,4 +1,3 @@
-grpcio==1.69.0
+grpcio==1.67.0
 protobuf
-certifi
-setuptools
+certifi
--- a/backend/python/vllm/install.sh
+++ b/backend/python/vllm/install.sh
@@ -22,7 +22,7 @@ if [ "x${BUILD_TYPE}" == "x" ] && [ "x${FROM_SOURCE}" == "xtrue" ]; then
            git clone https://github.com/vllm-project/vllm
        fi
        pushd vllm
-            uv pip install wheel packaging ninja "setuptools>=49.4.0" numpy typing-extensions pillow setuptools-scm grpcio==1.68.1 protobuf bitsandbytes
+            uv pip install wheel packaging ninja "setuptools>=49.4.0" numpy typing-extensions pillow setuptools-scm grpcio==1.67.0 protobuf bitsandbytes
            uv pip install -v -r requirements-cpu.txt --extra-index-url https://download.pytorch.org/whl/cpu
            VLLM_TARGET_DEVICE=cpu python setup.py install
        popd
--- a/backend/python/vllm/requirements-intel.txt
+++ b/backend/python/vllm/requirements-intel.txt
@@ -1,9 +1,8 @@
 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
+intel-extension-for-pytorch
 accelerate
-torch==2.3.1+cxx11.abi
+torch
 transformers
 optimum[openvino]
-setuptools
-bitsandbytes
-oneccl_bind_pt==2.3.100+xpu
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
+bitsandbytes
--- a/backend/python/vllm/requirements.txt
+++ b/backend/python/vllm/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.69.0
+grpcio==1.67.0
 protobuf
 certifi
 setuptools
--- a/core/application.go
+++ b/core/application.go
@@ -0,0 +1,38 @@
+package core
+
+import (
+	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/core/services"
+	"github.com/mudler/LocalAI/pkg/model"
+)
+
+// The purpose of this structure is to hold pointers to all initialized services, to make plumbing easy
+// Perhaps a proper DI system is worth it in the future, but for now keep things simple.
+type Application struct {
+
+	// Application-Level Config
+	ApplicationConfig *config.ApplicationConfig
+	// ApplicationState *ApplicationState
+
+	// Core Low-Level Services
+	BackendConfigLoader *config.BackendConfigLoader
+	ModelLoader         *model.ModelLoader
+
+	// Backend Services
+	// EmbeddingsBackendService      *backend.EmbeddingsBackendService
+	// ImageGenerationBackendService *backend.ImageGenerationBackendService
+	// LLMBackendService             *backend.LLMBackendService
+	// TranscriptionBackendService *backend.TranscriptionBackendService
+	// TextToSpeechBackendService  *backend.TextToSpeechBackendService
+
+	// LocalAI System Services
+	BackendMonitorService *services.BackendMonitorService
+	GalleryService        *services.GalleryService
+	LocalAIMetricsService *services.LocalAIMetricsService
+	// OpenAIService         *services.OpenAIService
+}
+
+// TODO [NEXT PR?]: Break up ApplicationConfig.
+// Migrate over stuff that is not set via config at all - especially runtime stuff
+type ApplicationState struct {
+}
--- a/core/application/application.go
+++ b/core/application/application.go
@@ -1,39 +0,0 @@
-package application
-
-import (
-	"github.com/mudler/LocalAI/core/config"
-	"github.com/mudler/LocalAI/pkg/model"
-	"github.com/mudler/LocalAI/pkg/templates"
-)
-
-type Application struct {
-	backendLoader      *config.BackendConfigLoader
-	modelLoader        *model.ModelLoader
-	applicationConfig  *config.ApplicationConfig
-	templatesEvaluator *templates.Evaluator
-}
-
-func newApplication(appConfig *config.ApplicationConfig) *Application {
-	return &Application{
-		backendLoader:      config.NewBackendConfigLoader(appConfig.ModelPath),
-		modelLoader:        model.NewModelLoader(appConfig.ModelPath),
-		applicationConfig:  appConfig,
-		templatesEvaluator: templates.NewEvaluator(appConfig.ModelPath),
-	}
-}
-
-func (a *Application) BackendLoader() *config.BackendConfigLoader {
-	return a.backendLoader
-}
-
-func (a *Application) ModelLoader() *model.ModelLoader {
-	return a.modelLoader
-}
-
-func (a *Application) ApplicationConfig() *config.ApplicationConfig {
-	return a.applicationConfig
-}
-
-func (a *Application) TemplatesEvaluator() *templates.Evaluator {
-	return a.templatesEvaluator
-}
--- a/core/backend/embeddings.go
+++ b/core/backend/embeddings.go
@@ -11,9 +11,17 @@ import (

 func ModelEmbedding(s string, tokens []int, loader *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (func() ([]float32, error), error) {

-	opts := ModelOptions(backendConfig, appConfig)
+	var inferenceModel interface{}
+	var err error

-	inferenceModel, err := loader.Load(opts...)
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{})
+
+	if backendConfig.Backend == "" {
+		inferenceModel, err = loader.GreedyLoader(opts...)
+	} else {
+		opts = append(opts, model.WithBackendString(backendConfig.Backend))
+		inferenceModel, err = loader.BackendLoader(opts...)
+	}
 	if err != nil {
 		return nil, err
 	}
--- a/core/backend/image.go
+++ b/core/backend/image.go
@@ -9,8 +9,9 @@ import (

 func ImageGeneration(height, width, mode, step, seed int, positive_prompt, negative_prompt, src, dst string, loader *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (func() error, error) {

-	opts := ModelOptions(backendConfig, appConfig)
-	inferenceModel, err := loader.Load(
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{})
+
+	inferenceModel, err := loader.BackendLoader(
 		opts...,
 	)
 	if err != nil {
--- a/core/backend/llm.go
+++ b/core/backend/llm.go
@@ -16,15 +16,15 @@ import (
 	"github.com/mudler/LocalAI/core/schema"

 	"github.com/mudler/LocalAI/core/gallery"
+	"github.com/mudler/LocalAI/pkg/grpc"
 	"github.com/mudler/LocalAI/pkg/grpc/proto"
 	model "github.com/mudler/LocalAI/pkg/model"
 	"github.com/mudler/LocalAI/pkg/utils"
 )

 type LLMResponse struct {
-	Response    string // should this be []byte?
-	Usage       TokenUsage
-	AudioOutput string
+	Response string // should this be []byte?
+	Usage    TokenUsage
 }

 type TokenUsage struct {
@@ -35,6 +35,15 @@ type TokenUsage struct {
 func ModelInference(ctx context.Context, s string, messages []schema.Message, images, videos, audios []string, loader *model.ModelLoader, c config.BackendConfig, o *config.ApplicationConfig, tokenCallback func(string, TokenUsage) bool) (func() (LLMResponse, error), error) {
 	modelFile := c.Model

+	var inferenceModel grpc.Backend
+	var err error
+
+	opts := ModelOptions(c, o, []model.Option{})
+
+	if c.Backend != "" {
+		opts = append(opts, model.WithBackendString(c.Backend))
+	}
+
 	// Check if the modelFile exists, if it doesn't try to load it from the gallery
 	if o.AutoloadGalleries { // experimental
 		if _, err := os.Stat(modelFile); os.IsNotExist(err) {
@@ -47,8 +56,12 @@ func ModelInference(ctx context.Context, s string, messages []schema.Message, im
 		}
 	}

-	opts := ModelOptions(c, o)
-	inferenceModel, err := loader.Load(opts...)
+	if c.Backend == "" {
+		inferenceModel, err = loader.GreedyLoader(opts...)
+	} else {
+		inferenceModel, err = loader.BackendLoader(opts...)
+	}
+
 	if err != nil {
 		return nil, err
 	}
@@ -118,12 +131,8 @@ func ModelInference(ctx context.Context, s string, messages []schema.Message, im
 			ss := ""

 			var partialRune []byte
-			err := inferenceModel.PredictStream(ctx, opts, func(reply *proto.Reply) {
-				msg := reply.Message
-				partialRune = append(partialRune, msg...)
-
-				tokenUsage.Prompt = int(reply.PromptTokens)
-				tokenUsage.Completion = int(reply.Tokens)
+			err := inferenceModel.PredictStream(ctx, opts, func(chars []byte) {
+				partialRune = append(partialRune, chars...)

 				for len(partialRune) > 0 {
 					r, size := utf8.DecodeRune(partialRune)
@@ -137,10 +146,6 @@ func ModelInference(ctx context.Context, s string, messages []schema.Message, im

 					partialRune = partialRune[size:]
 				}
-
-				if len(msg) == 0 {
-					tokenCallback("", tokenUsage)
-				}
 			})
 			return LLMResponse{
 				Response: ss,
--- a/core/backend/options.go
+++ b/core/backend/options.go
@@ -11,7 +11,7 @@ import (
 	"github.com/rs/zerolog/log"
 )

-func ModelOptions(c config.BackendConfig, so *config.ApplicationConfig, opts ...model.Option) []model.Option {
+func ModelOptions(c config.BackendConfig, so *config.ApplicationConfig, opts []model.Option) []model.Option {
 	name := c.Name
 	if name == "" {
 		name = c.Model
@@ -122,17 +122,14 @@ func grpcModelOpts(c config.BackendConfig) *pb.ModelOptions {
 		CUDA:                 c.CUDA || c.Diffusers.CUDA,
 		SchedulerType:        c.Diffusers.SchedulerType,
 		PipelineType:         c.Diffusers.PipelineType,
-		CFGScale:             c.CFGScale,
+		CFGScale:             c.Diffusers.CFGScale,
 		LoraAdapter:          c.LoraAdapter,
 		LoraScale:            c.LoraScale,
-		LoraAdapters:         c.LoraAdapters,
-		LoraScales:           c.LoraScales,
 		F16Memory:            f16,
 		LoraBase:             c.LoraBase,
 		IMG2IMG:              c.Diffusers.IMG2IMG,
 		CLIPModel:            c.Diffusers.ClipModel,
 		CLIPSubfolder:        c.Diffusers.ClipSubFolder,
-		Options:              c.Options,
 		CLIPSkip:             int32(c.Diffusers.ClipSkip),
 		ControlNet:           c.Diffusers.ControlNet,
 		ContextSize:          int32(ctxSize),
@@ -151,8 +148,6 @@ func grpcModelOpts(c config.BackendConfig) *pb.ModelOptions {
 		TensorParallelSize:   int32(c.TensorParallelSize),
 		MMProj:               c.MMProj,
 		FlashAttention:       c.FlashAttention,
-		CacheTypeKey:         c.CacheTypeK,
-		CacheTypeValue:       c.CacheTypeV,
 		NoKVOffload:          c.NoKVOffloading,
 		YarnExtFactor:        c.YarnExtFactor,
 		YarnAttnFactor:       c.YarnAttnFactor,
--- a/core/backend/rerank.go
+++ b/core/backend/rerank.go
@@ -11,8 +11,8 @@ import (

 func Rerank(modelFile string, request *proto.RerankRequest, loader *model.ModelLoader, appConfig *config.ApplicationConfig, backendConfig config.BackendConfig) (*proto.RerankResult, error) {

-	opts := ModelOptions(backendConfig, appConfig, model.WithModel(modelFile))
-	rerankModel, err := loader.Load(opts...)
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{model.WithModel(modelFile)})
+	rerankModel, err := loader.BackendLoader(opts...)
 	if err != nil {
 		return nil, err
 	}
--- a/core/backend/soundgeneration.go
+++ b/core/backend/soundgeneration.go
@@ -25,8 +25,9 @@ func SoundGeneration(
 	backendConfig config.BackendConfig,
 ) (string, *proto.Result, error) {

-	opts := ModelOptions(backendConfig, appConfig, model.WithModel(modelFile))
-	soundGenModel, err := loader.Load(opts...)
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{model.WithModel(modelFile)})
+
+	soundGenModel, err := loader.BackendLoader(opts...)
 	if err != nil {
 		return "", nil, err
 	}
--- a/core/backend/stores.go
+++ b/core/backend/stores.go
@@ -8,15 +8,16 @@ import (
 )

 func StoreBackend(sl *model.ModelLoader, appConfig *config.ApplicationConfig, storeName string) (grpc.Backend, error) {
-	if storeName == "" {
-		storeName = "default"
-	}
+    if storeName == "" {
+      storeName = "default"
+    }

-	sc := []model.Option{
-		model.WithBackendString(model.LocalStoreBackend),
-		model.WithAssetDir(appConfig.AssetsDestination),
-		model.WithModel(storeName),
-	}
+    sc := []model.Option{
+      model.WithBackendString(model.LocalStoreBackend),
+      model.WithAssetDir(appConfig.AssetsDestination),
+      model.WithModel(storeName),
+    }

-	return sl.Load(sc...)
+    return sl.BackendLoader(sc...)
 }
+
--- a/core/backend/token_metrics.go
+++ b/core/backend/token_metrics.go
@@ -15,8 +15,10 @@ func TokenMetrics(
 	appConfig *config.ApplicationConfig,
 	backendConfig config.BackendConfig) (*proto.MetricsResponse, error) {

-	opts := ModelOptions(backendConfig, appConfig, model.WithModel(modelFile))
-	model, err := loader.Load(opts...)
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{
+		model.WithModel(modelFile),
+	})
+	model, err := loader.BackendLoader(opts...)
 	if err != nil {
 		return nil, err
 	}
--- a/core/backend/tokenize.go
+++ b/core/backend/tokenize.go
@@ -14,13 +14,15 @@ func ModelTokenize(s string, loader *model.ModelLoader, backendConfig config.Bac
 	var inferenceModel grpc.Backend
 	var err error

-	opts := ModelOptions(backendConfig, appConfig, model.WithModel(modelFile))
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{
+		model.WithModel(modelFile),
+	})

 	if backendConfig.Backend == "" {
-		inferenceModel, err = loader.Load(opts...)
+		inferenceModel, err = loader.GreedyLoader(opts...)
 	} else {
 		opts = append(opts, model.WithBackendString(backendConfig.Backend))
-		inferenceModel, err = loader.Load(opts...)
+		inferenceModel, err = loader.BackendLoader(opts...)
 	}
 	if err != nil {
 		return schema.TokenizeResponse{}, err
--- a/core/backend/transcript.go
+++ b/core/backend/transcript.go
@@ -18,9 +18,9 @@ func ModelTranscription(audio, language string, translate bool, ml *model.ModelL
 		backendConfig.Backend = model.WhisperBackend
 	}

-	opts := ModelOptions(backendConfig, appConfig)
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{})

-	transcriptionModel, err := ml.Load(opts...)
+	transcriptionModel, err := ml.BackendLoader(opts...)
 	if err != nil {
 		return nil, err
 	}
--- a/core/backend/tts.go
+++ b/core/backend/tts.go
@@ -28,8 +28,11 @@ func ModelTTS(
 		bb = model.PiperBackend
 	}

-	opts := ModelOptions(backendConfig, appConfig, model.WithBackendString(bb), model.WithModel(modelFile))
-	ttsModel, err := loader.Load(opts...)
+	opts := ModelOptions(config.BackendConfig{}, appConfig, []model.Option{
+		model.WithBackendString(bb),
+		model.WithModel(modelFile),
+	})
+	ttsModel, err := loader.BackendLoader(opts...)
 	if err != nil {
 		return "", nil, err
 	}
--- a/core/cli/run.go
+++ b/core/cli/run.go
@@ -6,12 +6,12 @@ import (
 	"strings"
 	"time"

-	"github.com/mudler/LocalAI/core/application"
 	cli_api "github.com/mudler/LocalAI/core/cli/api"
 	cliContext "github.com/mudler/LocalAI/core/cli/context"
 	"github.com/mudler/LocalAI/core/config"
 	"github.com/mudler/LocalAI/core/http"
 	"github.com/mudler/LocalAI/core/p2p"
+	"github.com/mudler/LocalAI/core/startup"
 	"github.com/rs/zerolog"
 	"github.com/rs/zerolog/log"
 )
@@ -186,16 +186,16 @@ func (r *RunCMD) Run(ctx *cliContext.Context) error {
 	}

 	if r.PreloadBackendOnly {
-		_, err := application.New(opts...)
+		_, _, _, err := startup.Startup(opts...)
 		return err
 	}

-	app, err := application.New(opts...)
+	cl, ml, options, err := startup.Startup(opts...)
 	if err != nil {
 		return fmt.Errorf("failed basic startup tasks with error %s", err.Error())
 	}

-	appHTTP, err := http.API(app)
+	appHTTP, err := http.App(cl, ml, options)
 	if err != nil {
 		log.Error().Err(err).Msg("error during HTTP App construction")
 		return err
--- a/core/cli/worker/worker_p2p.go
+++ b/core/cli/worker/worker_p2p.go
@@ -76,14 +76,8 @@ func (r *P2P) Run(ctx *cliContext.Context) error {
 					"util",
 					"llama-cpp-rpc-server",
 				)
-				var extraArgs []string
-
-				if r.ExtraLLamaCPPArgs != "" {
-					extraArgs = strings.Split(r.ExtraLLamaCPPArgs, " ")
-				}
+				extraArgs := strings.Split(r.ExtraLLamaCPPArgs, " ")
 				args := append([]string{"--host", address, "--port", fmt.Sprint(port)}, extraArgs...)
-				log.Debug().Msgf("Starting llama-cpp-rpc-server on '%s:%d' with args: %+v (%d)", address, port, args, len(args))
-
 				args, grpcProcess = library.LoadLDSO(r.BackendAssetsPath, args, grpcProcess)

 				cmd := exec.Command(
--- a/core/config/backend_config.go
+++ b/core/config/backend_config.go
@@ -38,7 +38,6 @@ type BackendConfig struct {
 	TemplateConfig      TemplateConfig         `yaml:"template"`
 	KnownUsecaseStrings []string               `yaml:"known_usecases"`
 	KnownUsecases       *BackendConfigUsecases `yaml:"-"`
-	Pipeline            Pipeline               `yaml:"pipeline"`

 	PromptStrings, InputStrings                []string               `yaml:"-"`
 	InputToken                                 [][]int                `yaml:"-"`
@@ -73,20 +72,6 @@ type BackendConfig struct {

 	Description string `yaml:"description"`
 	Usage       string `yaml:"usage"`
-
-	Options []string `yaml:"options"`
-}
-
-// Pipeline defines other models to use for audio-to-audio
-type Pipeline struct {
-	TTS           string `yaml:"tts"`
-	LLM           string `yaml:"llm"`
-	Transcription string `yaml:"transcription"`
-	VAD           string `yaml:"vad"`
-}
-
-func (p Pipeline) IsNotConfigured() bool {
-	return p.LLM == "" || p.TTS == "" || p.Transcription == ""
 }

 type File struct {
@@ -112,15 +97,16 @@ type GRPC struct {
 }

 type Diffusers struct {
-	CUDA             bool   `yaml:"cuda"`
-	PipelineType     string `yaml:"pipeline_type"`
-	SchedulerType    string `yaml:"scheduler_type"`
-	EnableParameters string `yaml:"enable_parameters"` // A list of comma separated parameters to specify
-	IMG2IMG          bool   `yaml:"img2img"`           // Image to Image Diffuser
-	ClipSkip         int    `yaml:"clip_skip"`         // Skip every N frames
-	ClipModel        string `yaml:"clip_model"`        // Clip model to use
-	ClipSubFolder    string `yaml:"clip_subfolder"`    // Subfolder to use for clip model
-	ControlNet       string `yaml:"control_net"`
+	CUDA             bool    `yaml:"cuda"`
+	PipelineType     string  `yaml:"pipeline_type"`
+	SchedulerType    string  `yaml:"scheduler_type"`
+	EnableParameters string  `yaml:"enable_parameters"` // A list of comma separated parameters to specify
+	CFGScale         float32 `yaml:"cfg_scale"`         // Classifier-Free Guidance Scale
+	IMG2IMG          bool    `yaml:"img2img"`           // Image to Image Diffuser
+	ClipSkip         int     `yaml:"clip_skip"`         // Skip every N frames
+	ClipModel        string  `yaml:"clip_model"`        // Clip model to use
+	ClipSubFolder    string  `yaml:"clip_subfolder"`    // Subfolder to use for clip model
+	ControlNet       string  `yaml:"control_net"`
 }

 // LLMConfig is a struct that holds the configuration that are
@@ -148,30 +134,26 @@ type LLMConfig struct {
 	TrimSpace       []string `yaml:"trimspace"`
 	TrimSuffix      []string `yaml:"trimsuffix"`

-	ContextSize          *int      `yaml:"context_size"`
-	NUMA                 bool      `yaml:"numa"`
-	LoraAdapter          string    `yaml:"lora_adapter"`
-	LoraBase             string    `yaml:"lora_base"`
-	LoraAdapters         []string  `yaml:"lora_adapters"`
-	LoraScales           []float32 `yaml:"lora_scales"`
-	LoraScale            float32   `yaml:"lora_scale"`
-	NoMulMatQ            bool      `yaml:"no_mulmatq"`
-	DraftModel           string    `yaml:"draft_model"`
-	NDraft               int32     `yaml:"n_draft"`
-	Quantization         string    `yaml:"quantization"`
-	LoadFormat           string    `yaml:"load_format"`
-	GPUMemoryUtilization float32   `yaml:"gpu_memory_utilization"` // vLLM
-	TrustRemoteCode      bool      `yaml:"trust_remote_code"`      // vLLM
-	EnforceEager         bool      `yaml:"enforce_eager"`          // vLLM
-	SwapSpace            int       `yaml:"swap_space"`             // vLLM
-	MaxModelLen          int       `yaml:"max_model_len"`          // vLLM
-	TensorParallelSize   int       `yaml:"tensor_parallel_size"`   // vLLM
-	MMProj               string    `yaml:"mmproj"`
+	ContextSize          *int    `yaml:"context_size"`
+	NUMA                 bool    `yaml:"numa"`
+	LoraAdapter          string  `yaml:"lora_adapter"`
+	LoraBase             string  `yaml:"lora_base"`
+	LoraScale            float32 `yaml:"lora_scale"`
+	NoMulMatQ            bool    `yaml:"no_mulmatq"`
+	DraftModel           string  `yaml:"draft_model"`
+	NDraft               int32   `yaml:"n_draft"`
+	Quantization         string  `yaml:"quantization"`
+	LoadFormat           string  `yaml:"load_format"`
+	GPUMemoryUtilization float32 `yaml:"gpu_memory_utilization"` // vLLM
+	TrustRemoteCode      bool    `yaml:"trust_remote_code"`      // vLLM
+	EnforceEager         bool    `yaml:"enforce_eager"`          // vLLM
+	SwapSpace            int     `yaml:"swap_space"`             // vLLM
+	MaxModelLen          int     `yaml:"max_model_len"`          // vLLM
+	TensorParallelSize   int     `yaml:"tensor_parallel_size"`   // vLLM
+	MMProj               string  `yaml:"mmproj"`

-	FlashAttention bool   `yaml:"flash_attention"`
-	NoKVOffloading bool   `yaml:"no_kv_offloading"`
-	CacheTypeK     string `yaml:"cache_type_k"`
-	CacheTypeV     string `yaml:"cache_type_v"`
+	FlashAttention bool `yaml:"flash_attention"`
+	NoKVOffloading bool `yaml:"no_kv_offloading"`

 	RopeScaling string `yaml:"rope_scaling"`
 	ModelType   string `yaml:"type"`
@@ -180,8 +162,6 @@ type LLMConfig struct {
 	YarnAttnFactor float32 `yaml:"yarn_attn_factor"`
 	YarnBetaFast   float32 `yaml:"yarn_beta_fast"`
 	YarnBetaSlow   float32 `yaml:"yarn_beta_slow"`
-
-	CFGScale float32 `yaml:"cfg_scale"` // Classifier-Free Guidance Scale
 }

 // AutoGPTQ is a struct that holds the configuration specific to the AutoGPTQ backend
@@ -219,8 +199,6 @@ type TemplateConfig struct {
 	JoinChatMessagesByCharacter *string `yaml:"join_chat_messages_by_character"`

 	Multimodal string `yaml:"multimodal"`
-
-	JinjaTemplate bool `yaml:"jinja_template"`
 }

 func (c *BackendConfig) UnmarshalYAML(value *yaml.Node) error {
--- a/core/config/backend_config_loader.go
+++ b/core/config/backend_config_loader.go
@@ -140,7 +140,7 @@ func (bcl *BackendConfigLoader) LoadBackendConfigFileByName(modelName, modelPath
 		}
 	}

-	cfg.SetDefaults(append(opts, ModelPath(modelPath))...)
+	cfg.SetDefaults(opts...)

 	return cfg, nil
 }
--- a/Show More
+++ b/Show More