fix(go-grpc-server): always close resultChan

By not closing the channel, if a server not implementing PredictStream receives a client call would hang indefinetly as would wait for resultChan to be consumed. If the prediction stream returns we close the channel now and we wait for the goroutine to finish. Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
2026-02-05 04:02:45 -05:00 · 2024-10-05 00:07:58 +02:00
340 changed files with 10241 additions and 5189 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/.gitattributes
+++ b/.gitattributes
@@ -1,2 +1 @@
 *.sh text eol=lf
-backend/cpp/llama/*.hpp linguist-vendored
--- a/.github/ci/modelslist.go
+++ b/.github/ci/modelslist.go
@@ -6,7 +6,6 @@ import (
 	"io/ioutil"
 	"os"

-	"github.com/microcosm-cc/bluemonday"
 	"gopkg.in/yaml.v3"
 )

@@ -280,12 +279,6 @@ func main() {
 		return
 	}

-	// Ensure that all arbitrary text content is sanitized before display
-	for i, m := range models {
-		models[i].Name = bluemonday.StrictPolicy().Sanitize(m.Name)
-		models[i].Description = bluemonday.StrictPolicy().Sanitize(m.Description)
-	}
-
 	// render the template
 	data := struct {
 		Models          []*GalleryModel
--- a/.github/dependabot.yml
+++ b/.github/dependabot.yml
@@ -9,8 +9,6 @@ updates:
    directory: "/"
    schedule:
      interval: "weekly"
-    ignore:
-    - dependency-name: "github.com/mudler/LocalAI/pkg/grpc/proto"
  - package-ecosystem: "github-actions"
    # Workflow files stored in the default location of `.github/workflows`. (You don't need to specify `/.github/workflows` for `directory`. You can use `directory: "/"`.)
    directory: "/"
--- 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@v1.0.3
        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@v1.0.3
        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/notify-models.yaml
+++ b/.github/workflows/notify-models.yaml
@@ -79,7 +79,7 @@ jobs:
        args: ${{ steps.summarize.outputs.message }}
    - name: Setup tmate session if fails
      if: ${{ failure() }}
-      uses: mxschmitt/action-tmate@v3.19
+      uses: mxschmitt/action-tmate@v3.18
      with:
        detached: true
        connect-timeout-seconds: 180
@@ -161,7 +161,7 @@ jobs:
        TWITTER_ACCESS_TOKEN_SECRET: ${{ secrets.TWITTER_ACCESS_TOKEN_SECRET }}
    - name: Setup tmate session if fails
      if: ${{ failure() }}
-      uses: mxschmitt/action-tmate@v3.19
+      uses: mxschmitt/action-tmate@v3.18
      with:
        detached: true
        connect-timeout-seconds: 180
--- a/.github/workflows/release.yaml
+++ b/.github/workflows/release.yaml
@@ -123,7 +123,7 @@ jobs:
            release/*
      - name: Setup tmate session if tests fail
        if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.19
+        uses: mxschmitt/action-tmate@v3.18
        with:
          detached: true
          connect-timeout-seconds: 180
@@ -232,7 +232,7 @@ jobs:
            release/*
      - name: Setup tmate session if tests fail
        if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.19
+        uses: mxschmitt/action-tmate@v3.18
        with:
          detached: true
          connect-timeout-seconds: 180
@@ -308,7 +308,7 @@ jobs:
            release/*
      - name: Setup tmate session if tests fail
        if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.19
+        uses: mxschmitt/action-tmate@v3.18
        with:
          detached: true
          connect-timeout-seconds: 180
@@ -350,7 +350,7 @@ jobs:
            release/*
      - name: Setup tmate session if tests fail
        if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.19
+        uses: mxschmitt/action-tmate@v3.18
        with:
          detached: true
          connect-timeout-seconds: 180
--- a/.github/workflows/test-extra.yml
+++ b/.github/workflows/test-extra.yml
@@ -123,13 +123,6 @@ 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
@@ -133,7 +133,7 @@ jobs:
          PATH="$PATH:/root/go/bin" GO_TAGS="stablediffusion tts" make --jobs 5 --output-sync=target test
      - name: Setup tmate session if tests fail
        if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.19
+        uses: mxschmitt/action-tmate@v3.18
        with:
          detached: true
          connect-timeout-seconds: 180
@@ -197,7 +197,7 @@ jobs:
            make run-e2e-aio
      - name: Setup tmate session if tests fail
        if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.19
+        uses: mxschmitt/action-tmate@v3.18
        with:
          detached: true
          connect-timeout-seconds: 180
@@ -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
@@ -235,7 +235,7 @@ jobs:
          BUILD_TYPE="GITHUB_CI_HAS_BROKEN_METAL" CMAKE_ARGS="-DGGML_F16C=OFF -DGGML_AVX512=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF" make --jobs 4 --output-sync=target test
      - name: Setup tmate session if tests fail
        if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.19
+        uses: mxschmitt/action-tmate@v3.18
        with:
          detached: true
          connect-timeout-seconds: 180
--- a/.gitignore
+++ b/.gitignore
@@ -12,6 +12,7 @@ prepare-sources

 go-ggml-transformers
 go-gpt2
+go-rwkv
 whisper.cpp
 /bloomz
 go-bert
--- a/39
+++ b/39
@@ -9,8 +9,6 @@ FROM ${BASE_IMAGE} AS requirements-core
 USER root

 ARG GO_VERSION=1.22.6
-ARG CMAKE_VERSION=3.26.4
-ARG CMAKE_FROM_SOURCE=false
 ARG TARGETARCH
 ARG TARGETVARIANT

@@ -23,25 +21,13 @@ RUN apt-get update && \
        build-essential \
        ccache \
        ca-certificates \
-        curl libssl-dev \
+        cmake \
+        curl \
        git \
        unzip upx-ucl && \
    apt-get clean && \
    rm -rf /var/lib/apt/lists/*

-# Install CMake (the version in 22.04 is too old)
-RUN <<EOT bash
-    if [ "${CMAKE_FROM_SOURCE}}" = "true" ]; then
-        curl -L -s https://github.com/Kitware/CMake/releases/download/v${CMAKE_VERSION}/cmake-${CMAKE_VERSION}.tar.gz -o cmake.tar.gz && tar xvf cmake.tar.gz && cd cmake-${CMAKE_VERSION} && ./configure && make && make install
-    else
-        apt-get update && \
-        apt-get install -y \
-            cmake && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/*
-    fi
-EOT
-
 # Install Go
 RUN curl -L -s https://go.dev/dl/go${GO_VERSION}.linux-${TARGETARCH}.tar.gz | tar -C /usr/local -xz
 ENV PATH=$PATH:/root/go/bin:/usr/local/go/bin
@@ -85,8 +71,7 @@ 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
@@ -203,8 +188,6 @@ FROM ${GRPC_BASE_IMAGE} AS grpc
 # This is a bit of a hack, but it's required in order to be able to effectively cache this layer in CI
 ARG GRPC_MAKEFLAGS="-j4 -Otarget"
 ARG GRPC_VERSION=v1.65.0
-ARG CMAKE_FROM_SOURCE=false
-ARG CMAKE_VERSION=3.26.4

 ENV MAKEFLAGS=${GRPC_MAKEFLAGS}

@@ -213,24 +196,12 @@ WORKDIR /build
 RUN apt-get update && \
    apt-get install -y --no-install-recommends \
        ca-certificates \
-        build-essential curl libssl-dev \
+        build-essential \
+        cmake \
        git && \
    apt-get clean && \
    rm -rf /var/lib/apt/lists/*

-# Install CMake (the version in 22.04 is too old)
-RUN <<EOT bash
-    if [ "${CMAKE_FROM_SOURCE}}" = "true" ]; then
-        curl -L -s https://github.com/Kitware/CMake/releases/download/v${CMAKE_VERSION}/cmake-${CMAKE_VERSION}.tar.gz -o cmake.tar.gz && tar xvf cmake.tar.gz && cd cmake-${CMAKE_VERSION} && ./configure && make && make install
-    else
-        apt-get update && \
-        apt-get install -y \
-            cmake && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/*
-    fi
-EOT
-
 # We install GRPC to a different prefix here so that we can copy in only the build artifacts later
 # saves several hundred MB on the final docker image size vs copying in the entire GRPC source tree
 # and running make install in the target container
--- a/110
+++ b/110
@@ -8,11 +8,15 @@ DETECT_LIBS?=true
 # llama.cpp versions
 GOLLAMA_REPO?=https://github.com/go-skynet/go-llama.cpp
 GOLLAMA_VERSION?=2b57a8ae43e4699d3dc5d1496a1ccd42922993be
-CPPLLAMA_VERSION?=47f931c8f9a26c072d71224bc8013cc66ea9e445
+CPPLLAMA_VERSION?=d5ed2b929d85bbd7dbeecb690880f07d9d7a6077
+
+# 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?=ccc2547210e09e3a1785817383ab770389bb442b

 # bert.cpp version
 BERT_REPO?=https://github.com/go-skynet/go-bert.cpp
@@ -20,7 +24,7 @@ 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
@@ -30,10 +34,6 @@ STABLEDIFFUSION_VERSION?=4a3cd6aeae6f66ee57eae9a0075f8c58c3a6a38f
 TINYDREAM_REPO?=https://github.com/M0Rf30/go-tiny-dream
 TINYDREAM_VERSION?=c04fa463ace9d9a6464313aa5f9cd0f953b6c057

-ONNX_VERSION?=1.20.0
-ONNX_ARCH?=x64
-ONNX_OS?=linux
-
 export BUILD_TYPE?=
 export STABLE_BUILD_TYPE?=$(BUILD_TYPE)
 export CMAKE_ARGS?=
@@ -45,7 +45,6 @@ CGO_LDFLAGS_WHISPER+=-lggml
 CUDA_LIBPATH?=/usr/local/cuda/lib64/
 GO_TAGS?=
 BUILD_ID?=
-NATIVE?=false

 TEST_DIR=/tmp/test

@@ -84,25 +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 := $(shell security find-identity -v -p codesigning | grep '"' | head -n 1 | sed -E 's/.*"(.*)"/\1/')
@@ -157,10 +138,10 @@ ifeq ($(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

@@ -205,9 +186,9 @@ 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
 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))
@@ -267,6 +248,20 @@ sources/go-piper:
 sources/go-piper/libpiper_binding.a: sources/go-piper
 	$(MAKE) -C sources/go-piper libpiper_binding.a example/main piper.o

+
+## 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 $(RWKV_VERSION) && \
+	git submodule update --init --recursive --depth 1 --single-branch
+
+sources/go-rwkv.cpp/librwkv.a: sources/go-rwkv.cpp
+	cd sources/go-rwkv.cpp && cd rwkv.cpp &&	cmake . -DRWKV_BUILD_SHARED_LIBRARY=OFF &&	cmake --build . && 	cp librwkv.a ..
+
 ## stable diffusion
 sources/go-stable-diffusion:
 	mkdir -p sources/go-stable-diffusion
@@ -280,20 +275,6 @@ sources/go-stable-diffusion:
 sources/go-stable-diffusion/libstablediffusion.a: sources/go-stable-diffusion
 	CPATH="$(CPATH):/usr/include/opencv4" $(MAKE) -C sources/go-stable-diffusion libstablediffusion.a

-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
@@ -320,9 +301,10 @@ 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/whisper.cpp sources/go-bert.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
@@ -332,6 +314,7 @@ replace:
 	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(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
@@ -347,6 +330,7 @@ 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
@@ -455,6 +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
@@ -484,13 +470,13 @@ run-e2e-image:

 run-e2e-aio: protogen-go
 	@echo 'Running e2e AIO tests'
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts $(TEST_FLAKES) -v -r ./tests/e2e-aio
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts 5 -v -r ./tests/e2e-aio

 test-e2e:
 	@echo 'Running e2e tests'
 	BUILD_TYPE=$(BUILD_TYPE) \
 	LOCALAI_API=http://$(E2E_BRIDGE_IP):5390/v1 \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts $(TEST_FLAKES) -v -r ./tests/e2e
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts 5 -v -r ./tests/e2e

 teardown-e2e:
 	rm -rf $(TEST_DIR) || true
@@ -498,24 +484,24 @@ teardown-e2e:

 test-llama: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama" --flake-attempts 5 -v -r $(TEST_PATHS)

 test-llama-gguf: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama-gguf" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama-gguf" --flake-attempts 5 -v -r $(TEST_PATHS)

 test-tts: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="tts" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="tts" --flake-attempts 1 -v -r $(TEST_PATHS)

 test-stablediffusion: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stablediffusion" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stablediffusion" --flake-attempts 1 -v -r $(TEST_PATHS)

 test-stores: backend-assets/grpc/local-store
 	mkdir -p tests/integration/backend-assets/grpc
 	cp -f backend-assets/grpc/local-store tests/integration/backend-assets/grpc/
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stores" --flake-attempts $(TEST_FLAKES) -v -r tests/integration
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stores" --flake-attempts 1 -v -r tests/integration

 test-container:
 	docker build --target requirements -t local-ai-test-container .
@@ -775,7 +761,7 @@ backend-assets/grpc/llama-cpp-fallback: backend-assets/grpc backend/cpp/llama/ll
 	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/ggml-metal.metal backend-assets/grpc/
+	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
@@ -789,7 +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
@@ -831,6 +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 "$(GO_TAGS)" -o backend-assets/grpc/stablediffusion ./backend/go/image/stablediffusion
@@ -838,13 +831,6 @@ ifneq ($(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
@@ -905,7 +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)" \
@@ -913,7 +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)

@@ -60,40 +56,20 @@ 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
-```
-
-To load models:
-
-```bash
-# From the model gallery (see available models with `local-ai models list`, in the WebUI from the model tab, or visiting https://models.localai.io)
-local-ai run llama-3.2-1b-instruct:q4_k_m
-# Start LocalAI with the phi-2 model directly from huggingface
-local-ai run huggingface://TheBloke/phi-2-GGUF/phi-2.Q8_0.gguf
-# Install and run a model from the Ollama OCI registry
-local-ai run ollama://gemma:2b
-# Run a model from a configuration file
-local-ai run https://gist.githubusercontent.com/.../phi-2.yaml
-# Install and run a model from a standard OCI registry (e.g., Docker Hub)
-local-ai run oci://localai/phi-2:latest
+# 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 
 ```

 [💻 Getting started](https://localai.io/basics/getting_started/index.html)

 ## 📰 Latest project news

- 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
@@ -164,9 +140,6 @@ 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/
  
@@ -241,6 +214,7 @@ 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/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
@@ -221,7 +219,6 @@ message ModelOptions {
  int32  SwapSpace = 53;
  int32  MaxModelLen = 54;
  int32  TensorParallelSize = 55;
-  string LoadFormat = 58;

  string MMProj = 41;

@@ -235,11 +232,6 @@ message ModelOptions {

  bool FlashAttention = 56;
  bool NoKVOffload = 57;
-
-  string ModelPath = 59;
-
-  repeated string LoraAdapters = 60;
-  repeated float LoraScales = 61;
 }

 message Result {
@@ -295,19 +287,6 @@ 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)
--- a/backend/cpp/llama/grpc-server.cpp
+++ b/backend/cpp/llama/grpc-server.cpp
@@ -113,7 +113,7 @@ static std::string tokens_to_str(llama_context *ctx, Iter begin, Iter end)
    std::string ret;
    for (; begin != end; ++begin)
    {
-        ret += common_token_to_piece(ctx, *begin);
+        ret += llama_token_to_piece(ctx, *begin);
    }
    return ret;
 }
@@ -121,7 +121,7 @@ static std::string tokens_to_str(llama_context *ctx, Iter begin, Iter end)
 // format incomplete utf-8 multibyte character for output
 static std::string tokens_to_output_formatted_string(const llama_context *ctx, const llama_token token)
 {
-    std::string out = token == -1 ? "" : common_token_to_piece(ctx, token);
+    std::string out = token == -1 ? "" : llama_token_to_piece(ctx, token);
    // if the size is 1 and first bit is 1, meaning it's a partial character
    //   (size > 1 meaning it's already a known token)
    if (out.size() == 1 && (out[0] & 0x80) == 0x80)
@@ -203,8 +203,8 @@ struct llama_client_slot
    std::string stopping_word;

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

    int32_t ga_i = 0;   // group-attention state
    int32_t ga_n = 1;   // group-attention factor
@@ -257,7 +257,7 @@ struct llama_client_slot
        images.clear();
    }

-    bool has_budget(common_params &global_params) {
+    bool has_budget(gpt_params &global_params) {
        if (params.n_predict == -1 && global_params.n_predict == -1)
        {
            return true; // limitless
@@ -391,39 +391,6 @@ struct llama_metrics {
    }
 };

-struct llava_embd_batch {
-    std::vector<llama_pos>      pos;
-    std::vector<int32_t>        n_seq_id;
-    std::vector<llama_seq_id>   seq_id_0;
-    std::vector<llama_seq_id *> seq_ids;
-    std::vector<int8_t>         logits;
-    llama_batch batch;
-    llava_embd_batch(float * embd, int32_t n_tokens, llama_pos pos_0, llama_seq_id seq_id) {
-        pos     .resize(n_tokens);
-        n_seq_id.resize(n_tokens);
-        seq_ids .resize(n_tokens + 1);
-        logits  .resize(n_tokens);
-        seq_id_0.resize(1);
-        seq_id_0[0] = seq_id;
-        seq_ids [n_tokens] = nullptr;
-        batch = {
-            /*n_tokens       =*/ n_tokens,
-            /*tokens         =*/ nullptr,
-            /*embd           =*/ embd,
-            /*pos            =*/ pos.data(),
-            /*n_seq_id       =*/ n_seq_id.data(),
-            /*seq_id         =*/ seq_ids.data(),
-            /*logits         =*/ logits.data(),
-        };
-        for (int i = 0; i < n_tokens; i++) {
-            batch.pos     [i] = pos_0 + i;
-            batch.n_seq_id[i] = 1;
-            batch.seq_id  [i] = seq_id_0.data();
-            batch.logits  [i] = false;
-        }
-    }
-};
-
 struct llama_server_context
 {
    llama_model *model = nullptr;
@@ -431,7 +398,7 @@ struct llama_server_context

    clip_ctx *clp_ctx = nullptr;

-    common_params params;
+    gpt_params params;

    llama_batch batch;

@@ -474,7 +441,7 @@ struct llama_server_context
        }
    }

-    bool load_model(const common_params &params_)
+    bool load_model(const gpt_params &params_)
    {
        params = params_;
        if (!params.mmproj.empty()) {
@@ -491,9 +458,9 @@ struct llama_server_context
            }
        }

-        common_init_result common_init = common_init_from_params(params);
-        model = common_init.model;
-        ctx = common_init.context;
+        llama_init_result llama_init = llama_init_from_gpt_params(params);
+        model = llama_init.model;
+        ctx = llama_init.context;
        if (model == nullptr)
        {
            LOG_ERR("unable to load model: %s", params.model.c_str());
@@ -611,12 +578,12 @@ struct llama_server_context
                    std::vector<llama_token> p;
                    if (first)
                    {
-                        p = common_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
+                        p = ::llama_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
                        first = false;
                    }
                    else
                    {
-                        p = common_tokenize(ctx, s, false, TMP_FORCE_SPECIAL);
+                        p = ::llama_tokenize(ctx, s, false, TMP_FORCE_SPECIAL);
                    }
                    prompt_tokens.insert(prompt_tokens.end(), p.begin(), p.end());
                }
@@ -633,7 +600,7 @@ struct llama_server_context
        else
        {
            auto s = json_prompt.template get<std::string>();
-            prompt_tokens = common_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
+            prompt_tokens = ::llama_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
        }

        return prompt_tokens;
@@ -662,7 +629,7 @@ struct llama_server_context

    bool launch_slot_with_data(llama_client_slot* &slot, json data) {
        slot_params default_params;
-        common_params_sampling default_sparams;
+        gpt_sampler_params default_sparams;
 
        slot->params.stream             = json_value(data, "stream",            false);
        slot->params.cache_prompt       = json_value(data, "cache_prompt",      false);
@@ -670,6 +637,7 @@ 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);
@@ -801,7 +769,7 @@ struct llama_server_context
                    }
                    else if (el[0].is_string())
                    {
-                        auto toks = common_tokenize(model, el[0].get<std::string>(), false);
+                        auto toks = llama_tokenize(model, el[0].get<std::string>(), false);
                        for (auto tok : toks)
                        {
                            slot->sparams.logit_bias.push_back({tok, bias});
@@ -833,7 +801,7 @@ struct llama_server_context
                        sampler_names.emplace_back(name);
                    }
                }
-                slot->sparams.samplers = common_sampler_types_from_names(sampler_names, false);
+                slot->sparams.samplers = gpt_sampler_types_from_names(sampler_names, false);
        }
        else
        {
@@ -917,9 +885,9 @@ struct llama_server_context

        if (slot->ctx_sampling != nullptr)
        {
-            common_sampler_free(slot->ctx_sampling);
+            gpt_sampler_free(slot->ctx_sampling);
        }
-        slot->ctx_sampling = common_sampler_init(model, slot->sparams);
+        slot->ctx_sampling = gpt_sampler_init(model, slot->sparams);
        //llama_set_rng_seed(ctx, slot->params.seed);
        slot->command = LOAD_PROMPT;

@@ -946,13 +914,13 @@ struct llama_server_context
        system_tokens.clear();

        if (!system_prompt.empty()) {
-            system_tokens = common_tokenize(ctx, system_prompt, add_bos_token);
+            system_tokens = ::llama_tokenize(ctx, system_prompt, add_bos_token);

-            common_batch_clear(batch);
+            llama_batch_clear(batch);

            for (int i = 0; i < (int)system_tokens.size(); ++i)
            {
-                common_batch_add(batch, system_tokens[i], i, { 0 }, false);
+                llama_batch_add(batch, system_tokens[i], i, { 0 }, false);
            }

            for (int32_t i = 0; i < (int32_t) batch.n_tokens; i += params.n_batch)
@@ -966,6 +934,7 @@ struct llama_server_context
                    batch.n_seq_id + i,
                    batch.seq_id   + i,
                    batch.logits   + i,
+                    0, 0, 0, // unused
                };
                if (llama_decode(ctx, batch_view) != 0)
                {
@@ -1040,7 +1009,7 @@ struct llama_server_context

    bool process_token(completion_token_output &result, llama_client_slot &slot) {
        // remember which tokens were sampled - used for repetition penalties during sampling
-        const std::string token_str = common_token_to_piece(ctx, result.tok);
+        const std::string token_str = llama_token_to_piece(ctx, result.tok);
        slot.sampled = result.tok;

        // search stop word and delete it
@@ -1191,7 +1160,7 @@ struct llama_server_context
        samplers.reserve(slot.sparams.samplers.size());
        for (const auto & sampler : slot.sparams.samplers)
        {
-            samplers.emplace_back(common_sampler_type_to_str(sampler));
+            samplers.emplace_back(gpt_sampler_type_to_str(sampler));
        }

        return json {
@@ -1205,6 +1174,7 @@ 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},
@@ -1246,7 +1216,7 @@ struct llama_server_context
        if (slot.sparams.n_probs > 0)
        {
            std::vector<completion_token_output> probs_output = {};
-            const std::vector<llama_token> to_send_toks = common_tokenize(ctx, tkn.text_to_send, false);
+            const std::vector<llama_token> to_send_toks = llama_tokenize(ctx, tkn.text_to_send, false);
            size_t probs_pos      = std::min(slot.sent_token_probs_index,                       slot.generated_token_probs.size());
            size_t probs_stop_pos = std::min(slot.sent_token_probs_index + to_send_toks.size(), slot.generated_token_probs.size());
            if (probs_pos < probs_stop_pos)
@@ -1298,7 +1268,7 @@ struct llama_server_context
            std::vector<completion_token_output> probs = {};
            if (!slot.params.stream && slot.stopped_word)
            {
-                const std::vector<llama_token> stop_word_toks = common_tokenize(ctx, slot.stopping_word, false);
+                const std::vector<llama_token> stop_word_toks = llama_tokenize(ctx, slot.stopping_word, false);
                probs = std::vector<completion_token_output>(slot.generated_token_probs.begin(), slot.generated_token_probs.end() - stop_word_toks.size());
            }
            else
@@ -1409,6 +1379,7 @@ struct llama_server_context
                    batch.n_seq_id + i,
                    batch.seq_id   + i,
                    batch.logits   + i,
+                    0, 0, 0, // unused
                };
                if (llama_decode(ctx, batch_view))
                {
@@ -1427,9 +1398,8 @@ struct llama_server_context
                }

                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))
+                llama_batch batch_img = { n_eval, nullptr, (img.image_embedding + i * n_embd), nullptr, nullptr, nullptr, nullptr, slot.n_past, 1, 0, };
+                if (llama_decode(ctx, batch_img))
                {
                    LOG("%s : failed to eval image\n", __func__);
                    return false;
@@ -1438,7 +1408,7 @@ struct llama_server_context
            }
            image_idx++;

-            common_batch_clear(batch);
+            llama_batch_clear(batch);

            // append prefix of next image
            const auto json_prompt = (image_idx >= (int) slot.images.size()) ?
@@ -1448,7 +1418,7 @@ struct llama_server_context
            std::vector<llama_token> append_tokens = tokenize(json_prompt, false); // has next image
            for (int i = 0; i < (int) append_tokens.size(); ++i)
            {
-                common_batch_add(batch, append_tokens[i], system_tokens.size() + slot.n_past, { slot.id }, true);
+                llama_batch_add(batch, append_tokens[i], system_tokens.size() + slot.n_past, { slot.id }, true);
                slot.n_past += 1;
            }
        }
@@ -1580,7 +1550,7 @@ struct llama_server_context
            update_system_prompt();
        }

-        common_batch_clear(batch);
+        llama_batch_clear(batch);

        if (all_slots_are_idle)
        {
@@ -1658,7 +1628,7 @@ struct llama_server_context

            // TODO: we always have to take into account the "system_tokens"
            //       this is not great and needs to be improved somehow
-            common_batch_add(batch, slot.sampled, system_tokens.size() + slot_npast, { slot.id }, true);
+            llama_batch_add(batch, slot.sampled, system_tokens.size() + slot_npast, { slot.id }, true);
            slot.n_past += 1;
        }

@@ -1752,7 +1722,7 @@ struct llama_server_context

                    if (!slot.params.cache_prompt)
                    {
-                        common_sampler_reset(slot.ctx_sampling);
+                        gpt_sampler_reset(slot.ctx_sampling);

                        slot.n_past = 0;
                        slot.n_past_se = 0;
@@ -1764,7 +1734,7 @@ struct llama_server_context
                        // push the prompt into the sampling context (do not apply grammar)
                        for (auto &token : prompt_tokens)
                        {
-                            common_sampler_accept(slot.ctx_sampling, token, false);
+                            gpt_sampler_accept(slot.ctx_sampling, token, false);
                        }

                        slot.n_past = common_part(slot.cache_tokens, prompt_tokens);
@@ -1856,7 +1826,7 @@ struct llama_server_context
                                ga_i += ga_w/ga_n;
                            }
                        }
-                        common_batch_add(batch, prefix_tokens[slot.n_past], system_tokens.size() + slot_npast, {slot.id }, false);
+                        llama_batch_add(batch, prefix_tokens[slot.n_past], system_tokens.size() + slot_npast, {slot.id }, false);
                        slot_npast++;
                    }

@@ -1934,6 +1904,7 @@ struct llama_server_context
                batch.n_seq_id + i,
                batch.seq_id   + i,
                batch.logits   + i,
+                0, 0, 0, // unused
            };

            const int ret = llama_decode(ctx, batch_view);
@@ -1972,9 +1943,9 @@ struct llama_server_context
                }

                completion_token_output result;
-                const llama_token id = common_sampler_sample(slot.ctx_sampling, ctx, slot.i_batch - i);
+                const llama_token id = gpt_sampler_sample(slot.ctx_sampling, ctx, slot.i_batch - i);

-                common_sampler_accept(slot.ctx_sampling, id, true);
+                gpt_sampler_accept(slot.ctx_sampling, id, true);

                slot.n_decoded += 1;
                if (slot.n_decoded == 1)
@@ -1985,7 +1956,7 @@ struct llama_server_context
                }

                result.tok = id;
-                const auto * cur_p = common_sampler_get_candidates(slot.ctx_sampling);
+                const auto * cur_p = gpt_sampler_get_candidates(slot.ctx_sampling);

                for (size_t i = 0; i < (size_t) slot.sparams.n_probs; ++i) {
                    result.probs.push_back({
@@ -2038,7 +2009,7 @@ static json format_partial_response(
 struct token_translator
 {
    llama_context * ctx;
-    std::string operator()(llama_token tok)                    const { return common_token_to_piece(ctx, tok); }
+    std::string operator()(llama_token tok)                    const { return llama_token_to_piece(ctx, tok); }
    std::string operator()(const completion_token_output &cto) const { return (*this)(cto.tok); }
 };

@@ -2103,6 +2074,7 @@ 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);
@@ -2126,6 +2098,7 @@ 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();
@@ -2172,6 +2145,7 @@ 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();
@@ -2229,7 +2203,7 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
 // }

 static void params_parse(const backend::ModelOptions* request,
-                                common_params & params) {
+                                gpt_params & params) {
   
    // this is comparable to: https://github.com/ggerganov/llama.cpp/blob/d9b33fe95bd257b36c84ee5769cc048230067d6f/examples/server/server.cpp#L1809

@@ -2299,7 +2273,6 @@ 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();

@@ -2338,7 +2311,7 @@ public:

  grpc::Status LoadModel(ServerContext* context, const backend::ModelOptions* request, backend::Result* result) {
    // Implement LoadModel RPC
-    common_params params;
+    gpt_params params;
    params_parse(request, params);

    llama_backend_init();
--- 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: 0,
-		SpeechPadMs:          0,
-	})
-	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-cublas11.txt
+++ b/backend/python/autogptq/requirements-cublas11.txt
@@ -1,2 +1,2 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.4.1+cu118
+torch
--- a/backend/python/autogptq/requirements-cublas12.txt
+++ b/backend/python/autogptq/requirements-cublas12.txt
@@ -1 +1 @@
-torch==2.4.1
+torch
--- a/backend/python/autogptq/requirements-hipblas.txt
+++ b/backend/python/autogptq/requirements-hipblas.txt
@@ -1,2 +1,2 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch==2.4.1+rocm6.0
+torch
--- a/backend/python/autogptq/requirements.txt
+++ b/backend/python/autogptq/requirements.txt
@@ -1,6 +1,6 @@
 accelerate
 auto-gptq==0.7.1
-grpcio==1.68.0
+grpcio==1.66.2
 protobuf
 certifi
 transformers
--- a/backend/python/bark/requirements-cpu.txt
+++ b/backend/python/bark/requirements-cpu.txt
@@ -1,4 +1,4 @@
 transformers
 accelerate
-torch==2.4.1
-torchaudio==2.4.1
+torch
+torchaudio
--- a/backend/python/bark/requirements-cublas11.txt
+++ b/backend/python/bark/requirements-cublas11.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.4.1+cu118
-torchaudio==2.4.1+cu118
+torch
+torchaudio
 transformers
 accelerate
--- a/backend/python/bark/requirements-cublas12.txt
+++ b/backend/python/bark/requirements-cublas12.txt
@@ -1,4 +1,4 @@
-torch==2.4.1
-torchaudio==2.4.1
+torch
+torchaudio
 transformers
 accelerate
--- a/backend/python/bark/requirements-hipblas.txt
+++ b/backend/python/bark/requirements-hipblas.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch==2.4.1+rocm6.0
-torchaudio==2.4.1+rocm6.0
+torch
+torchaudio
 transformers
 accelerate
--- a/backend/python/bark/requirements.txt
+++ b/backend/python/bark/requirements.txt
@@ -1,4 +1,4 @@
 bark==0.1.5
-grpcio==1.68.0
+grpcio==1.66.2
 protobuf
 certifi
--- 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.txt
+++ b/backend/python/common/template/requirements.txt
@@ -1,3 +1,2 @@
-grpcio==1.68.0
-protobuf
-grpcio-tools
+grpcio==1.66.2
+protobuf
--- a/backend/python/coqui/requirements-cpu.txt
+++ b/backend/python/coqui/requirements-cpu.txt
@@ -1,4 +1,3 @@
 transformers
 accelerate
-torch==2.4.1
-coqui-tts
+torch
--- a/backend/python/coqui/requirements-cublas11.txt
+++ b/backend/python/coqui/requirements-cublas11.txt
@@ -1,6 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.4.1+cu118
-torchaudio==2.4.1+cu118
+torch
+torchaudio
 transformers
-accelerate
-coqui-tts
+accelerate
--- a/backend/python/coqui/requirements-cublas12.txt
+++ b/backend/python/coqui/requirements-cublas12.txt
@@ -1,5 +1,4 @@
-torch==2.4.1
-torchaudio==2.4.1
+torch
+torchaudio
 transformers
-accelerate
-coqui-tts
+accelerate
--- a/backend/python/coqui/requirements-hipblas.txt
+++ b/backend/python/coqui/requirements-hipblas.txt
@@ -1,6 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch==2.4.1+rocm6.0
-torchaudio==2.4.1+rocm6.0
+torch
+torchaudio
 transformers
-accelerate
-coqui-tts
+accelerate
--- a/backend/python/coqui/requirements-intel.txt
+++ b/backend/python/coqui/requirements-intel.txt
@@ -5,5 +5,4 @@ torchaudio
 optimum[openvino]
 setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
 transformers
-accelerate
-coqui-tts
+accelerate
--- a/backend/python/coqui/requirements.txt
+++ b/backend/python/coqui/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.68.0
+coqui-tts
+grpcio==1.66.2
 protobuf
-certifi
-packaging==24.1
+certifi
--- a/backend/python/coqui/test.py
+++ b/backend/python/coqui/test.py
@@ -19,7 +19,7 @@ class TestBackendServicer(unittest.TestCase):
        This method sets up the gRPC service by starting the server
        """
        self.service = subprocess.Popen(["python3", "backend.py", "--addr", "localhost:50051"])
-        time.sleep(30)
+        time.sleep(10)

    def tearDown(self) -> None:
        """
--- a/backend/python/diffusers/backend.py
+++ b/backend/python/diffusers/backend.py
@@ -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"
@@ -301,34 +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')
@@ -409,6 +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,
        }

@@ -426,13 +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:
@@ -443,12 +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)
@@ -468,7 +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-cpu.txt
+++ b/backend/python/diffusers/requirements-cpu.txt
@@ -5,5 +5,5 @@ accelerate
 compel
 peft
 sentencepiece
-torch==2.4.1
+torch
 optimum-quanto
--- a/backend/python/diffusers/requirements-cublas11.txt
+++ b/backend/python/diffusers/requirements-cublas11.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.4.1+cu118
+torch
 diffusers
 opencv-python
 transformers
--- a/backend/python/diffusers/requirements-cublas12.txt
+++ b/backend/python/diffusers/requirements-cublas12.txt
@@ -1,4 +1,4 @@
-torch==2.4.1
+torch
 diffusers
 opencv-python
 transformers
--- a/backend/python/diffusers/requirements.txt
+++ b/backend/python/diffusers/requirements.txt
@@ -1,5 +1,5 @@
 setuptools
-grpcio==1.68.0
+grpcio==1.66.2
 pillow
 protobuf
 certifi
--- a/backend/python/exllama2/requirements-cpu.txt
+++ b/backend/python/exllama2/requirements-cpu.txt
@@ -1,3 +1,3 @@
 transformers
 accelerate
-torch==2.4.1
+torch
--- a/backend/python/exllama2/requirements-cublas11.txt
+++ b/backend/python/exllama2/requirements-cublas11.txt
@@ -1,4 +1,4 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.4.1+cu118
+torch
 transformers
 accelerate
--- a/backend/python/exllama2/requirements-cublas12.txt
+++ b/backend/python/exllama2/requirements-cublas12.txt
@@ -1,3 +1,3 @@
-torch==2.4.1
+torch
 transformers
 accelerate
--- a/backend/python/exllama2/requirements.txt
+++ b/backend/python/exllama2/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.68.0
+grpcio==1.66.2
 protobuf
 certifi
 wheel
--- a/backend/python/mamba/requirements-cpu.txt
+++ b/backend/python/mamba/requirements-cpu.txt
@@ -1,2 +1,2 @@
-torch==2.4.1
+torch
 transformers
--- a/backend/python/mamba/requirements-cublas11.txt
+++ b/backend/python/mamba/requirements-cublas11.txt
@@ -1,3 +1,3 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.4.1+cu118
+torch
 transformers
--- a/backend/python/mamba/requirements-cublas12.txt
+++ b/backend/python/mamba/requirements-cublas12.txt
@@ -1,2 +1,2 @@
-torch==2.4.1
+torch
 transformers
--- a/backend/python/mamba/requirements.txt
+++ b/backend/python/mamba/requirements.txt
@@ -1,3 +1,3 @@
-grpcio==1.68.0
+grpcio==1.66.2
 protobuf
 certifi
--- a/backend/python/openvoice/requirements-cpu.txt
+++ b/backend/python/openvoice/requirements-cpu.txt
@@ -1,3 +1 @@
-torch==2.4.1
-git+https://github.com/myshell-ai/MeloTTS.git
-git+https://github.com/myshell-ai/OpenVoice.git
+torch
--- a/backend/python/openvoice/requirements-cublas11.txt
+++ b/backend/python/openvoice/requirements-cublas11.txt
@@ -1,4 +1,2 @@
 --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
+torch
--- a/backend/python/openvoice/requirements-cublas12.txt
+++ b/backend/python/openvoice/requirements-cublas12.txt
@@ -1,3 +1 @@
-torch==2.4.1
-git+https://github.com/myshell-ai/MeloTTS.git
-git+https://github.com/myshell-ai/OpenVoice.git
+torch
--- a/backend/python/openvoice/requirements-hipblas.txt
+++ b/backend/python/openvoice/requirements-hipblas.txt
@@ -1,4 +1,2 @@
 --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
+torch
--- a/backend/python/openvoice/requirements-intel.txt
+++ b/backend/python/openvoice/requirements-intel.txt
@@ -2,22 +2,22 @@
 intel-extension-for-pytorch
 torch
 optimum[openvino]
-grpcio==1.68.0
+grpcio==1.66.2
 protobuf
 librosa==0.9.1
-faster-whisper==0.9.0
+faster-whisper==1.0.3
 pydub==0.25.1
 wavmark==0.0.3
-numpy==1.22.0
+numpy==1.26.4
 eng_to_ipa==0.0.2
 inflect==7.0.0
 unidecode==1.3.7
-whisper-timestamped==1.14.2
+whisper-timestamped==1.15.4
 openai
 python-dotenv
 pypinyin==0.50.0
 cn2an==0.5.22
 jieba==0.42.1
+gradio==4.44.1
 langid==1.1.6
 git+https://github.com/myshell-ai/MeloTTS.git
-git+https://github.com/myshell-ai/OpenVoice.git
--- a/backend/python/openvoice/requirements.txt
+++ b/backend/python/openvoice/requirements.txt
@@ -1,10 +1,10 @@
-grpcio==1.68.0
+grpcio==1.66.2
 protobuf
 librosa
 faster-whisper
 pydub==0.25.1
 wavmark==0.0.3
-numpy==1.22.0
+numpy
 eng_to_ipa==0.0.2
 inflect
 unidecode
@@ -13,8 +13,8 @@ openai
 python-dotenv
 pypinyin
 cn2an==0.5.22
-networkx==2.8.8
 jieba==0.42.1
-gradio==3.48.0
+gradio
 langid==1.1.6
-llvmlite==0.43.0
+git+https://github.com/myshell-ai/MeloTTS.git
+git+https://github.com/myshell-ai/OpenVoice.git
--- a/backend/python/openvoice/test.py
+++ b/backend/python/openvoice/test.py
@@ -19,7 +19,7 @@ class TestBackendServicer(unittest.TestCase):
        This method sets up the gRPC service by starting the server
        """
        self.service = subprocess.Popen(["python3", "backend.py", "--addr", "localhost:50051"])
-        time.sleep(30)
+        time.sleep(10)

    def tearDown(self) -> None:
        """
--- 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
@@ -11,10 +11,8 @@ 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
--- 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 +1,3 @@
 git+https://github.com/huggingface/parler-tts.git@8e465f1b5fcd223478e07175cb40494d19ffbe17
 llvmlite==0.43.0
 numba==0.60.0
-grpcio-tools==1.42.0
--- a/backend/python/parler-tts/requirements-cpu.txt
+++ b/backend/python/parler-tts/requirements-cpu.txt
@@ -1,3 +1,3 @@
 transformers
 accelerate
-torch==2.4.1
+torch
--- a/backend/python/parler-tts/requirements-cublas11.txt
+++ b/backend/python/parler-tts/requirements-cublas11.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.4.1+cu118
-torchaudio==2.4.1+cu118
+torch
+torchaudio
 transformers
 accelerate
--- a/backend/python/parler-tts/requirements-cublas12.txt
+++ b/backend/python/parler-tts/requirements-cublas12.txt
@@ -1,4 +1,4 @@
-torch==2.4.1
-torchaudio==2.4.1
+torch
+torchaudio
 transformers
 accelerate
--- a/backend/python/parler-tts/requirements.txt
+++ b/backend/python/parler-tts/requirements.txt
@@ -1,3 +1,4 @@
-grpcio==1.68.0
+grpcio==1.66.2
+protobuf
 certifi
-llvmlite==0.43.0
+llvmlite==0.43.0
--- a/backend/python/rerankers/requirements-cpu.txt
+++ b/backend/python/rerankers/requirements-cpu.txt
@@ -1,4 +1,4 @@
 transformers
 accelerate
-torch==2.4.1
+torch
 rerankers[transformers]
--- a/backend/python/rerankers/requirements-cublas11.txt
+++ b/backend/python/rerankers/requirements-cublas11.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
 transformers
 accelerate
-torch==2.4.1+cu118
+torch
 rerankers[transformers]
--- a/backend/python/rerankers/requirements-cublas12.txt
+++ b/backend/python/rerankers/requirements-cublas12.txt
@@ -1,4 +1,4 @@
 transformers
 accelerate
-torch==2.4.1
+torch
 rerankers[transformers]
--- a/backend/python/rerankers/requirements-hipblas.txt
+++ b/backend/python/rerankers/requirements-hipblas.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
 transformers
 accelerate
-torch==2.4.1+rocm6.0
+torch
 rerankers[transformers]
--- a/backend/python/rerankers/requirements.txt
+++ b/backend/python/rerankers/requirements.txt
@@ -1,3 +1,3 @@
-grpcio==1.68.0
+grpcio==1.66.2
 protobuf
 certifi
--- a/backend/python/sentencetransformers/requirements-cpu.txt
+++ b/backend/python/sentencetransformers/requirements-cpu.txt
@@ -1,6 +1,6 @@
-torch==2.4.1
+torch
 accelerate
 transformers
 bitsandbytes
-sentence-transformers==3.3.1
+sentence-transformers==3.1.1
 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
+torch
 accelerate
-sentence-transformers==3.3.1
+sentence-transformers==3.1.1
 transformers
--- a/backend/python/sentencetransformers/requirements-cublas12.txt
+++ b/backend/python/sentencetransformers/requirements-cublas12.txt
@@ -1,4 +1,4 @@
-torch==2.4.1
+torch
 accelerate
-sentence-transformers==3.3.1
+sentence-transformers==3.1.1
 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
+torch
 accelerate
-sentence-transformers==3.3.1
+sentence-transformers==3.1.1
 transformers
--- a/backend/python/sentencetransformers/requirements-intel.txt
+++ b/backend/python/sentencetransformers/requirements-intel.txt
@@ -4,5 +4,5 @@ torch
 optimum[openvino]
 setuptools==69.5.1 # https://github.com/mudler/LocalAI/issues/2406
 accelerate
-sentence-transformers==3.3.1
+sentence-transformers==3.1.1
 transformers
--- a/backend/python/sentencetransformers/requirements.txt
+++ b/backend/python/sentencetransformers/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.68.0
+grpcio==1.66.2
 protobuf
 certifi
 datasets
--- a/backend/python/transformers-musicgen/requirements-cpu.txt
+++ b/backend/python/transformers-musicgen/requirements-cpu.txt
@@ -1,3 +1,3 @@
 transformers
 accelerate
-torch==2.4.1
+torch
--- a/backend/python/transformers-musicgen/requirements-cublas11.txt
+++ b/backend/python/transformers-musicgen/requirements-cublas11.txt
@@ -1,4 +1,4 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
 transformers
 accelerate
-torch==2.4.1+cu118
+torch
--- a/backend/python/transformers-musicgen/requirements-cublas12.txt
+++ b/backend/python/transformers-musicgen/requirements-cublas12.txt
@@ -1,3 +1,3 @@
 transformers
 accelerate
-torch==2.4.1
+torch
--- a/backend/python/transformers-musicgen/requirements-hipblas.txt
+++ b/backend/python/transformers-musicgen/requirements-hipblas.txt
@@ -1,4 +1,4 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
 transformers
 accelerate
-torch==2.4.1+rocm6.0
+torch
--- a/backend/python/transformers-musicgen/requirements.txt
+++ b/backend/python/transformers-musicgen/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.68.0
+grpcio==1.66.2
 protobuf
 scipy==1.14.0
 certifi
--- a/backend/python/transformers/backend.py
+++ b/backend/python/transformers/backend.py
@@ -72,12 +72,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
        Returns:
            A Result object that contains the result of the LoadModel operation.
        """
-
        model_name = request.Model
-        
-        # Check to see if the Model exists in the filesystem already.
-        if os.path.exists(request.ModelFile):
-            model_name = request.ModelFile

        compute = torch.float16
        if request.F16Memory == True:
--- a/backend/python/transformers/requirements-cpu.txt
+++ b/backend/python/transformers/requirements-cpu.txt
@@ -1,4 +1,4 @@
-torch==2.4.1
+torch
 accelerate
 transformers
 bitsandbytes
--- a/backend/python/transformers/requirements-cublas11.txt
+++ b/backend/python/transformers/requirements-cublas11.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.4.1+cu118
+torch
 accelerate
 transformers
 bitsandbytes
--- a/backend/python/transformers/requirements-cublas12.txt
+++ b/backend/python/transformers/requirements-cublas12.txt
@@ -1,4 +1,4 @@
-torch==2.4.1
+torch
 accelerate
 transformers
 bitsandbytes
--- a/backend/python/transformers/requirements-hipblas.txt
+++ b/backend/python/transformers/requirements-hipblas.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch==2.4.1+rocm6.0
+torch
 accelerate
 transformers
 bitsandbytes
--- a/backend/python/transformers/requirements.txt
+++ b/backend/python/transformers/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.68.0
+grpcio==1.66.2
 protobuf
 certifi
 setuptools==69.5.1 # https://github.com/mudler/LocalAI/issues/2406
--- a/backend/python/vall-e-x/requirements-cpu.txt
+++ b/backend/python/vall-e-x/requirements-cpu.txt
@@ -1,3 +1,3 @@
 accelerate
-torch==2.4.1
-torchaudio==2.4.1
+torch
+torchaudio
--- a/backend/python/vall-e-x/requirements-cublas11.txt
+++ b/backend/python/vall-e-x/requirements-cublas11.txt
@@ -1,4 +1,4 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
 accelerate
-torch==2.4.1+cu118
-torchaudio==2.4.1+cu118
+torch
+torchaudio
--- a/backend/python/vall-e-x/requirements-cublas12.txt
+++ b/backend/python/vall-e-x/requirements-cublas12.txt
@@ -1,3 +1,3 @@
 accelerate
-torch==2.4.1
-torchaudio==2.4.1
+torch
+torchaudio
--- a/backend/python/vall-e-x/requirements.txt
+++ b/backend/python/vall-e-x/requirements.txt
@@ -1,3 +1,3 @@
-grpcio==1.68.0
+grpcio==1.66.2
 protobuf
 certifi
--- a/backend/python/vllm/backend.py
+++ b/backend/python/vllm/backend.py
@@ -19,8 +19,6 @@ from vllm.utils import random_uuid
 from vllm.transformers_utils.tokenizer import get_tokenizer
 from vllm.multimodal.utils import fetch_image
 from vllm.assets.video import VideoAsset
-import base64
-import io

 _ONE_DAY_IN_SECONDS = 60 * 60 * 24

@@ -95,8 +93,6 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):

        if request.Quantization != "":
            engine_args.quantization = request.Quantization
-        if request.LoadFormat != "":
-            engine_args.load_format = request.LoadFormat
        if request.GPUMemoryUtilization != 0:
            engine_args.gpu_memory_utilization = request.GPUMemoryUtilization
        if request.TrustRemoteCode:
@@ -221,15 +217,13 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
        # Generate text using the LLM engine
        request_id = random_uuid()
        print(f"Generating text with request_id: {request_id}", file=sys.stderr)
-        multi_modal_data = {}
-        if image_data:
-            multi_modal_data["image"] = image_data
-        if video_data:
-            multi_modal_data["video"] = video_data
        outputs = self.llm.generate(
            {
-            "prompt": prompt,
-            "multi_modal_data": multi_modal_data if multi_modal_data else None,
+                "prompt": prompt,
+                "multi_modal_data": {
+                    "image": image_data if image_data else None,
+                    "video": video_data if video_data else None,
+                } if image_data or video_data else None,
            },
            sampling_params=sampling_params,
            request_id=request_id,
@@ -268,22 +262,19 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):

    def load_image(self, image_path: str):
        """
-        Load an image from the given file path or base64 encoded data.
+        Load an image from the given file path.
        
        Args:
-            image_path (str): The path to the image file or base64 encoded data.
+            image_path (str): The path to the image file.

        Returns:
            Image: The loaded image.
        """
        try:
-
-            image_data = base64.b64decode(image_path)
-            image = Image.open(io.BytesIO(image_data))
-            return image
+            return Image.open(image_path)
        except Exception as e:
            print(f"Error loading image {image_path}: {e}", file=sys.stderr)
-            return None
+            return self.load_video(image_path)

    def load_video(self, video_path: str):
        """
@@ -296,15 +287,10 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
            Video: The loaded video.
        """
        try:
-            timestamp = str(int(time.time() * 1000))  # Generate timestamp
-            p = f"/tmp/vl-{timestamp}.data"  # Use timestamp in filename
-            with open(p, "wb") as f:
-                f.write(base64.b64decode(video_path))
-            video = VideoAsset(name=p).np_ndarrays
-            os.remove(p)
+            video = VideoAsset(name=video_path).np_ndarrays
            return video
        except Exception as e:
-            print(f"Error loading video {video_path}: {e}", file=sys.stderr)
+            print(f"Error loading video {image_path}: {e}", file=sys.stderr)
            return None

 async def serve(address):
--- a/backend/python/vllm/install.sh
+++ b/backend/python/vllm/install.sh
@@ -13,16 +13,14 @@ if [ "x${BUILD_PROFILE}" == "xintel" ]; then
    EXTRA_PIP_INSTALL_FLAGS+=" --upgrade --index-strategy=unsafe-first-match"
 fi

-# We don't embed this into the images as it is a large dependency and not always needed.
-# Besides, the speed inference are not actually usable in the current state for production use-cases.
-if [ "x${BUILD_TYPE}" == "x" ] && [ "x${FROM_SOURCE}" == "xtrue" ]; then
+if [ "x${BUILD_TYPE}" == "x" ]; then
        ensureVenv
        # https://docs.vllm.ai/en/v0.6.1/getting_started/cpu-installation.html
        if [ ! -d vllm ]; 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==1.68.0 protobuf bitsandbytes
+            uv pip install wheel packaging ninja "setuptools>=49.4.0" numpy typing-extensions pillow setuptools-scm grpcio==1.66.2 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-cpu.txt
+++ b/backend/python/vllm/requirements-cpu.txt
@@ -1,3 +1,3 @@
 accelerate
-torch==2.4.1
+torch
 transformers
--- a/backend/python/vllm/requirements-cublas11.txt
+++ b/backend/python/vllm/requirements-cublas11.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
 accelerate
-torch==2.4.1+cu118
+torch
 transformers
 bitsandbytes
--- a/Show More
+++ b/Show More