fix: roll out bluemonday Sanitize more widely (#3794)

* initial pass: roll out bluemonday sanitization more widely

Signed-off-by: Dave Lee <dave@gray101.com>

* add one additional sanitize - the overall modelslist used by the docs site

Signed-off-by: Dave Lee <dave@gray101.com>

---------

Signed-off-by: Dave Lee <dave@gray101.com>

.devcontainer-scripts/utils.sh

@@ -9,6 +9,7 @@
 # Param 2: email
 #
 config_user() {
+    echo "Configuring git for $1 <$2>"
     local gcn=$(git config --global user.name)
     if [ -z "${gcn}" ]; then
         echo "Setting up git user / remote"
@@ -24,6 +25,7 @@ config_user() {
 # Param 2: remote url
 #
 config_remote() {
+    echo "Adding git remote and fetching $2 as $1"
     local gr=$(git remote -v | grep $1)
     if [ -z "${gr}" ]; then
         git remote add $1 $2
@@ -32,18 +34,22 @@ config_remote() {
 }
 
 # Setup special .ssh files
-#
+# Prints out lines of text to make things pretty
 # Param 1: bash array, filenames relative to the customization directory that should be copied to ~/.ssh
 setup_ssh() {
+    echo "starting ~/.ssh directory setup..."
+    mkdir -p "${HOME}.ssh"
+    chmod 0700 "${HOME}/.ssh"
+    echo "-----"
     local files=("$@")
-    for file in "${files[@]}"; then
+    for file in "${files[@]}" ; do
         local cfile="/devcontainer-customization/${file}"
-        local hfile="~/.ssh/${file}"
+        local hfile="${HOME}/.ssh/${file}"
         if [ ! -f "${hfile}" ]; then
-            echo "copying ${file}"
+            echo "copying \"${file}\""
             cp "${cfile}" "${hfile}"
             chmod 600 "${hfile}"
         fi
     done
-    ls ~/.ssh
+    echo "~/.ssh directory setup complete!"
 }

.github/check_and_update.py

@@ -29,9 +29,14 @@ def calculate_sha256(file_path):
 def manual_safety_check_hf(repo_id):
     scanResponse = requests.get('https://huggingface.co/api/models/' + repo_id + "/scan")
     scan = scanResponse.json()
-    if scan['hasUnsafeFile']:
-        return scan
-    return None
+    # Check if 'hasUnsafeFile' exists in the response
+    if 'hasUnsafeFile' in scan:
+        if scan['hasUnsafeFile']:
+            return scan
+        else:
+            return None
+    else:
+        return None
 
 download_type, repo_id_or_url = parse_uri(uri)
 

.github/ci/modelslist.go

@@ -6,6 +6,7 @@ import (
 	"io/ioutil"
 	"os"
 
+	"github.com/microcosm-cc/bluemonday"
 	"gopkg.in/yaml.v3"
 )
 
@@ -279,6 +280,12 @@ 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

.github/workflows/bump_deps.yaml

@@ -56,7 +56,7 @@ jobs:
           rm -rfv ${{ matrix.variable }}_message.txt
           rm -rfv ${{ matrix.variable }}_commit.txt
       - name: Create Pull Request
-        uses: peter-evans/create-pull-request@v6
+        uses: peter-evans/create-pull-request@v7
         with:
           token: ${{ secrets.UPDATE_BOT_TOKEN }}
           push-to-fork: ci-forks/LocalAI

.github/workflows/bump_docs.yaml

@@ -17,7 +17,7 @@ jobs:
         run: |
           bash .github/bump_docs.sh ${{ matrix.repository }}
       - name: Create Pull Request
-        uses: peter-evans/create-pull-request@v6
+        uses: peter-evans/create-pull-request@v7
         with:
           token: ${{ secrets.UPDATE_BOT_TOKEN }}
           push-to-fork: ci-forks/LocalAI

.github/workflows/checksum_checker.yaml

@@ -36,7 +36,7 @@ jobs:
           sudo chmod 777 /hf_cache
           bash .github/checksum_checker.sh gallery/index.yaml
       - name: Create Pull Request
-        uses: peter-evans/create-pull-request@v6
+        uses: peter-evans/create-pull-request@v7
         with:
           token: ${{ secrets.UPDATE_BOT_TOKEN }}
           push-to-fork: ci-forks/LocalAI

.github/workflows/deploy-explorer.yaml

@@ -33,7 +33,7 @@ jobs:
         run: |
           CGO_ENABLED=0 make build-api
       - name: rm
-        uses: appleboy/ssh-action@v1.0.3
+        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@v1.0.3
+        uses: appleboy/ssh-action@v1.1.0
         with:
             host: ${{ secrets.EXPLORER_SSH_HOST }}
             username: ${{ secrets.EXPLORER_SSH_USERNAME }}

.github/workflows/image.yml

@@ -13,6 +13,78 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
+  hipblas-jobs:
+    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 }}
+      base-image: ${{ matrix.base-image }}
+      grpc-base-image: ${{ matrix.grpc-base-image }}
+      aio: ${{ matrix.aio }}
+      makeflags: ${{ matrix.makeflags }}
+      latest-image: ${{ matrix.latest-image }}
+      latest-image-aio: ${{ matrix.latest-image-aio }}
+    secrets:
+      dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
+      dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
+      quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
+      quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
+    strategy:
+      # Pushing with all jobs in parallel
+      # eats the bandwidth of all the nodes
+      max-parallel: 2
+      matrix:
+        include:
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'auto'
+            tag-suffix: '-hipblas-ffmpeg'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            aio: "-aio-gpu-hipblas"
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            grpc-base-image: "ubuntu:22.04"
+            latest-image: 'latest-gpu-hipblas'
+            latest-image-aio: 'latest-aio-gpu-hipblas'
+            runs-on: 'arc-runner-set'
+            makeflags: "--jobs=3 --output-sync=target"
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas'
+            ffmpeg: 'false'
+            image-type: 'extras'
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            grpc-base-image: "ubuntu:22.04"
+            runs-on: 'arc-runner-set'
+            makeflags: "--jobs=3 --output-sync=target"
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas-ffmpeg-core'
+            ffmpeg: 'true'
+            image-type: 'core'
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            grpc-base-image: "ubuntu:22.04"
+            runs-on: 'arc-runner-set'
+            makeflags: "--jobs=3 --output-sync=target"
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas-core'
+            ffmpeg: 'false'
+            image-type: 'core'
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            grpc-base-image: "ubuntu:22.04"
+            runs-on: 'arc-runner-set'
+            makeflags: "--jobs=3 --output-sync=target"
   self-hosted-jobs:
     uses: ./.github/workflows/image_build.yml
     with:
@@ -39,7 +111,7 @@ jobs:
     strategy:
       # Pushing with all jobs in parallel
       # eats the bandwidth of all the nodes
-      max-parallel: ${{ github.event_name != 'pull_request' && 6 || 10 }}
+      max-parallel: ${{ github.event_name != 'pull_request' && 5 || 8 }}
       matrix:
         include:
           # Extra images
@@ -122,29 +194,6 @@ jobs:
             base-image: "ubuntu:22.04"
             runs-on: 'arc-runner-set'
             makeflags: "--jobs=3 --output-sync=target"
-          - build-type: 'hipblas'
-            platforms: 'linux/amd64'
-            tag-latest: 'auto'
-            tag-suffix: '-hipblas-ffmpeg'
-            ffmpeg: 'true'
-            image-type: 'extras'
-            aio: "-aio-gpu-hipblas"
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
-            grpc-base-image: "ubuntu:22.04"
-            latest-image: 'latest-gpu-hipblas'
-            latest-image-aio: 'latest-aio-gpu-hipblas'
-            runs-on: 'arc-runner-set'
-            makeflags: "--jobs=3 --output-sync=target"
-          - build-type: 'hipblas'
-            platforms: 'linux/amd64'
-            tag-latest: 'false'
-            tag-suffix: '-hipblas'
-            ffmpeg: 'false'
-            image-type: 'extras'
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
-            grpc-base-image: "ubuntu:22.04"
-            runs-on: 'arc-runner-set'
-            makeflags: "--jobs=3 --output-sync=target"
           - build-type: 'sycl_f16'
             platforms: 'linux/amd64'
             tag-latest: 'auto'
@@ -212,26 +261,6 @@ jobs:
             image-type: 'core'
             runs-on: 'arc-runner-set'
             makeflags: "--jobs=3 --output-sync=target"
-          - build-type: 'hipblas'
-            platforms: 'linux/amd64'
-            tag-latest: 'false'
-            tag-suffix: '-hipblas-ffmpeg-core'
-            ffmpeg: 'true'
-            image-type: 'core'
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
-            grpc-base-image: "ubuntu:22.04"
-            runs-on: 'arc-runner-set'
-            makeflags: "--jobs=3 --output-sync=target"
-          - build-type: 'hipblas'
-            platforms: 'linux/amd64'
-            tag-latest: 'false'
-            tag-suffix: '-hipblas-core'
-            ffmpeg: 'false'
-            image-type: 'core'
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
-            grpc-base-image: "ubuntu:22.04"
-            runs-on: 'arc-runner-set'
-            makeflags: "--jobs=3 --output-sync=target"
 
   core-image-build:
     uses: ./.github/workflows/image_build.yml

.github/workflows/release.yaml

@@ -294,7 +294,7 @@ jobs:
           export C_INCLUDE_PATH=/usr/local/include
           export CPLUS_INCLUDE_PATH=/usr/local/include
           export PATH=$PATH:$GOPATH/bin
-
+          export SKIP_GRPC_BACKEND=backend-assets/grpc/whisper
           make dist
       - uses: actions/upload-artifact@v4
         with:
@@ -327,7 +327,7 @@ jobs:
           cache: false
       - name: Dependencies
         run: |
-          brew install protobuf grpc
+          brew install protobuf grpc libomp llvm
           go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
           go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
       - name: Build
@@ -336,7 +336,7 @@ jobs:
           export C_INCLUDE_PATH=/usr/local/include
           export CPLUS_INCLUDE_PATH=/usr/local/include
           export PATH=$PATH:$GOPATH/bin
-
+          export CC=/opt/homebrew/opt/llvm/bin/clang
           make dist
       - uses: actions/upload-artifact@v4
         with:

.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@master
+        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 ./...'

.github/workflows/test.yml

@@ -178,13 +178,22 @@ jobs:
         uses: actions/checkout@v4
         with:
           submodules: true
+      - name: Dependencies
+        run: |
+          # Install protoc
+          curl -L -s https://github.com/protocolbuffers/protobuf/releases/download/v26.1/protoc-26.1-linux-x86_64.zip -o protoc.zip && \
+          unzip -j -d /usr/local/bin protoc.zip bin/protoc && \
+          rm protoc.zip
+          go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
+          go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
+          PATH="$PATH:$HOME/go/bin" make protogen-go
       - name: Build images
         run: |
           docker build --build-arg FFMPEG=true --build-arg IMAGE_TYPE=extras --build-arg EXTRA_BACKENDS=rerankers --build-arg MAKEFLAGS="--jobs=5 --output-sync=target" -t local-ai:tests -f Dockerfile .
           BASE_IMAGE=local-ai:tests DOCKER_AIO_IMAGE=local-ai-aio:test make docker-aio
       - name: Test
         run: |
-          LOCALAI_MODELS_DIR=$PWD/models LOCALAI_IMAGE_TAG=test LOCALAI_IMAGE=local-ai-aio \
+            PATH="$PATH:$HOME/go/bin" LOCALAI_MODELS_DIR=$PWD/models LOCALAI_IMAGE_TAG=test LOCALAI_IMAGE=local-ai-aio \
             make run-e2e-aio
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
@@ -214,12 +223,13 @@ jobs:
         run: go version
       - name: Dependencies
         run: |
-          brew install protobuf grpc make protoc-gen-go protoc-gen-go-grpc
+          brew install protobuf grpc make protoc-gen-go protoc-gen-go-grpc libomp llvm
           pip install --user --no-cache-dir grpcio-tools==1.64.1
       - name: Test
         run: |
           export C_INCLUDE_PATH=/usr/local/include
           export CPLUS_INCLUDE_PATH=/usr/local/include
+          export CC=/opt/homebrew/opt/llvm/bin/clang
           # Used to run the newer GNUMake version from brew that supports --output-sync
           export PATH="/opt/homebrew/opt/make/libexec/gnubin:$PATH"
           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

.github/workflows/update_swagger.yaml

@@ -25,7 +25,7 @@ jobs:
         run: |
           make protogen-go swagger
       - name: Create Pull Request
-        uses: peter-evans/create-pull-request@v6
+        uses: peter-evans/create-pull-request@v7
         with:
           token: ${{ secrets.UPDATE_BOT_TOKEN }}
           push-to-fork: ci-forks/LocalAI

CONTRIBUTING.md

@@ -15,8 +15,6 @@ Thank you for your interest in contributing to LocalAI! We appreciate your time
 - [Documentation](#documentation)
 - [Community and Communication](#community-and-communication)
 
-
-
 ## Getting Started
 
 ### Prerequisites
@@ -54,7 +52,7 @@ If you find a bug, have a feature request, or encounter any issues, please check
 
 ## Coding Guidelines
 
-- No specific coding guidelines at the moment. Please make sure the code can be tested. The most popular lint tools like []`golangci-lint`](https://golangci-lint.run) can help you here.
+- No specific coding guidelines at the moment. Please make sure the code can be tested. The most popular lint tools like [`golangci-lint`](https://golangci-lint.run) can help you here.
 
 ## Testing
 
@@ -84,5 +82,3 @@ We are welcome the contribution of the documents, please open new PR or create a
 - You can reach out via the Github issue tracker.
 - Open a new discussion at [Discussion](https://github.com/go-skynet/LocalAI/discussions)
 - Join the Discord channel [Discord](https://discord.gg/uJAeKSAGDy)
-
----

Dockerfile

@@ -9,11 +9,13 @@ 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
 
 ENV DEBIAN_FRONTEND=noninteractive
-ENV EXTERNAL_GRPC_BACKENDS="coqui:/build/backend/python/coqui/run.sh,huggingface-embeddings:/build/backend/python/sentencetransformers/run.sh,transformers:/build/backend/python/transformers/run.sh,sentencetransformers:/build/backend/python/sentencetransformers/run.sh,rerankers:/build/backend/python/rerankers/run.sh,autogptq:/build/backend/python/autogptq/run.sh,bark:/build/backend/python/bark/run.sh,diffusers:/build/backend/python/diffusers/run.sh,exllama:/build/backend/python/exllama/run.sh,openvoice:/build/backend/python/openvoice/run.sh,vall-e-x:/build/backend/python/vall-e-x/run.sh,vllm:/build/backend/python/vllm/run.sh,mamba:/build/backend/python/mamba/run.sh,exllama2:/build/backend/python/exllama2/run.sh,transformers-musicgen:/build/backend/python/transformers-musicgen/run.sh,parler-tts:/build/backend/python/parler-tts/run.sh"
+ENV EXTERNAL_GRPC_BACKENDS="coqui:/build/backend/python/coqui/run.sh,huggingface-embeddings:/build/backend/python/sentencetransformers/run.sh,transformers:/build/backend/python/transformers/run.sh,sentencetransformers:/build/backend/python/sentencetransformers/run.sh,rerankers:/build/backend/python/rerankers/run.sh,autogptq:/build/backend/python/autogptq/run.sh,bark:/build/backend/python/bark/run.sh,diffusers:/build/backend/python/diffusers/run.sh,openvoice:/build/backend/python/openvoice/run.sh,vall-e-x:/build/backend/python/vall-e-x/run.sh,vllm:/build/backend/python/vllm/run.sh,mamba:/build/backend/python/mamba/run.sh,exllama2:/build/backend/python/exllama2/run.sh,transformers-musicgen:/build/backend/python/transformers-musicgen/run.sh,parler-tts:/build/backend/python/parler-tts/run.sh"
 
 
 RUN apt-get update && \
@@ -21,13 +23,25 @@ RUN apt-get update && \
         build-essential \
         ccache \
         ca-certificates \
-        cmake \
-        curl \
+        curl libssl-dev \
         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
@@ -188,6 +202,8 @@ 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}
 
@@ -196,12 +212,24 @@ WORKDIR /build
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
         ca-certificates \
-        build-essential \
-        cmake \
+        build-essential curl libssl-dev \
         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
@@ -263,14 +291,20 @@ EOT
 # In most cases, builder is the image you should be using - however, this can save build time if one just needs to copy backend-assets/grpc/stablediffusion and nothing else.
 FROM builder-base AS builder-sd
 
-COPY . .
-COPY .git .
+# stablediffusion does not tolerate a newer version of abseil, copy only over enough elements to build it
+COPY Makefile .
+COPY go.mod .
+COPY go.sum .
+COPY backend/backend.proto ./backend/backend.proto
+COPY backend/go/image/stablediffusion ./backend/go/image/stablediffusion
+COPY pkg/grpc ./pkg/grpc
+COPY pkg/stablediffusion ./pkg/stablediffusion
+RUN git init
+RUN make sources/go-stable-diffusion
+RUN touch prepare-sources
 
-RUN make prepare
-
-
-# stablediffusion does not tolerate a newer version of abseil, build it first
-RUN GRPC_BACKENDS=backend-assets/grpc/stablediffusion make build
+# Actually build the backend
+RUN GRPC_BACKENDS=backend-assets/grpc/stablediffusion make backend-assets/grpc/stablediffusion
 
 ###################################
 ###################################
@@ -285,8 +319,20 @@ COPY --from=grpc /opt/grpc /usr/local
 # Rebuild with defaults backends
 WORKDIR /build
 
+COPY . .
+COPY .git .
+
+RUN make prepare
+
 ## Build the binary
-RUN make build
+## If it's CUDA or hipblas, we want to skip some of the llama-compat backends to save space
+## We only leave the most CPU-optimized variant and the fallback for the cublas/hipblas build
+## (both will use CUDA or hipblas for the actual computation)
+RUN if [ "${BUILD_TYPE}" = "cublas" ] || [ "${BUILD_TYPE}" = "hipblas" ]; then \
+        SKIP_GRPC_BACKEND="backend-assets/grpc/llama-cpp-avx backend-assets/grpc/llama-cpp-avx2" make build; \
+    else \
+        make build; \
+    fi
 
 RUN if [ ! -d "/build/sources/go-piper/piper-phonemize/pi/lib/" ]; then \
         mkdir -p /build/sources/go-piper/piper-phonemize/pi/lib/ \
@@ -320,9 +366,8 @@ RUN if [ "${FFMPEG}" = "true" ]; then \
 
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
-        ssh less && \
-    apt-get clean && \
-    rm -rf /var/lib/apt/lists/*
+        ssh less wget
+# For the devcontainer, leave apt functional in case additional devtools are needed at runtime.
 
 RUN go install github.com/go-delve/delve/cmd/dlv@latest
 
@@ -400,9 +445,6 @@ RUN if [[ ( "${EXTRA_BACKENDS}" =~ "coqui" || -z "${EXTRA_BACKENDS}" ) && "$IMAG
     ; fi && \
     if [[ ( "${EXTRA_BACKENDS}" =~ "transformers-musicgen" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
         make -C backend/python/transformers-musicgen \
-    ; fi && \
-    if [[ ( "${EXTRA_BACKENDS}" =~ "exllama1" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
-        make -C backend/python/exllama \
     ; fi
 
 RUN if [[ ( "${EXTRA_BACKENDS}" =~ "vall-e-x" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \

Makefile

@@ -8,7 +8,7 @@ DETECT_LIBS?=true
 # llama.cpp versions
 GOLLAMA_REPO?=https://github.com/go-skynet/go-llama.cpp
 GOLLAMA_VERSION?=2b57a8ae43e4699d3dc5d1496a1ccd42922993be
-CPPLLAMA_VERSION?=2f3c1466ff46a2413b0e363a5005c46538186ee6
+CPPLLAMA_VERSION?=96776405a17034dcfd53d3ddf5d142d34bdbb657
 
 # go-rwkv version
 RWKV_REPO?=https://github.com/donomii/go-rwkv.cpp
@@ -16,7 +16,7 @@ RWKV_VERSION?=661e7ae26d442f5cfebd2a0881b44e8c55949ec6
 
 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggerganov/whisper.cpp
-WHISPER_CPP_VERSION?=d65786ea540a5aef21f67cacfa6f134097727780
+WHISPER_CPP_VERSION?=fdbfb460ed546452a5d53611bba66d10d842e719
 
 # bert.cpp version
 BERT_REPO?=https://github.com/go-skynet/go-bert.cpp
@@ -338,7 +338,7 @@ rebuild: ## Rebuilds the project
 	$(MAKE) -C sources/go-tiny-dream clean
 	$(MAKE) build
 
-prepare: prepare-sources gen-assets $(OPTIONAL_TARGETS)
+prepare: prepare-sources $(OPTIONAL_TARGETS)
 
 clean: ## Remove build related file
 	$(GOCMD) clean -cache
@@ -359,6 +359,9 @@ clean-tests:
 	rm -rf test-dir
 	rm -rf core/http/backend-assets
 
+clean-dc: clean
+	cp -r /build/backend-assets /workspace/backend-assets
+
 ## Build:
 build: prepare backend-assets grpcs ## Build the project
 	$(info ${GREEN}I local-ai build info:${RESET})
@@ -465,15 +468,15 @@ run-e2e-image:
 	ls -liah $(abspath ./tests/e2e-fixtures)
 	docker run -p 5390:8080 -e MODELS_PATH=/models -e THREADS=1 -e DEBUG=true -d --rm -v $(TEST_DIR):/models --gpus all --name e2e-tests-$(RANDOM) localai-tests
 
-run-e2e-aio:
+run-e2e-aio: protogen-go
 	@echo 'Running e2e AIO tests'
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts 5 -v -r ./tests/e2e-aio
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts $(TEST_FLAKES) -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 5 -v -r ./tests/e2e
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts $(TEST_FLAKES) -v -r ./tests/e2e
 
 teardown-e2e:
 	rm -rf $(TEST_DIR) || true
@@ -481,24 +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 5 -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama" --flake-attempts $(TEST_FLAKES) -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 5 -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama-gguf" --flake-attempts $(TEST_FLAKES) -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 1 -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="tts" --flake-attempts $(TEST_FLAKES) -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 1 -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stablediffusion" --flake-attempts $(TEST_FLAKES) -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 1 -v -r tests/integration
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stores" --flake-attempts $(TEST_FLAKES) -v -r tests/integration
 
 test-container:
 	docker build --target requirements -t local-ai-test-container .
@@ -534,10 +537,10 @@ protogen-go-clean:
 	$(RM) bin/*
 
 .PHONY: protogen-python
-protogen-python: autogptq-protogen bark-protogen coqui-protogen diffusers-protogen exllama-protogen exllama2-protogen mamba-protogen rerankers-protogen sentencetransformers-protogen transformers-protogen parler-tts-protogen transformers-musicgen-protogen vall-e-x-protogen vllm-protogen openvoice-protogen
+protogen-python: autogptq-protogen bark-protogen coqui-protogen diffusers-protogen exllama2-protogen mamba-protogen rerankers-protogen sentencetransformers-protogen transformers-protogen parler-tts-protogen transformers-musicgen-protogen vall-e-x-protogen vllm-protogen openvoice-protogen
 
 .PHONY: protogen-python-clean
-protogen-python-clean: autogptq-protogen-clean bark-protogen-clean coqui-protogen-clean diffusers-protogen-clean exllama-protogen-clean exllama2-protogen-clean mamba-protogen-clean sentencetransformers-protogen-clean rerankers-protogen-clean transformers-protogen-clean transformers-musicgen-protogen-clean parler-tts-protogen-clean vall-e-x-protogen-clean vllm-protogen-clean openvoice-protogen-clean
+protogen-python-clean: autogptq-protogen-clean bark-protogen-clean coqui-protogen-clean diffusers-protogen-clean  exllama2-protogen-clean mamba-protogen-clean sentencetransformers-protogen-clean rerankers-protogen-clean transformers-protogen-clean transformers-musicgen-protogen-clean parler-tts-protogen-clean vall-e-x-protogen-clean vllm-protogen-clean openvoice-protogen-clean
 
 .PHONY: autogptq-protogen
 autogptq-protogen:
@@ -571,14 +574,6 @@ diffusers-protogen:
 diffusers-protogen-clean:
 	$(MAKE) -C backend/python/diffusers protogen-clean
 
-.PHONY: exllama-protogen
-exllama-protogen:
-	$(MAKE) -C backend/python/exllama protogen
-
-.PHONY: exllama-protogen-clean
-exllama-protogen-clean:
-	$(MAKE) -C backend/python/exllama protogen-clean
-
 .PHONY: exllama2-protogen
 exllama2-protogen:
 	$(MAKE) -C backend/python/exllama2 protogen
@@ -675,7 +670,6 @@ prepare-extra-conda-environments: protogen-python
 	$(MAKE) -C backend/python/parler-tts
 	$(MAKE) -C backend/python/vall-e-x
 	$(MAKE) -C backend/python/openvoice
-	$(MAKE) -C backend/python/exllama
 	$(MAKE) -C backend/python/exllama2
 
 prepare-test-extra: protogen-python
@@ -846,7 +840,7 @@ endif
 
 backend-assets/grpc/whisper: sources/whisper.cpp sources/whisper.cpp/libwhisper.a backend-assets/grpc
 	CGO_LDFLAGS="$(CGO_LDFLAGS) $(CGO_LDFLAGS_WHISPER)" C_INCLUDE_PATH="$(CURDIR)/sources/whisper.cpp/include:$(CURDIR)/sources/whisper.cpp/ggml/include" LIBRARY_PATH=$(CURDIR)/sources/whisper.cpp \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/whisper ./backend/go/transcribe/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/whisper ./backend/go/transcribe/whisper
 ifneq ($(UPX),)
 	$(UPX) backend-assets/grpc/whisper
 endif

README.md

@@ -40,7 +40,7 @@
 
 > :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/) [📣 News](https://localai.io/basics/news/) [ 🛫 Examples ](https://github.com/go-skynet/LocalAI/tree/master/examples/) [ 🖼️ Models ](https://localai.io/models/) [ 🚀 Roadmap ](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap)
+> [💻 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)
 
@@ -68,10 +68,9 @@ docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-aio-cpu
 
 [💻 Getting started](https://localai.io/basics/getting_started/index.html)
 
-## 🔥🔥 Hot topics / Roadmap
-
-[Roadmap](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap)
+## 📰 Latest project news
 
+- 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
 - June 2024: Support for models from OCI registries: https://github.com/mudler/LocalAI/pull/2628
@@ -82,8 +81,12 @@ docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-aio-cpu
 - May 2024: Chat, TTS, and Image generation in the WebUI: https://github.com/mudler/LocalAI/pull/2222
 - April 2024: Reranker API: https://github.com/mudler/LocalAI/pull/2121
 
-Hot topics (looking for contributors):
+Roadmap items: [List of issues](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap)
 
+## 🔥🔥 Hot topics (looking for help):
+
+- Multimodal with vLLM and Video understanding: https://github.com/mudler/LocalAI/pull/3729
+- Realtime API https://github.com/mudler/LocalAI/issues/3714
 - 🔥🔥 Distributed, P2P Global community pools: https://github.com/mudler/LocalAI/issues/3113
 - WebUI improvements: https://github.com/mudler/LocalAI/issues/2156
 - Backends v2: https://github.com/mudler/LocalAI/issues/1126

aio/cpu/vision.yaml

@@ -2,7 +2,7 @@ backend: llama-cpp
 context_size: 4096
 f16: true
 mmap: true
-name: gpt-4-vision-preview
+name: gpt-4o
 
 roles:
   user: "USER:"

aio/gpu-8g/vision.yaml

@@ -2,7 +2,7 @@ backend: llama-cpp
 context_size: 4096
 f16: true
 mmap: true
-name: gpt-4-vision-preview
+name: gpt-4o
 
 roles:
   user: "USER:"

aio/intel/image-gen.yaml

@@ -1,6 +1,6 @@
 name: stablediffusion
 parameters:
-  model: runwayml/stable-diffusion-v1-5
+  model: Lykon/dreamshaper-8
 backend: diffusers
 step: 25
 f16: true

aio/intel/vision.yaml

@@ -2,7 +2,7 @@ backend: llama-cpp
 context_size: 4096
 mmap: false
 f16: false
-name: gpt-4-vision-preview
+name: gpt-4o
 
 roles:
   user: "USER:"

backend/backend.proto

@@ -16,6 +16,7 @@ service Backend {
   rpc GenerateImage(GenerateImageRequest) returns (Result) {}
   rpc AudioTranscription(TranscriptRequest) returns (TranscriptResult) {}
   rpc TTS(TTSRequest) returns (Result) {}
+  rpc SoundGeneration(SoundGenerationRequest) returns (Result) {}
   rpc TokenizeString(PredictOptions) returns (TokenizationResponse) {}
   rpc Status(HealthMessage) returns (StatusResponse) {}
 
@@ -25,6 +26,19 @@ service Backend {
   rpc StoresFind(StoresFindOptions) returns (StoresFindResult) {}
 
   rpc Rerank(RerankRequest) returns (RerankResult) {}
+
+  rpc GetMetrics(MetricsRequest) returns (MetricsResponse);
+}
+
+// Define the empty request
+message MetricsRequest {}
+
+message MetricsResponse {
+  int32 slot_id = 1;
+  string prompt_json_for_slot = 2;  // Stores the prompt as a JSON string.
+  float tokens_per_second = 3;
+  int32 tokens_generated = 4;
+  int32 prompt_tokens_processed = 5;
 }
 
 message RerankRequest {
@@ -133,6 +147,9 @@ message PredictOptions {
   repeated string Images = 42;
   bool UseTokenizerTemplate = 43;
   repeated Message Messages = 44;
+  repeated string Videos = 45;
+  repeated string Audios = 46;
+  string CorrelationId = 47;
 }
 
 // The response message containing the result
@@ -270,6 +287,17 @@ message TTSRequest {
   optional string language = 5;
 }
 
+message SoundGenerationRequest {
+  string text = 1;
+  string model = 2;
+  string dst = 3;
+  optional float duration = 4;
+  optional float temperature = 5;
+  optional bool sample = 6;
+  optional string src = 7;
+  optional int32 src_divisor = 8;
+}
+
 message TokenizationResponse {
   int32 length = 1;
   repeated int32 tokens = 2;

backend/cpp/llama/grpc-server.cpp

@@ -13,15 +13,15 @@
 #include <getopt.h>
 #include "clip.h"
 #include "llava.h"
+#include "log.h"
 #include "stb_image.h"
 #include "common.h"
 #include "json.hpp"
 #include "llama.h"
-#include "grammar-parser.h"
 #include "backend.pb.h"
 #include "backend.grpc.pb.h"
 #include "utils.hpp"
-
+#include "sampling.h"
 // include std::regex
 #include <cstddef>
 #include <thread>
@@ -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 += llama_token_to_piece(ctx, *begin);
+        ret += common_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 ? "" : llama_token_to_piece(ctx, token);
+    std::string out = token == -1 ? "" : common_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 llama_sampling_params sparams;
-    llama_sampling_context *ctx_sampling = nullptr;
+    struct common_sampler_params sparams;
+    common_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(gpt_params &global_params) {
+    bool has_budget(common_params &global_params) {
         if (params.n_predict == -1 && global_params.n_predict == -1)
         {
             return true; // limitless
@@ -398,7 +398,7 @@ struct llama_server_context
 
     clip_ctx *clp_ctx = nullptr;
 
-    gpt_params params;
+    common_params params;
 
     llama_batch batch;
 
@@ -441,7 +441,7 @@ struct llama_server_context
         }
     }
 
-    bool load_model(const gpt_params &params_)
+    bool load_model(const common_params &params_)
     {
         params = params_;
         if (!params.mmproj.empty()) {
@@ -449,7 +449,7 @@ struct llama_server_context
             LOG_INFO("Multi Modal Mode Enabled", {});
             clp_ctx = clip_model_load(params.mmproj.c_str(), /*verbosity=*/ 1);
             if(clp_ctx == nullptr) {
-                LOG_ERROR("unable to load clip model", {{"model", params.mmproj}});
+                LOG_ERR("unable to load clip model: %s", params.mmproj.c_str());
                 return false;
             }
 
@@ -458,12 +458,12 @@ struct llama_server_context
             }
         }
 
-        llama_init_result llama_init = llama_init_from_gpt_params(params);
-        model = llama_init.model;
-        ctx = llama_init.context;
+        common_init_result common_init = common_init_from_params(params);
+        model = common_init.model;
+        ctx = common_init.context;
         if (model == nullptr)
         {
-            LOG_ERROR("unable to load model", {{"model", params.model}});
+            LOG_ERR("unable to load model: %s", params.model.c_str());
             return false;
         }
 
@@ -471,7 +471,7 @@ struct llama_server_context
             const int n_embd_clip = clip_n_mmproj_embd(clp_ctx);
             const int n_embd_llm  = llama_n_embd(model);
             if (n_embd_clip != n_embd_llm) {
-                LOG_TEE("%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);
+                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);
                 llama_free_model(model);
                 return false;
@@ -490,11 +490,21 @@ struct llama_server_context
         std::vector<char> buf(1);
         int res = llama_chat_apply_template(model, nullptr, chat, 1, true, buf.data(), buf.size());
         if (res < 0) {
-            LOG_ERROR("The chat template comes with this model is not yet supported, falling back to chatml. This may cause the model to output suboptimal responses", {});
+            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
+            if (slot.is_processing()) {
+                return &slot;  // Return the active slot
+            }
+        }
+        return nullptr;  // No active slot found
+    }
+
     void initialize() {
         // create slots
         all_slots_are_idle = true;
@@ -568,12 +578,12 @@ struct llama_server_context
                     std::vector<llama_token> p;
                     if (first)
                     {
-                        p = ::llama_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
+                        p = common_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
                         first = false;
                     }
                     else
                     {
-                        p = ::llama_tokenize(ctx, s, false, TMP_FORCE_SPECIAL);
+                        p = common_tokenize(ctx, s, false, TMP_FORCE_SPECIAL);
                     }
                     prompt_tokens.insert(prompt_tokens.end(), p.begin(), p.end());
                 }
@@ -590,7 +600,7 @@ struct llama_server_context
         else
         {
             auto s = json_prompt.template get<std::string>();
-            prompt_tokens = ::llama_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
+            prompt_tokens = common_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
         }
 
         return prompt_tokens;
@@ -619,7 +629,7 @@ struct llama_server_context
 
     bool launch_slot_with_data(llama_client_slot* &slot, json data) {
         slot_params default_params;
-        llama_sampling_params 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);
@@ -628,7 +638,7 @@ struct llama_server_context
         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.typical_p         = json_value(data, "typical_p",         default_sparams.typical_p);
+        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);
         slot->sparams.dynatemp_exponent = json_value(data, "dynatemp_exponent", default_sparams.dynatemp_exponent);
@@ -641,7 +651,7 @@ struct llama_server_context
         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.seed               = json_value(data, "seed",              default_sparams.seed);
         slot->sparams.grammar           = json_value(data, "grammar",           default_sparams.grammar);
         slot->sparams.n_probs           = json_value(data, "n_probs",           default_sparams.n_probs);
         slot->sparams.min_keep          = json_value(data, "min_keep",          default_sparams.min_keep);
@@ -665,6 +675,7 @@ struct llama_server_context
             slot->params.input_prefix = "";
         }
 
+
         if (data.count("input_suffix") != 0)
         {
             slot->params.input_suffix = data["input_suffix"];
@@ -683,6 +694,10 @@ struct llama_server_context
             slot->prompt = "";
         }
 
+        if (json_value(data, "ignore_eos", false)) {
+                slot->sparams.logit_bias.push_back({llama_token_eos(model), -INFINITY});
+        }
+        /*
         slot->sparams.penalty_prompt_tokens.clear();
         slot->sparams.use_penalty_prompt_tokens = false;
         const auto &penalty_prompt = data.find("penalty_prompt");
@@ -718,14 +733,10 @@ struct llama_server_context
                 slot->sparams.use_penalty_prompt_tokens = true;
             }
         }
+      */
 
         slot->sparams.logit_bias.clear();
 
-        if (json_value(data, "ignore_eos", false))
-        {
-            slot->sparams.logit_bias[llama_token_eos(model)] = -INFINITY;
-        }
-
         const auto &logit_bias = data.find("logit_bias");
         if (logit_bias != data.end() && logit_bias->is_array())
         {
@@ -753,21 +764,21 @@ struct llama_server_context
                         llama_token tok = el[0].get<llama_token>();
                         if (tok >= 0 && tok < n_vocab)
                         {
-                            slot->sparams.logit_bias[tok] = bias;
+                            slot->sparams.logit_bias.push_back({tok, bias});
                         }
                     }
                     else if (el[0].is_string())
                     {
-                        auto toks = llama_tokenize(model, 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[tok] = bias;
+                            slot->sparams.logit_bias.push_back({tok, bias});
                         }
                     }
                 }
             }
         }
-
+        
         slot->params.antiprompt.clear();
 
         const auto &stop = data.find("stop");
@@ -781,24 +792,22 @@ struct llama_server_context
                 }
             }
         }
-
-        const auto &samplers_sequence = data.find("samplers");
-        if (samplers_sequence != data.end() && samplers_sequence->is_array())
-        {
+        
+        const auto & samplers = data.find("samplers");
+        if (samplers != data.end() && samplers->is_array()) {
             std::vector<std::string> sampler_names;
-            for (const auto &sampler_name : *samplers_sequence)
-            {
-                if (sampler_name.is_string())
-                {
-                    sampler_names.emplace_back(sampler_name);
+                for (const auto & name : *samplers) {
+                    if (name.is_string()) {
+                        sampler_names.emplace_back(name);
+                    }
                 }
-            }
-            slot->sparams.samplers_sequence = llama_sampling_types_from_names(sampler_names, false);
+                slot->sparams.samplers = common_sampler_types_from_names(sampler_names, false);
         }
         else
         {
-            slot->sparams.samplers_sequence = default_sparams.samplers_sequence;
+                slot->sparams.samplers = default_sparams.samplers;
         }
+        
 
         if (multimodal)
         {
@@ -814,10 +823,11 @@ struct llama_server_context
                     img_sl.img_data = clip_image_u8_init();
                     if (!clip_image_load_from_bytes(image_buffer.data(), image_buffer.size(), img_sl.img_data))
                     {
-                        LOG_ERROR("failed to load image", {
-                            {"slot_id",   slot->id},
-                            {"img_sl_id", img_sl.id}
-                        });
+                        LOG_ERR("%s: failed to load image, slot_id: %d, img_sl_id: %d", 
+                             __func__,
+                             slot->id,
+                             img_sl.id
+                        );
                         return false;
                     }
                     LOG_VERBOSE("image loaded", {
@@ -855,12 +865,12 @@ struct llama_server_context
                                     }
                                 }
                                 if (!found) {
-                                    LOG_TEE("ERROR: Image with id: %i, not found.\n", img_id);
+                                    LOG("ERROR: Image with id: %i, not found.\n", img_id);
                                     slot->images.clear();
                                     return false;
                                 }
                             } catch (const std::invalid_argument& e) {
-                                LOG_TEE("Invalid image number id in prompt\n");
+                                LOG("Invalid image number id in prompt\n");
                                 slot->images.clear();
                                 return false;
                             }
@@ -875,10 +885,10 @@ struct llama_server_context
 
         if (slot->ctx_sampling != nullptr)
         {
-            llama_sampling_free(slot->ctx_sampling);
+            common_sampler_free(slot->ctx_sampling);
         }
-        slot->ctx_sampling = llama_sampling_init(slot->sparams);
-        llama_set_rng_seed(ctx, slot->params.seed);
+        slot->ctx_sampling = common_sampler_init(model, slot->sparams);
+        //llama_set_rng_seed(ctx, slot->params.seed);
         slot->command = LOAD_PROMPT;
 
         all_slots_are_idle = false;
@@ -888,7 +898,7 @@ struct llama_server_context
             {"task_id", slot->task_id},
         });
 
-        LOG_TEE("sampling: \n%s\n", llama_sampling_print(slot->sparams).c_str());
+      //  LOG("sampling: \n%s\n", llama_sampling_print(slot->sparams).c_str());
 
         return true;
     }
@@ -904,13 +914,13 @@ struct llama_server_context
         system_tokens.clear();
 
         if (!system_prompt.empty()) {
-            system_tokens = ::llama_tokenize(ctx, system_prompt, add_bos_token);
+            system_tokens = common_tokenize(ctx, system_prompt, add_bos_token);
 
-            llama_batch_clear(batch);
+            common_batch_clear(batch);
 
             for (int i = 0; i < (int)system_tokens.size(); ++i)
             {
-                llama_batch_add(batch, system_tokens[i], i, { 0 }, false);
+                common_batch_add(batch, system_tokens[i], i, { 0 }, false);
             }
 
             for (int32_t i = 0; i < (int32_t) batch.n_tokens; i += params.n_batch)
@@ -928,7 +938,7 @@ struct llama_server_context
                 };
                 if (llama_decode(ctx, batch_view) != 0)
                 {
-                    LOG_TEE("%s: llama_decode() failed\n", __func__);
+                    LOG("%s: llama_decode() failed\n", __func__);
                     return;
                 }
             }
@@ -940,7 +950,7 @@ struct llama_server_context
             }
         }
 
-        LOG_TEE("system prompt updated\n");
+        LOG("system prompt updated\n");
         system_need_update = false;
     }
 
@@ -999,18 +1009,20 @@ 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 = llama_token_to_piece(ctx, result.tok);
+        const std::string token_str = common_token_to_piece(ctx, result.tok);
         slot.sampled = result.tok;
 
         // search stop word and delete it
         slot.generated_text += token_str;
         slot.has_next_token = true;
 
+/*
         if (slot.ctx_sampling->params.use_penalty_prompt_tokens && result.tok != -1)
         {
             // we can change penalty_prompt_tokens because it is always created from scratch each request
             slot.ctx_sampling->params.penalty_prompt_tokens.push_back(result.tok);
         }
+        */
 
         // check if there is incomplete UTF-8 character at the end
         bool incomplete = false;
@@ -1119,8 +1131,8 @@ struct llama_server_context
                 continue;
             }
 
-            if (!llava_image_embed_make_with_clip_img(clp_ctx, params.n_threads, img.img_data, &img.image_embedding, &img.image_tokens)) {
-                LOG_TEE("Error processing the given image");
+            if (!llava_image_embed_make_with_clip_img(clp_ctx, params.cpuparams.n_threads, img.img_data, &img.image_embedding, &img.image_tokens)) {
+                LOG("Error processing the given image");
                 return false;
             }
 
@@ -1132,7 +1144,7 @@ struct llama_server_context
 
     void send_error(task_server& task, const std::string &error)
     {
-        LOG_TEE("task %i - error: %s\n", task.id, error.c_str());
+        LOG("task %i - error: %s\n", task.id, error.c_str());
         task_result res;
         res.id = task.id;
         res.multitask_id = task.multitask_id;
@@ -1144,13 +1156,11 @@ struct llama_server_context
 
     json get_formated_generation(llama_client_slot &slot)
     {
-        const auto eos_bias = slot.sparams.logit_bias.find(llama_token_eos(model));
-        const bool ignore_eos = eos_bias != slot.sparams.logit_bias.end() &&
-                                eos_bias->second < 0.0f && std::isinf(eos_bias->second);
-        std::vector<std::string> samplers_sequence;
-        for (const auto &sampler_type : slot.sparams.samplers_sequence)
+        std::vector<std::string> samplers;
+        samplers.reserve(slot.sparams.samplers.size());
+        for (const auto & sampler : slot.sparams.samplers)
         {
-            samplers_sequence.emplace_back(llama_sampling_type_to_str(sampler_type));
+            samplers.emplace_back(common_sampler_type_to_str(sampler));
         }
 
         return json {
@@ -1165,13 +1175,11 @@ struct llama_server_context
             {"top_p",             slot.sparams.top_p},
             {"min_p",             slot.sparams.min_p},
             {"tfs_z",             slot.sparams.tfs_z},
-            {"typical_p",         slot.sparams.typical_p},
+            {"typical_p",         slot.sparams.typ_p},
             {"repeat_last_n",     slot.sparams.penalty_last_n},
             {"repeat_penalty",    slot.sparams.penalty_repeat},
             {"presence_penalty",  slot.sparams.penalty_present},
             {"frequency_penalty", slot.sparams.penalty_freq},
-            {"penalty_prompt_tokens", slot.sparams.penalty_prompt_tokens},
-            {"use_penalty_prompt_tokens", slot.sparams.use_penalty_prompt_tokens},
             {"mirostat",          slot.sparams.mirostat},
             {"mirostat_tau",      slot.sparams.mirostat_tau},
             {"mirostat_eta",      slot.sparams.mirostat_eta},
@@ -1179,13 +1187,13 @@ struct llama_server_context
             {"stop",              slot.params.antiprompt},
             {"n_predict",         slot.params.n_predict},
             {"n_keep",            params.n_keep},
-            {"ignore_eos",        ignore_eos},
+            {"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},
-            {"samplers",          samplers_sequence}
+            {"samplers",          samplers}
         };
     }
 
@@ -1208,7 +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 = llama_tokenize(ctx, tkn.text_to_send, false);
+            const std::vector<llama_token> to_send_toks = common_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)
@@ -1260,7 +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 = llama_tokenize(ctx, slot.stopping_word, false);
+                const std::vector<llama_token> stop_word_toks = common_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
@@ -1375,7 +1383,7 @@ struct llama_server_context
                 };
                 if (llama_decode(ctx, batch_view))
                 {
-                    LOG_TEE("%s : failed to eval\n", __func__);
+                    LOG("%s : failed to eval\n", __func__);
                     return false;
                 }
             }
@@ -1393,14 +1401,14 @@ struct llama_server_context
                 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_TEE("%s : failed to eval image\n", __func__);
+                    LOG("%s : failed to eval image\n", __func__);
                     return false;
                 }
                 slot.n_past += n_eval;
             }
             image_idx++;
 
-            llama_batch_clear(batch);
+            common_batch_clear(batch);
 
             // append prefix of next image
             const auto json_prompt = (image_idx >= (int) slot.images.size()) ?
@@ -1410,7 +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)
             {
-                llama_batch_add(batch, append_tokens[i], system_tokens.size() + slot.n_past, { slot.id }, true);
+                common_batch_add(batch, append_tokens[i], system_tokens.size() + slot.n_past, { slot.id }, true);
                 slot.n_past += 1;
             }
         }
@@ -1542,7 +1550,7 @@ struct llama_server_context
             update_system_prompt();
         }
 
-        llama_batch_clear(batch);
+        common_batch_clear(batch);
 
         if (all_slots_are_idle)
         {
@@ -1576,7 +1584,7 @@ struct llama_server_context
                     slot.n_past = 0;
                     slot.truncated = false;
                     slot.has_next_token = true;
-                    LOG_TEE("Context exhausted. Slot %d released (%d tokens in cache)\n", slot.id, (int) slot.cache_tokens.size());
+                    LOG("Context exhausted. Slot %d released (%d tokens in cache)\n", slot.id, (int) slot.cache_tokens.size());
 
                     continue;
                     // END LOCALAI changes
@@ -1620,7 +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
-            llama_batch_add(batch, slot.sampled, system_tokens.size() + slot_npast, { slot.id }, true);
+            common_batch_add(batch, slot.sampled, system_tokens.size() + slot_npast, { slot.id }, true);
             slot.n_past += 1;
         }
 
@@ -1714,7 +1722,7 @@ struct llama_server_context
 
                     if (!slot.params.cache_prompt)
                     {
-                        llama_sampling_reset(slot.ctx_sampling);
+                        common_sampler_reset(slot.ctx_sampling);
 
                         slot.n_past = 0;
                         slot.n_past_se = 0;
@@ -1726,7 +1734,7 @@ struct llama_server_context
                         // push the prompt into the sampling context (do not apply grammar)
                         for (auto &token : prompt_tokens)
                         {
-                            llama_sampling_accept(slot.ctx_sampling, ctx, token, false);
+                            common_sampler_accept(slot.ctx_sampling, token, false);
                         }
 
                         slot.n_past = common_part(slot.cache_tokens, prompt_tokens);
@@ -1818,16 +1826,17 @@ struct llama_server_context
                                 ga_i += ga_w/ga_n;
                             }
                         }
-                        llama_batch_add(batch, prefix_tokens[slot.n_past], system_tokens.size() + slot_npast, {slot.id }, false);
+                        common_batch_add(batch, prefix_tokens[slot.n_past], system_tokens.size() + slot_npast, {slot.id }, false);
                         slot_npast++;
                     }
 
                     if (has_images && !ingest_images(slot, n_batch))
                     {
-                        LOG_ERROR("failed processing images", {
-                            "slot_id", slot.id,
-                            "task_id", slot.task_id,
-                        });
+                        LOG_ERR("%s: failed processing images Slot id : %d, Task id: %d", 
+                            __func__,
+                            slot.id,
+                            slot.task_id
+                        );
                         // FIXME @phymbert: to be properly tested
                         //  early returning without changing the slot state will block the slot for ever
                         // no one at the moment is checking the return value
@@ -1867,10 +1876,10 @@ struct llama_server_context
                         const int bd = (slot.ga_w / slot.ga_n) * (slot.ga_n - 1);
                         const int dd = (slot.ga_w / slot.ga_n) - ib * bd - slot.ga_w;
 
-                        LOG_TEE("\n");
-                        LOG_TEE("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i, slot.n_past_se, ib * bd, slot.ga_i + ib * bd, slot.n_past_se + ib * bd);
-                        LOG_TEE("div:   [%6d, %6d] / %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd, slot.ga_i + ib * bd + slot.ga_w, slot.ga_n, (slot.ga_i + ib * bd) / slot.ga_n, (slot.ga_i + ib * bd + slot.ga_w) / slot.ga_n);
-                        LOG_TEE("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd + slot.ga_w, slot.n_past_se + ib * bd, dd, slot.ga_i + ib * bd + slot.ga_w + dd, slot.n_past_se + ib * bd + dd);
+                        LOG("\n");
+                        LOG("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i, slot.n_past_se, ib * bd, slot.ga_i + ib * bd, slot.n_past_se + ib * bd);
+                        LOG("div:   [%6d, %6d] / %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd, slot.ga_i + ib * bd + slot.ga_w, slot.ga_n, (slot.ga_i + ib * bd) / slot.ga_n, (slot.ga_i + ib * bd + slot.ga_w) / slot.ga_n);
+                        LOG("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd + slot.ga_w, slot.n_past_se + ib * bd, dd, slot.ga_i + ib * bd + slot.ga_w + dd, slot.n_past_se + ib * bd + dd);
 
                         llama_kv_cache_seq_add(ctx, slot.id, slot.ga_i, slot.n_past_se, ib * bd);
                         llama_kv_cache_seq_div(ctx, slot.id, slot.ga_i + ib * bd, slot.ga_i + ib * bd + slot.ga_w,slot.ga_n);
@@ -1880,7 +1889,7 @@ struct llama_server_context
 
                         slot.ga_i += slot.ga_w / slot.ga_n;
 
-                        LOG_TEE("\nn_past_old = %d, n_past = %d, ga_i = %d\n\n", slot.n_past_se + bd, slot.n_past_se, slot.ga_i);
+                        LOG("\nn_past_old = %d, n_past = %d, ga_i = %d\n\n", slot.n_past_se + bd, slot.n_past_se, slot.ga_i);
                     }
                     slot.n_past_se += n_tokens;
                 }
@@ -1905,11 +1914,11 @@ struct llama_server_context
                 if (n_batch == 1 || ret < 0)
                 {
                     // if you get here, it means the KV cache is full - try increasing it via the context size
-                    LOG_TEE("%s : failed to decode the batch, n_batch = %d, ret = %d\n", __func__, n_batch, ret);
+                    LOG("%s : failed to decode the batch, n_batch = %d, ret = %d\n", __func__, n_batch, ret);
                     return false;
                 }
 
-                LOG_TEE("%s : failed to find free space in the KV cache, retrying with smaller n_batch = %d\n", __func__, n_batch / 2);
+                LOG("%s : failed to find free space in the KV cache, retrying with smaller n_batch = %d\n", __func__, n_batch / 2);
 
                 // retry with half the batch size to try to find a free slot in the KV cache
                 n_batch /= 2;
@@ -1934,9 +1943,9 @@ struct llama_server_context
                 }
 
                 completion_token_output result;
-                const llama_token id = llama_sampling_sample(slot.ctx_sampling, ctx, NULL, slot.i_batch - i);
+                const llama_token id = common_sampler_sample(slot.ctx_sampling, ctx, slot.i_batch - i);
 
-                llama_sampling_accept(slot.ctx_sampling, ctx, id, true);
+                common_sampler_accept(slot.ctx_sampling, id, true);
 
                 slot.n_decoded += 1;
                 if (slot.n_decoded == 1)
@@ -1946,19 +1955,14 @@ struct llama_server_context
                     metrics.on_prompt_eval(slot);
                 }
 
-                llama_token_data_array cur_p = { slot.ctx_sampling->cur.data(), slot.ctx_sampling->cur.size(), false };
                 result.tok = id;
+                const auto * cur_p = common_sampler_get_candidates(slot.ctx_sampling);
 
-                const int32_t n_probs = slot.sparams.n_probs;
-                if (slot.sparams.temp <= 0 && n_probs > 0)
-                {
-                    // for llama_sample_token_greedy we need to sort candidates
-                    llama_sample_softmax(ctx, &cur_p);
-                }
-
-                for (size_t i = 0; i < std::min(cur_p.size, (size_t)n_probs); ++i)
-                {
-                    result.probs.push_back({cur_p.data[i].id, cur_p.data[i].p});
+                for (size_t i = 0; i < (size_t) slot.sparams.n_probs; ++i) {
+                    result.probs.push_back({
+                        cur_p->data[i].id,
+                        i >= cur_p->size ? 0.0f : cur_p->data[i].p,
+                    });
                 }
 
                 if (!process_token(result, slot))
@@ -2005,7 +2009,7 @@ static json format_partial_response(
 struct token_translator
 {
     llama_context * ctx;
-    std::string operator()(llama_token tok)                    const { return llama_token_to_piece(ctx, tok); }
+    std::string operator()(llama_token tok)                    const { return common_token_to_piece(ctx, tok); }
     std::string operator()(const completion_token_output &cto) const { return (*this)(cto.tok); }
 };
 
@@ -2112,6 +2116,9 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
     data["ignore_eos"] = predict->ignoreeos();
     data["embeddings"] = predict->embeddings();
 
+    // Add the correlationid to json data
+    data["correlation_id"] = predict->correlationid();
+
     // for each image in the request, add the image data
     //
     for (int i = 0; i < predict->images_size(); i++) {
@@ -2196,7 +2203,7 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
 // }
 
 static void params_parse(const backend::ModelOptions* request,
-                                gpt_params & params) {
+                                common_params & params) {
    
     // this is comparable to: https://github.com/ggerganov/llama.cpp/blob/d9b33fe95bd257b36c84ee5769cc048230067d6f/examples/server/server.cpp#L1809
 
@@ -2210,7 +2217,7 @@ static void params_parse(const backend::ModelOptions* request,
     params.model_alias =  request->modelfile();
     params.n_ctx = request->contextsize();
     //params.memory_f16 = request->f16memory();
-    params.n_threads = request->threads();
+    params.cpuparams.n_threads = request->threads();
     params.n_gpu_layers = request->ngpulayers();
     params.n_batch = request->nbatch();
     // Set params.n_parallel by environment variable (LLAMA_PARALLEL), defaults to 1
@@ -2304,7 +2311,7 @@ public:
 
   grpc::Status LoadModel(ServerContext* context, const backend::ModelOptions* request, backend::Result* result) {
     // Implement LoadModel RPC
-    gpt_params params;
+    common_params params;
     params_parse(request, params);
 
     llama_backend_init();
@@ -2350,6 +2357,11 @@ public:
                 int32_t tokens_evaluated = result.result_json.value("tokens_evaluated", 0);
                 reply.set_prompt_tokens(tokens_evaluated);
 
+                // Log Request Correlation Id
+                LOG_VERBOSE("correlation:", {
+                    { "id", data["correlation_id"] }
+                });
+
                 // Send the reply
                 writer->Write(reply);
 
@@ -2373,6 +2385,12 @@ public:
         std::string completion_text;
         task_result result = llama.queue_results.recv(task_id);
         if (!result.error && result.stop) {
+            
+            // Log Request Correlation Id
+            LOG_VERBOSE("correlation:", {
+                { "id", data["correlation_id"] }
+            });
+
             completion_text = result.result_json.value("content", "");
             int32_t tokens_predicted = result.result_json.value("tokens_predicted", 0);
             int32_t tokens_evaluated = result.result_json.value("tokens_evaluated", 0);
@@ -2412,6 +2430,31 @@ public:
 
         return grpc::Status::OK;
     }
+
+    grpc::Status GetMetrics(ServerContext* context, const backend::MetricsRequest* request, backend::MetricsResponse* response) {
+        llama_client_slot* active_slot = llama.get_active_slot();
+
+        if (active_slot != nullptr) {
+            // Calculate the tokens per second using existing logic
+            double tokens_per_second = 1e3 / active_slot->t_token_generation * active_slot->n_decoded;
+
+            // Populate the response with metrics
+            response->set_slot_id(active_slot->id);
+            response->set_prompt_json_for_slot(active_slot->prompt.dump());
+            response->set_tokens_per_second(tokens_per_second);
+            response->set_tokens_generated(active_slot->n_decoded);
+            response->set_prompt_tokens_processed(active_slot->num_prompt_tokens_processed);
+        } else {
+            // Handle case when no active slot exists
+            response->set_slot_id(0);
+            response->set_prompt_json_for_slot("");
+            response->set_tokens_per_second(0);
+            response->set_tokens_generated(0);
+            response->set_prompt_tokens_processed(0);
+        }
+
+        return grpc::Status::OK;
+    } 
 };
 
 void RunServer(const std::string& server_address) {

backend/cpp/llama/patches/01-llava.patch

@@ -0,0 +1,13 @@
+diff --git a/examples/llava/clip.cpp b/examples/llava/clip.cpp
+index 342042ff..224db9b5 100644
+--- a/examples/llava/clip.cpp
++++ b/examples/llava/clip.cpp
+@@ -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);

backend/cpp/llama/prepare.sh

@@ -1,5 +1,12 @@
 #!/bin/bash
 
+## Patches
+## Apply patches from the `patches` directory
+for patch in $(ls patches); do
+    echo "Applying patch $patch"
+    patch -d llama.cpp/ -p1 < patches/$patch
+done 
+
 cp -r CMakeLists.txt llama.cpp/examples/grpc-server/
 cp -r grpc-server.cpp llama.cpp/examples/grpc-server/
 cp -rfv json.hpp llama.cpp/examples/grpc-server/

backend/cpp/llama/utils.hpp

@@ -480,31 +480,4 @@ static inline std::vector<uint8_t> base64_decode(const std::string & encoded_str
     }
 
     return ret;
-}
-
-//
-// random string / id
-//
-
-static std::string random_string()
-{
-    static const std::string str("0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz");
-
-    std::random_device rd;
-    std::mt19937 generator(rd());
-
-    std::string result(32, ' ');
-
-    for (int i = 0; i < 32; ++i) {
-        result[i] = str[generator() % str.size()];
-    }
-
-    return result;
-}
-
-static std::string gen_chatcmplid()
-{
-    std::stringstream chatcmplid;
-    chatcmplid << "chatcmpl-" << random_string();
-    return chatcmplid.str();
 }

backend/go/transcribe/transcript.go

@@ -1,104 +0,0 @@
-package main
-
-import (
-	"fmt"
-	"os"
-	"os/exec"
-	"path/filepath"
-
-	"github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
-	"github.com/go-audio/wav"
-	"github.com/mudler/LocalAI/core/schema"
-)
-
-func ffmpegCommand(args []string) (string, error) {
-	cmd := exec.Command("ffmpeg", args...) // Constrain this to ffmpeg to permit security scanner to see that the command is safe.
-	cmd.Env = os.Environ()
-	out, err := cmd.CombinedOutput()
-	return string(out), err
-}
-
-// AudioToWav converts audio to wav for transcribe.
-// TODO: use https://github.com/mccoyst/ogg?
-func audioToWav(src, dst string) error {
-	commandArgs := []string{"-i", src, "-format", "s16le", "-ar", "16000", "-ac", "1", "-acodec", "pcm_s16le", dst}
-	out, err := ffmpegCommand(commandArgs)
-	if err != nil {
-		return fmt.Errorf("error: %w out: %s", err, out)
-	}
-	return nil
-}
-
-func Transcript(model whisper.Model, audiopath, language string, translate bool, threads uint) (schema.TranscriptionResult, error) {
-	res := schema.TranscriptionResult{}
-
-	dir, err := os.MkdirTemp("", "whisper")
-	if err != nil {
-		return res, err
-	}
-	defer os.RemoveAll(dir)
-
-	convertedPath := filepath.Join(dir, "converted.wav")
-
-	if err := audioToWav(audiopath, convertedPath); err != nil {
-		return res, err
-	}
-
-	// Open samples
-	fh, err := os.Open(convertedPath)
-	if err != nil {
-		return res, err
-	}
-	defer fh.Close()
-
-	// Read samples
-	d := wav.NewDecoder(fh)
-	buf, err := d.FullPCMBuffer()
-	if err != nil {
-		return res, err
-	}
-
-	data := buf.AsFloat32Buffer().Data
-
-	// Process samples
-	context, err := model.NewContext()
-	if err != nil {
-		return res, err
-
-	}
-
-	context.SetThreads(threads)
-
-	if language != "" {
-		context.SetLanguage(language)
-	} else {
-		context.SetLanguage("auto")
-	}
-
-	if translate {
-		context.SetTranslate(true)
-	}
-
-	if err := context.Process(data, nil, nil); err != nil {
-		return res, err
-	}
-
-	for {
-		s, err := context.NextSegment()
-		if err != nil {
-			break
-		}
-
-		var tokens []int
-		for _, t := range s.Tokens {
-			tokens = append(tokens, t.Id)
-		}
-
-		segment := schema.Segment{Id: s.Num, Text: s.Text, Start: s.Start, End: s.End, Tokens: tokens}
-		res.Segments = append(res.Segments, segment)
-
-		res.Text += s.Text
-	}
-
-	return res, nil
-}

backend/go/transcribe/whisper.go

@@ -1,26 +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 (
-	"github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
-	"github.com/mudler/LocalAI/core/schema"
-	"github.com/mudler/LocalAI/pkg/grpc/base"
-	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
-)
-
-type Whisper struct {
-	base.SingleThread
-	whisper whisper.Model
-}
-
-func (sd *Whisper) Load(opts *pb.ModelOptions) error {
-	// Note: the Model here is a path to a directory containing the model files
-	w, err := whisper.New(opts.ModelFile)
-	sd.whisper = w
-	return err
-}
-
-func (sd *Whisper) AudioTranscription(opts *pb.TranscriptRequest) (schema.TranscriptionResult, error) {
-	return Transcript(sd.whisper, opts.Dst, opts.Language, opts.Translate, uint(opts.Threads))
-}

backend/go/transcribe/whisper/whisper.go

@@ -0,0 +1,105 @@
+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 (
+	"os"
+	"path/filepath"
+
+	"github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
+	"github.com/go-audio/wav"
+	"github.com/mudler/LocalAI/pkg/grpc/base"
+	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
+	"github.com/mudler/LocalAI/pkg/utils"
+)
+
+type Whisper struct {
+	base.SingleThread
+	whisper whisper.Model
+}
+
+func (sd *Whisper) Load(opts *pb.ModelOptions) error {
+	// Note: the Model here is a path to a directory containing the model files
+	w, err := whisper.New(opts.ModelFile)
+	sd.whisper = w
+	return err
+}
+
+func (sd *Whisper) AudioTranscription(opts *pb.TranscriptRequest) (pb.TranscriptResult, error) {
+
+	dir, err := os.MkdirTemp("", "whisper")
+	if err != nil {
+		return pb.TranscriptResult{}, err
+	}
+	defer os.RemoveAll(dir)
+
+	convertedPath := filepath.Join(dir, "converted.wav")
+
+	if err := utils.AudioToWav(opts.Dst, convertedPath); err != nil {
+		return pb.TranscriptResult{}, err
+	}
+
+	// Open samples
+	fh, err := os.Open(convertedPath)
+	if err != nil {
+		return pb.TranscriptResult{}, err
+	}
+	defer fh.Close()
+
+	// Read samples
+	d := wav.NewDecoder(fh)
+	buf, err := d.FullPCMBuffer()
+	if err != nil {
+		return pb.TranscriptResult{}, err
+	}
+
+	data := buf.AsFloat32Buffer().Data
+
+	// Process samples
+	context, err := sd.whisper.NewContext()
+	if err != nil {
+		return pb.TranscriptResult{}, err
+
+	}
+
+	context.SetThreads(uint(opts.Threads))
+
+	if opts.Language != "" {
+		context.SetLanguage(opts.Language)
+	} else {
+		context.SetLanguage("auto")
+	}
+
+	if opts.Translate {
+		context.SetTranslate(true)
+	}
+
+	if err := context.Process(data, nil, nil); err != nil {
+		return pb.TranscriptResult{}, err
+	}
+
+	segments := []*pb.TranscriptSegment{}
+	text := ""
+	for {
+		s, err := context.NextSegment()
+		if err != nil {
+			break
+		}
+
+		var tokens []int32
+		for _, t := range s.Tokens {
+			tokens = append(tokens, int32(t.Id))
+		}
+
+		segment := &pb.TranscriptSegment{Id: int32(s.Num), Text: s.Text, Start: int64(s.Start), End: int64(s.End), Tokens: tokens}
+		segments = append(segments, segment)
+
+		text += s.Text
+	}
+
+	return pb.TranscriptResult{
+		Segments: segments,
+		Text:     text,
+	}, nil
+
+}

backend/python/autogptq/requirements-intel.txt

@@ -2,4 +2,4 @@
 intel-extension-for-pytorch
 torch
 optimum[openvino]
-setuptools==72.1.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406

backend/python/autogptq/requirements.txt

@@ -1,6 +1,6 @@
 accelerate
 auto-gptq==0.7.1
-grpcio==1.65.4
+grpcio==1.66.2
 protobuf
 certifi
 transformers

backend/python/bark/requirements-intel.txt

@@ -3,6 +3,6 @@ intel-extension-for-pytorch
 torch
 torchaudio
 optimum[openvino]
-setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
 transformers
 accelerate

backend/python/bark/requirements.txt

@@ -1,4 +1,4 @@
 bark==0.1.5
-grpcio==1.65.5
+grpcio==1.66.2
 protobuf
 certifi

backend/python/common/template/requirements.txt

@@ -1,2 +1,2 @@
-grpcio==1.65.5
+grpcio==1.66.2
 protobuf

backend/python/coqui/requirements-intel.txt

@@ -3,6 +3,6 @@ intel-extension-for-pytorch
 torch
 torchaudio
 optimum[openvino]
-setuptools==72.1.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
 transformers
 accelerate

backend/python/coqui/requirements.txt

@@ -1,4 +1,4 @@
-TTS==0.22.0
-grpcio==1.65.5
+coqui-tts
+grpcio==1.66.2
 protobuf
 certifi

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(10)
+        time.sleep(30)
 
     def tearDown(self) -> None:
         """

backend/python/diffusers/backend.py

@@ -168,7 +168,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             if request.CFGScale != 0:
                 self.cfg_scale = request.CFGScale
 
-            clipmodel = "runwayml/stable-diffusion-v1-5"
+            clipmodel = "Lykon/dreamshaper-8"
             if request.CLIPModel != "":
                 clipmodel = request.CLIPModel
             clipsubfolder = "text_encoder"

backend/python/diffusers/requirements-intel.txt

@@ -3,7 +3,7 @@ intel-extension-for-pytorch
 torch
 torchvision
 optimum[openvino]
-setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
 diffusers
 opencv-python
 transformers

backend/python/diffusers/requirements.txt

@@ -1,5 +1,5 @@
 setuptools
-grpcio==1.65.4
+grpcio==1.66.2
 pillow
 protobuf
 certifi

backend/python/diffusers/test.py

@@ -53,7 +53,7 @@ class TestBackendServicer(unittest.TestCase):
             self.setUp()
             with grpc.insecure_channel("localhost:50051") as channel:
                 stub = backend_pb2_grpc.BackendStub(channel)
-                response = stub.LoadModel(backend_pb2.ModelOptions(Model="runwayml/stable-diffusion-v1-5"))
+                response = stub.LoadModel(backend_pb2.ModelOptions(Model="Lykon/dreamshaper-8"))
                 self.assertTrue(response.success)
                 self.assertEqual(response.message, "Model loaded successfully")
         except Exception as err:
@@ -71,7 +71,7 @@ class TestBackendServicer(unittest.TestCase):
             self.setUp()
             with grpc.insecure_channel("localhost:50051") as channel:
                 stub = backend_pb2_grpc.BackendStub(channel)
-                response = stub.LoadModel(backend_pb2.ModelOptions(Model="runwayml/stable-diffusion-v1-5"))
+                response = stub.LoadModel(backend_pb2.ModelOptions(Model="Lykon/dreamshaper-8"))
                 print(response.message)
                 self.assertTrue(response.success)
                 image_req = backend_pb2.GenerateImageRequest(positive_prompt="cat", width=16,height=16, dst="test.jpg")
@@ -81,4 +81,4 @@ class TestBackendServicer(unittest.TestCase):
             print(err)
             self.fail("Image gen service failed")
         finally:
-            self.tearDown()
+            self.tearDown()

backend/python/exllama/.gitignore

@@ -1 +0,0 @@
-source

backend/python/exllama/Makefile

@@ -1,25 +0,0 @@
-export CONDA_ENV_PATH = "exllama.yml"
-
-.PHONY: exllama
-exllama: protogen
-	bash install.sh ${CONDA_ENV_PATH}
-
-.PHONY: run
-run: protogen
-	@echo "Running exllama..."
-	bash run.sh
-	@echo "exllama run."
-
-.PHONY: protogen
-protogen: backend_pb2_grpc.py backend_pb2.py
-
-.PHONY: protogen-clean
-protogen-clean:
-	$(RM) backend_pb2_grpc.py backend_pb2.py
-
-backend_pb2_grpc.py backend_pb2.py:
-	python3 -m grpc_tools.protoc -I../.. --python_out=. --grpc_python_out=. backend.proto
-
-.PHONY: clean
-clean: protogen-clean
-	$(RM) -r venv source __pycache__

backend/python/exllama/README.md

@@ -1,5 +0,0 @@
-# Creating a separate environment for the exllama project
-
-```
-make exllama
-```

backend/python/exllama/backend.py

@@ -1,159 +0,0 @@
-#!/usr/bin/env python3
-import grpc
-from concurrent import futures
-import time
-import backend_pb2
-import backend_pb2_grpc
-import argparse
-import signal
-import sys
-import os, glob
-
-from pathlib import Path
-import torch
-import torch.nn.functional as F
-from torch import version as torch_version
-
-from source.tokenizer import ExLlamaTokenizer
-from source.generator import ExLlamaGenerator
-from source.model import ExLlama, ExLlamaCache, ExLlamaConfig
-
-_ONE_DAY_IN_SECONDS = 60 * 60 * 24
-
-# If MAX_WORKERS are specified in the environment use it, otherwise default to 1
-MAX_WORKERS = int(os.environ.get('PYTHON_GRPC_MAX_WORKERS', '1'))
-
-# Implement the BackendServicer class with the service methods
-class BackendServicer(backend_pb2_grpc.BackendServicer):
-    def generate(self,prompt, max_new_tokens):
-        self.generator.end_beam_search()
-
-        # Tokenizing the input
-        ids = self.generator.tokenizer.encode(prompt)
-
-        self.generator.gen_begin_reuse(ids)
-        initial_len = self.generator.sequence[0].shape[0]
-        has_leading_space = False
-        decoded_text = ''
-        for i in range(max_new_tokens):
-            token = self.generator.gen_single_token()
-            if i == 0 and self.generator.tokenizer.tokenizer.IdToPiece(int(token)).startswith('▁'):
-                has_leading_space = True
-
-            decoded_text = self.generator.tokenizer.decode(self.generator.sequence[0][initial_len:])
-            if has_leading_space:
-                decoded_text = ' ' + decoded_text
-
-            if token.item() == self.generator.tokenizer.eos_token_id:
-                break
-        return decoded_text
-    def Health(self, request, context):
-        return backend_pb2.Reply(message=bytes("OK", 'utf-8'))
-    def LoadModel(self, request, context):
-        try:
-            # https://github.com/turboderp/exllama/blob/master/example_cfg.py
-            model_directory = request.ModelFile
-
-            # Locate files we need within that directory
-            tokenizer_path = os.path.join(model_directory, "tokenizer.model")
-            model_config_path = os.path.join(model_directory, "config.json")
-            st_pattern = os.path.join(model_directory, "*.safetensors")
-            model_path = glob.glob(st_pattern)[0]
-
-            # Create config, model, tokenizer and generator
-
-            config = ExLlamaConfig(model_config_path)               # create config from config.json
-            config.model_path = model_path                          # supply path to model weights file
-            if (request.ContextSize):
-                config.max_seq_len = request.ContextSize            # override max sequence length
-                config.max_attention_size = request.ContextSize**2  # Should be set to context_size^2. 
-                # https://github.com/turboderp/exllama/issues/220#issuecomment-1720324163
-
-            # Set Rope scaling.
-            if (request.RopeFreqScale):
-                # Alpha value for Rope scaling. 
-                # Higher value increases context but adds perplexity.
-                # alpha_value and compress_pos_emb are mutually exclusive.
-                # https://github.com/turboderp/exllama/issues/115
-                config.alpha_value = request.RopeFreqScale
-                config.calculate_rotary_embedding_base()
-
-            model = ExLlama(config)                                 # create ExLlama instance and load the weights
-            tokenizer = ExLlamaTokenizer(tokenizer_path)            # create tokenizer from tokenizer model file
-
-            cache = ExLlamaCache(model, batch_size = 2)             # create cache for inference
-            generator = ExLlamaGenerator(model, tokenizer, cache)   # create generator
-
-            self.generator= generator
-            self.model = model
-            self.tokenizer = tokenizer
-            self.cache = cache
-        except Exception as err:
-            return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
-        return backend_pb2.Result(message="Model loaded successfully", success=True)
-
-    def Predict(self, request, context):
-        penalty = 1.15
-        if request.Penalty != 0.0:
-            penalty = request.Penalty
-        self.generator.settings.token_repetition_penalty_max = penalty
-        self.generator.settings.temperature = request.Temperature
-        self.generator.settings.top_k = request.TopK
-        self.generator.settings.top_p = request.TopP
-
-        tokens = 512
-        if request.Tokens != 0:
-            tokens = request.Tokens
-
-        if self.cache.batch_size == 1:
-            del self.cache
-            self.cache = ExLlamaCache(self.model, batch_size=2)
-            self.generator = ExLlamaGenerator(self.model, self.tokenizer, self.cache)
-
-        t = self.generate(request.Prompt, tokens)
-
-        # Remove prompt from response if present
-        if request.Prompt in t:
-            t = t.replace(request.Prompt, "")
-
-        return backend_pb2.Result(message=bytes(t, encoding='utf-8'))
-
-    def PredictStream(self, request, context):
-        # Implement PredictStream RPC
-        #for reply in some_data_generator():
-        #    yield reply
-        # Not implemented yet
-        return self.Predict(request, context)
-
-
-def serve(address):
-    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS))
-    backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
-    server.add_insecure_port(address)
-    server.start()
-    print("Server started. Listening on: " + address, file=sys.stderr)
-
-    # Define the signal handler function
-    def signal_handler(sig, frame):
-        print("Received termination signal. Shutting down...")
-        server.stop(0)
-        sys.exit(0)
-
-    # Set the signal handlers for SIGINT and SIGTERM
-    signal.signal(signal.SIGINT, signal_handler)
-    signal.signal(signal.SIGTERM, signal_handler)
-
-    try:
-        while True:
-            time.sleep(_ONE_DAY_IN_SECONDS)
-    except KeyboardInterrupt:
-        server.stop(0)
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser(description="Run the gRPC server.")
-    parser.add_argument(
-        "--addr", default="localhost:50051", help="The address to bind the server to."
-    )
-    args = parser.parse_args()
-
-    serve(args.addr)

backend/python/exllama/install.sh

@@ -1,13 +0,0 @@
-#!/bin/bash
-set -e
-
-LIMIT_TARGETS="cublas"
-
-source $(dirname $0)/../common/libbackend.sh
-
-installRequirements
-
-git clone https://github.com/turboderp/exllama $MY_DIR/source
-uv pip install ${BUILD_ISOLATION_FLAG} --requirement ${MY_DIR}/source/requirements.txt
-
-cp -v ./*py $MY_DIR/source/

backend/python/exllama/requirements-cpu.txt

@@ -1,3 +0,0 @@
-transformers
-accelerate
-torch

backend/python/exllama/requirements-cublas11.txt

@@ -1,4 +0,0 @@
---extra-index-url https://download.pytorch.org/whl/cu118
-torch
-transformers
-accelerate

backend/python/exllama/requirements-cublas12.txt

@@ -1,3 +0,0 @@
-torch
-transformers
-accelerate

backend/python/exllama/requirements.txt

@@ -1,4 +0,0 @@
-grpcio==1.65.5
-protobuf
-certifi
-setuptools

backend/python/exllama/run.sh

@@ -1,7 +0,0 @@
-#!/bin/bash
-LIMIT_TARGETS="cublas"
-BACKEND_FILE="${MY_DIR}/source/backend.py"
-
-source $(dirname $0)/../common/libbackend.sh
-
-startBackend $@

backend/python/exllama/test.sh

@@ -1,6 +0,0 @@
-#!/bin/bash
-set -e
-
-source $(dirname $0)/../common/libbackend.sh
-
-runUnittests

backend/python/exllama2/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.65.4
+grpcio==1.66.2
 protobuf
 certifi
 wheel

backend/python/mamba/requirements.txt

@@ -1,3 +1,3 @@
-grpcio==1.65.5
+grpcio==1.66.2
 protobuf
 certifi

backend/python/openvoice/requirements-intel.txt

@@ -2,7 +2,7 @@
 intel-extension-for-pytorch
 torch
 optimum[openvino]
-grpcio==1.65.5
+grpcio==1.66.2
 protobuf
 librosa==0.9.1
 faster-whisper==1.0.3
@@ -18,6 +18,6 @@ python-dotenv
 pypinyin==0.50.0
 cn2an==0.5.22
 jieba==0.42.1
-gradio==4.38.1
+gradio==4.44.1
 langid==1.1.6
 git+https://github.com/myshell-ai/MeloTTS.git

backend/python/openvoice/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.65.5
+grpcio==1.66.2
 protobuf
 librosa
 faster-whisper

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(10)
+        time.sleep(30)
 
     def tearDown(self) -> None:
         """

backend/python/parler-tts/install.sh

@@ -15,5 +15,12 @@ installRequirements
 
 # https://github.com/descriptinc/audiotools/issues/101
 # incompatible protobuf versions.
-PYDIR=$(ls ${MY_DIR}/venv/lib)
-curl -L https://raw.githubusercontent.com/protocolbuffers/protobuf/main/python/google/protobuf/internal/builder.py -o ${MY_DIR}/venv/lib/${PYDIR}/site-packages/google/protobuf/internal/builder.py
+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
+
+curl -L https://raw.githubusercontent.com/protocolbuffers/protobuf/main/python/google/protobuf/internal/builder.py -o ${pyenv}/builder.py

backend/python/parler-tts/requirements-hipblas.txt

@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch
-torchaudio
+torch==2.3.0+rocm6.0
+torchaudio==2.3.0+rocm6.0
 transformers
-accelerate
+accelerate

backend/python/parler-tts/requirements-intel.txt

@@ -3,6 +3,6 @@ intel-extension-for-pytorch
 torch
 torchaudio
 optimum[openvino]
-setuptools==72.1.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
 transformers
 accelerate

backend/python/parler-tts/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.65.5
+grpcio==1.66.2
 protobuf
 certifi
 llvmlite==0.43.0

backend/python/rerankers/requirements-intel.txt

@@ -5,4 +5,4 @@ accelerate
 torch
 rerankers[transformers]
 optimum[openvino]
-setuptools==72.1.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406

backend/python/rerankers/requirements.txt

@@ -1,3 +1,3 @@
-grpcio==1.65.4
+grpcio==1.66.2
 protobuf
 certifi

backend/python/sentencetransformers/backend.py

@@ -55,7 +55,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         """
         model_name = request.Model
         try:
-            self.model = SentenceTransformer(model_name)
+            self.model = SentenceTransformer(model_name, trust_remote_code=request.TrustRemoteCode)
         except Exception as err:
             return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
 

backend/python/sentencetransformers/requirements-cpu.txt

@@ -2,5 +2,5 @@ torch
 accelerate
 transformers
 bitsandbytes
-sentence-transformers==3.0.1
+sentence-transformers==3.1.1
 transformers

backend/python/sentencetransformers/requirements-cublas11.txt

@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
 torch
 accelerate
-sentence-transformers==3.0.1
+sentence-transformers==3.1.1
 transformers

backend/python/sentencetransformers/requirements-cublas12.txt

@@ -1,4 +1,4 @@
 torch
 accelerate
-sentence-transformers==3.0.1
+sentence-transformers==3.1.1
 transformers

backend/python/sentencetransformers/requirements-hipblas.txt

@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
 torch
 accelerate
-sentence-transformers==3.0.1
+sentence-transformers==3.1.1
 transformers

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==3.0.1
+sentence-transformers==3.1.1
 transformers

backend/python/sentencetransformers/requirements.txt

@@ -1,3 +1,5 @@
-grpcio==1.65.5
+grpcio==1.66.2
 protobuf
-certifi
+certifi
+datasets
+einops

backend/python/transformers-musicgen/backend.py

@@ -15,7 +15,7 @@ import backend_pb2_grpc
 
 import grpc
 
-from scipy.io.wavfile import write as write_wav
+from scipy.io import wavfile
 from transformers import AutoProcessor, MusicgenForConditionalGeneration
 
 _ONE_DAY_IN_SECONDS = 60 * 60 * 24
@@ -63,6 +63,61 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
 
         return backend_pb2.Result(message="Model loaded successfully", success=True)
 
+    def SoundGeneration(self, request, context):
+        model_name = request.model
+        if model_name == "":
+            return backend_pb2.Result(success=False, message="request.model is required")
+        try:
+            self.processor = AutoProcessor.from_pretrained(model_name)
+            self.model = MusicgenForConditionalGeneration.from_pretrained(model_name)
+            inputs = None
+            if request.text == "":
+                inputs = self.model.get_unconditional_inputs(num_samples=1)
+            elif request.HasField('src'):
+                # TODO SECURITY CODE GOES HERE LOL
+                # WHO KNOWS IF THIS WORKS???
+                sample_rate, wsamples = wavfile.read('path_to_your_file.wav')
+                
+                if request.HasField('src_divisor'):
+                    wsamples = wsamples[: len(wsamples) // request.src_divisor]
+                
+                inputs = self.processor(
+                    audio=wsamples,
+                    sampling_rate=sample_rate,
+                    text=[request.text],
+                    padding=True,
+                    return_tensors="pt",
+                )
+            else:
+                inputs = self.processor(
+                    text=[request.text],
+                    padding=True,
+                    return_tensors="pt",
+                )
+            
+            tokens = 256
+            if request.HasField('duration'):
+                tokens = int(request.duration * 51.2) # 256 tokens = 5 seconds, therefore 51.2 tokens is one second
+            guidance = 3.0
+            if request.HasField('temperature'):
+                guidance = request.temperature
+            dosample = True
+            if request.HasField('sample'):
+                dosample = request.sample
+            audio_values = self.model.generate(**inputs, do_sample=dosample, guidance_scale=guidance, max_new_tokens=tokens)
+            print("[transformers-musicgen] SoundGeneration generated!", file=sys.stderr)
+            sampling_rate = self.model.config.audio_encoder.sampling_rate
+            wavfile.write(request.dst, rate=sampling_rate, data=audio_values[0, 0].numpy())
+            print("[transformers-musicgen] SoundGeneration saved to", request.dst, file=sys.stderr)
+            print("[transformers-musicgen] SoundGeneration for", file=sys.stderr)
+            print("[transformers-musicgen] SoundGeneration requested tokens", tokens, file=sys.stderr)
+            print(request, file=sys.stderr)
+        except Exception as err:
+            return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
+        return backend_pb2.Result(success=True)
+
+
+# The TTS endpoint is older, and provides fewer features, but exists for compatibility reasons
     def TTS(self, request, context):
         model_name = request.model
         if model_name == "":
@@ -75,8 +130,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                 padding=True,
                 return_tensors="pt",
             )
-            tokens = 256
-            # TODO get tokens from request?
+            tokens = 512 # No good place to set the "length" in TTS, so use 10s as a sane default
             audio_values = self.model.generate(**inputs, max_new_tokens=tokens)
             print("[transformers-musicgen] TTS generated!", file=sys.stderr)
             sampling_rate = self.model.config.audio_encoder.sampling_rate

backend/python/transformers-musicgen/requirements-intel.txt

@@ -4,4 +4,4 @@ transformers
 accelerate
 torch
 optimum[openvino]
-setuptools==69.5.1 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406

backend/python/transformers-musicgen/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.65.5
+grpcio==1.66.2
 protobuf
 scipy==1.14.0
 certifi

backend/python/transformers-musicgen/test.py

@@ -63,7 +63,7 @@ class TestBackendServicer(unittest.TestCase):
 
     def test_tts(self):
         """
-        This method tests if the embeddings are generated successfully
+        This method tests if TTS is generated successfully
         """
         try:
             self.setUp()
@@ -77,5 +77,24 @@ class TestBackendServicer(unittest.TestCase):
         except Exception as err:
             print(err)
             self.fail("TTS service failed")
+        finally:
+            self.tearDown()
+
+    def test_sound_generation(self):
+        """
+        This method tests if SoundGeneration is generated successfully
+        """
+        try:
+            self.setUp()
+            with grpc.insecure_channel("localhost:50051") as channel:
+                stub = backend_pb2_grpc.BackendStub(channel)
+                response = stub.LoadModel(backend_pb2.ModelOptions(Model="facebook/musicgen-small"))
+                self.assertTrue(response.success)
+                sg_request = backend_pb2.SoundGenerationRequest(text="80s TV news production music hit for tonight's biggest story")
+                sg_response = stub.SoundGeneration(sg_request)
+                self.assertIsNotNone(sg_response)
+        except Exception as err:
+            print(err)
+            self.fail("SoundGeneration service failed")
         finally:
             self.tearDown()

backend/python/transformers/backend.py

@@ -72,7 +72,12 @@ 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:

backend/python/transformers/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.65.5
+grpcio==1.66.2
 protobuf
 certifi
 setuptools==69.5.1 # https://github.com/mudler/LocalAI/issues/2406

backend/python/vall-e-x/requirements-intel.txt

@@ -4,4 +4,4 @@ accelerate
 torch
 torchaudio
 optimum[openvino]
-setuptools==72.1.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406

backend/python/vall-e-x/requirements.txt

@@ -1,3 +1,3 @@
-grpcio==1.65.5
+grpcio==1.66.2
 protobuf
 certifi

backend/python/vllm/backend.py

@@ -5,6 +5,8 @@ import argparse
 import signal
 import sys
 import os
+from typing import List
+from PIL import Image
 
 import backend_pb2
 import backend_pb2_grpc
@@ -15,6 +17,8 @@ from vllm.engine.async_llm_engine import AsyncLLMEngine
 from vllm.sampling_params import SamplingParams
 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
 
 _ONE_DAY_IN_SECONDS = 60 * 60 * 24
 
@@ -105,6 +109,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         try:
             self.llm = AsyncLLMEngine.from_engine_args(engine_args)
         except Exception as err:
+            print(f"Unexpected {err=}, {type(err)=}", file=sys.stderr)
             return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
 
         try:
@@ -117,7 +122,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
            )
         except Exception as err:
             return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
-
+        print("Model loaded successfully", file=sys.stderr)
         return backend_pb2.Result(message="Model loaded successfully", success=True)
 
     async def Predict(self, request, context):
@@ -135,6 +140,26 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         res = await gen.__anext__()
         return res
 
+    def Embedding(self, request, context):
+        """
+        A gRPC method that calculates embeddings for a given sentence.
+
+        Args:
+            request: An EmbeddingRequest object that contains the request parameters.
+            context: A grpc.ServicerContext object that provides information about the RPC.
+
+        Returns:
+            An EmbeddingResult object that contains the calculated embeddings.
+        """
+        print("Calculated embeddings for: " + request.Embeddings, file=sys.stderr)
+        outputs = self.model.encode(request.Embeddings)
+        # Check if we have one result at least
+        if len(outputs) == 0:
+            context.set_code(grpc.StatusCode.INVALID_ARGUMENT)
+            context.set_details("No embeddings were calculated.")
+            return backend_pb2.EmbeddingResult()
+        return backend_pb2.EmbeddingResult(embeddings=outputs[0].outputs.embedding)
+
     async def PredictStream(self, request, context):
         """
         Generates text based on the given prompt and sampling parameters, and streams the results.
@@ -176,15 +201,33 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         if request.Seed != 0:
             sampling_params.seed = request.Seed
 
+        # Extract image paths and process images
         prompt = request.Prompt
-        
-        # If tokenizer template is enabled and messages are provided instead of prompt apply the tokenizer template
+
+        image_paths = request.Images
+        image_data = [self.load_image(img_path) for img_path in image_paths]
+
+        videos_path = request.Videos
+        video_data = [self.load_video(video_path) for video_path in videos_path]
+
+        # If tokenizer template is enabled and messages are provided instead of prompt, apply the tokenizer template
         if not request.Prompt and request.UseTokenizerTemplate and request.Messages:
             prompt = self.tokenizer.apply_chat_template(request.Messages, tokenize=False, add_generation_prompt=True)
 
-        # Generate text
+        # Generate text using the LLM engine
         request_id = random_uuid()
-        outputs = self.llm.generate(prompt, sampling_params, request_id)
+        print(f"Generating text with request_id: {request_id}", file=sys.stderr)
+        outputs = self.llm.generate(
+            {
+                "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,
+        )
 
         # Stream the results
         generated_text = ""
@@ -207,9 +250,49 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         if streaming:
             return
 
+        # Remove the image files from /tmp folder
+        for img_path in image_paths:
+            try:
+                os.remove(img_path)
+            except Exception as e:
+                print(f"Error removing image file: {img_path}, {e}", file=sys.stderr)
+
         # Sending the final generated text
         yield backend_pb2.Reply(message=bytes(generated_text, encoding='utf-8'))
 
+    def load_image(self, image_path: str):
+        """
+        Load an image from the given file path.
+        
+        Args:
+            image_path (str): The path to the image file.
+
+        Returns:
+            Image: The loaded image.
+        """
+        try:
+            return Image.open(image_path)
+        except Exception as e:
+            print(f"Error loading image {image_path}: {e}", file=sys.stderr)
+            return self.load_video(image_path)
+
+    def load_video(self, video_path: str):
+        """
+        Load a video from the given file path.
+        
+        Args:
+            video_path (str): The path to the image file.
+
+        Returns:
+            Video: The loaded video.
+        """
+        try:
+            video = VideoAsset(name=video_path).np_ndarrays
+            return video
+        except Exception as e:
+            print(f"Error loading video {image_path}: {e}", file=sys.stderr)
+            return None
+
 async def serve(address):
     # Start asyncio gRPC server
     server = grpc.aio.server(migration_thread_pool=futures.ThreadPoolExecutor(max_workers=MAX_WORKERS))

backend/python/vllm/install.sh

@@ -13,4 +13,20 @@ if [ "x${BUILD_PROFILE}" == "xintel" ]; then
     EXTRA_PIP_INSTALL_FLAGS+=" --upgrade --index-strategy=unsafe-first-match"
 fi
 
-installRequirements
+# 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
+        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==1.66.2 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
+        rm -rf vllm
+    else
+        installRequirements
+fi

backend/python/vllm/requirements-cublas11.txt

@@ -1,4 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
 accelerate
 torch
-transformers
+transformers
+bitsandbytes

backend/python/vllm/requirements-cublas12.txt

@@ -1,3 +1,4 @@
 accelerate
 torch
-transformers
+transformers
+bitsandbytes

backend/python/vllm/requirements-hipblas.txt

@@ -1,4 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
 accelerate
 torch
-transformers
+transformers
+bitsandbytes

backend/python/vllm/requirements-intel.txt

@@ -4,4 +4,5 @@ accelerate
 torch
 transformers
 optimum[openvino]
-setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
+bitsandbytes

backend/python/vllm/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.65.5
+grpcio==1.66.2
 protobuf
 certifi
 setuptools

backend/python/vllm/test.py

@@ -72,5 +72,28 @@ class TestBackendServicer(unittest.TestCase):
         except Exception as err:
             print(err)
             self.fail("text service failed")
+        finally:
+            self.tearDown()
+
+    def test_embedding(self):
+        """
+        This method tests if the embeddings are generated successfully
+        """
+        try:
+            self.setUp()
+            with grpc.insecure_channel("localhost:50051") as channel:
+                stub = backend_pb2_grpc.BackendStub(channel)
+                response = stub.LoadModel(backend_pb2.ModelOptions(Model="intfloat/e5-mistral-7b-instruct"))
+                self.assertTrue(response.success)
+                embedding_request = backend_pb2.PredictOptions(Embeddings="This is a test sentence.")
+                embedding_response = stub.Embedding(embedding_request)
+                self.assertIsNotNone(embedding_response.embeddings)
+                # assert that is a list of floats
+                self.assertIsInstance(embedding_response.embeddings, list)
+                # assert that the list is not empty
+                self.assertTrue(len(embedding_response.embeddings) > 0)
+        except Exception as err:
+            print(err)
+            self.fail("Embedding service failed")
         finally:
             self.tearDown()

core/backend/backend_suite_test.go

@@ -0,0 +1,13 @@
+package backend_test
+
+import (
+	"testing"
+
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+func TestBackend(t *testing.T) {
+	RegisterFailHandler(Fail)
+	RunSpecs(t, "Backend test suite")
+}

core/backend/embeddings.go

@@ -10,20 +10,11 @@ import (
 )
 
 func ModelEmbedding(s string, tokens []int, loader *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (func() ([]float32, error), error) {
-	modelFile := backendConfig.Model
-
-	grpcOpts := gRPCModelOpts(backendConfig)
 
 	var inferenceModel interface{}
 	var err error
 
-	opts := modelOpts(backendConfig, appConfig, []model.Option{
-		model.WithLoadGRPCLoadModelOpts(grpcOpts),
-		model.WithThreads(uint32(*backendConfig.Threads)),
-		model.WithAssetDir(appConfig.AssetsDestination),
-		model.WithModel(modelFile),
-		model.WithContext(appConfig.Context),
-	})
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{})
 
 	if backendConfig.Backend == "" {
 		inferenceModel, err = loader.GreedyLoader(opts...)

core/backend/image.go

@@ -8,19 +8,8 @@ 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) {
-	threads := backendConfig.Threads
-	if *threads == 0 && appConfig.Threads != 0 {
-		threads = &appConfig.Threads
-	}
-	gRPCOpts := gRPCModelOpts(backendConfig)
-	opts := modelOpts(backendConfig, appConfig, []model.Option{
-		model.WithBackendString(backendConfig.Backend),
-		model.WithAssetDir(appConfig.AssetsDestination),
-		model.WithThreads(uint32(*threads)),
-		model.WithContext(appConfig.Context),
-		model.WithModel(backendConfig.Model),
-		model.WithLoadGRPCLoadModelOpts(gRPCOpts),
-	})
+
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{})
 
 	inferenceModel, err := loader.BackendLoader(
 		opts...,

core/backend/llm.go

@@ -9,6 +9,8 @@ import (
 	"sync"
 	"unicode/utf8"
 
+	"github.com/rs/zerolog/log"
+
 	"github.com/mudler/LocalAI/core/config"
 	"github.com/mudler/LocalAI/core/schema"
 
@@ -29,24 +31,13 @@ type TokenUsage struct {
 	Completion int
 }
 
-func ModelInference(ctx context.Context, s string, messages []schema.Message, images []string, loader *model.ModelLoader, c config.BackendConfig, o *config.ApplicationConfig, tokenCallback func(string, TokenUsage) bool) (func() (LLMResponse, error), error) {
+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
-	threads := c.Threads
-	if *threads == 0 && o.Threads != 0 {
-		threads = &o.Threads
-	}
-	grpcOpts := gRPCModelOpts(c)
 
 	var inferenceModel grpc.Backend
 	var err error
 
-	opts := modelOpts(c, o, []model.Option{
-		model.WithLoadGRPCLoadModelOpts(grpcOpts),
-		model.WithThreads(uint32(*threads)), // some models uses this to allocate threads during startup
-		model.WithAssetDir(o.AssetsDestination),
-		model.WithModel(modelFile),
-		model.WithContext(o.Context),
-	})
+	opts := ModelOptions(c, o, []model.Option{})
 
 	if c.Backend != "" {
 		opts = append(opts, model.WithBackendString(c.Backend))
@@ -87,7 +78,7 @@ func ModelInference(ctx context.Context, s string, messages []schema.Message, im
 			case string:
 				protoMessages[i].Content = ct
 			default:
-				return nil, fmt.Errorf("Unsupported type for schema.Message.Content for inference: %T", ct)
+				return nil, fmt.Errorf("unsupported type for schema.Message.Content for inference: %T", ct)
 			}
 		}
 	}
@@ -99,6 +90,8 @@ func ModelInference(ctx context.Context, s string, messages []schema.Message, im
 		opts.Messages = protoMessages
 		opts.UseTokenizerTemplate = c.TemplateConfig.UseTokenizerTemplate
 		opts.Images = images
+		opts.Videos = videos
+		opts.Audios = audios
 
 		tokenUsage := TokenUsage{}
 
@@ -181,13 +174,37 @@ func Finetune(config config.BackendConfig, input, prediction string) string {
 		mu.Lock()
 		reg, ok := cutstrings[c]
 		if !ok {
-			cutstrings[c] = regexp.MustCompile(c)
+			r, err := regexp.Compile(c)
+			if err != nil {
+				log.Fatal().Err(err).Msg("failed to compile regex")
+			}
+			cutstrings[c] = r
 			reg = cutstrings[c]
 		}
 		mu.Unlock()
 		prediction = reg.ReplaceAllString(prediction, "")
 	}
 
+	// extract results from the response which can be for instance inside XML tags
+	var predResult string
+	for _, r := range config.ExtractRegex {
+		mu.Lock()
+		reg, ok := cutstrings[r]
+		if !ok {
+			regex, err := regexp.Compile(r)
+			if err != nil {
+				log.Fatal().Err(err).Msg("failed to compile regex")
+			}
+			cutstrings[r] = regex
+			reg = regex
+		}
+		mu.Unlock()
+		predResult += reg.FindString(prediction)
+	}
+	if predResult != "" {
+		prediction = predResult
+	}
+
 	for _, c := range config.TrimSpace {
 		prediction = strings.TrimSpace(strings.TrimPrefix(prediction, c))
 	}

core/backend/llm_test.go

@@ -0,0 +1,109 @@
+package backend_test
+
+import (
+	. "github.com/mudler/LocalAI/core/backend"
+	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/core/schema"
+
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+var _ = Describe("LLM tests", func() {
+	Context("Finetune LLM output", func() {
+		var (
+			testConfig config.BackendConfig
+			input      string
+			prediction string
+			result     string
+		)
+
+		BeforeEach(func() {
+			testConfig = config.BackendConfig{
+				PredictionOptions: schema.PredictionOptions{
+					Echo: false,
+				},
+				LLMConfig: config.LLMConfig{
+					Cutstrings:   []string{`<.*?>`},                  // Example regex for removing XML tags
+					ExtractRegex: []string{`<result>(.*?)</result>`}, // Example regex to extract from tags
+					TrimSpace:    []string{" ", "\n"},
+					TrimSuffix:   []string{".", "!"},
+				},
+			}
+		})
+
+		Context("when echo is enabled", func() {
+			BeforeEach(func() {
+				testConfig.Echo = true
+				input = "Hello"
+				prediction = "World"
+			})
+
+			It("should prepend input to prediction", func() {
+				result = Finetune(testConfig, input, prediction)
+				Expect(result).To(Equal("HelloWorld"))
+			})
+		})
+
+		Context("when echo is disabled", func() {
+			BeforeEach(func() {
+				testConfig.Echo = false
+				input = "Hello"
+				prediction = "World"
+			})
+
+			It("should not modify the prediction with input", func() {
+				result = Finetune(testConfig, input, prediction)
+				Expect(result).To(Equal("World"))
+			})
+		})
+
+		Context("when cutstrings regex is applied", func() {
+			BeforeEach(func() {
+				input = ""
+				prediction = "<div>Hello</div> World"
+			})
+
+			It("should remove substrings matching cutstrings regex", func() {
+				result = Finetune(testConfig, input, prediction)
+				Expect(result).To(Equal("Hello World"))
+			})
+		})
+
+		Context("when extract regex is applied", func() {
+			BeforeEach(func() {
+				input = ""
+				prediction = "<response><result>42</result></response>"
+			})
+
+			It("should extract substrings matching the extract regex", func() {
+				result = Finetune(testConfig, input, prediction)
+				Expect(result).To(Equal("42"))
+			})
+		})
+
+		Context("when trimming spaces", func() {
+			BeforeEach(func() {
+				input = ""
+				prediction = "   Hello World   "
+			})
+
+			It("should trim spaces from the prediction", func() {
+				result = Finetune(testConfig, input, prediction)
+				Expect(result).To(Equal("Hello World"))
+			})
+		})
+
+		Context("when trimming suffixes", func() {
+			BeforeEach(func() {
+				input = ""
+				prediction = "Hello World."
+			})
+
+			It("should trim suffixes from the prediction", func() {
+				result = Finetune(testConfig, input, prediction)
+				Expect(result).To(Equal("Hello World"))
+			})
+		})
+	})
+})

core/backend/options.go

@@ -11,32 +11,65 @@ import (
 	"github.com/rs/zerolog/log"
 )
 
-func modelOpts(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
+	}
+
+	defOpts := []model.Option{
+		model.WithBackendString(c.Backend),
+		model.WithModel(c.Model),
+		model.WithAssetDir(so.AssetsDestination),
+		model.WithContext(so.Context),
+		model.WithModelID(name),
+	}
+
+	threads := 1
+
+	if c.Threads != nil {
+		threads = *c.Threads
+	}
+
+	if so.Threads != 0 {
+		threads = so.Threads
+	}
+
+	c.Threads = &threads
+
+	grpcOpts := grpcModelOpts(c)
+	defOpts = append(defOpts, model.WithLoadGRPCLoadModelOpts(grpcOpts))
+
 	if so.SingleBackend {
-		opts = append(opts, model.WithSingleActiveBackend())
+		defOpts = append(defOpts, model.WithSingleActiveBackend())
 	}
 
 	if so.ParallelBackendRequests {
-		opts = append(opts, model.EnableParallelRequests)
+		defOpts = append(defOpts, model.EnableParallelRequests)
 	}
 
 	if c.GRPC.Attempts != 0 {
-		opts = append(opts, model.WithGRPCAttempts(c.GRPC.Attempts))
+		defOpts = append(defOpts, model.WithGRPCAttempts(c.GRPC.Attempts))
 	}
 
 	if c.GRPC.AttemptsSleepTime != 0 {
-		opts = append(opts, model.WithGRPCAttemptsDelay(c.GRPC.AttemptsSleepTime))
+		defOpts = append(defOpts, model.WithGRPCAttemptsDelay(c.GRPC.AttemptsSleepTime))
 	}
 
 	for k, v := range so.ExternalGRPCBackends {
-		opts = append(opts, model.WithExternalBackend(k, v))
+		defOpts = append(defOpts, model.WithExternalBackend(k, v))
 	}
 
-	return opts
+	return append(defOpts, opts...)
 }
 
 func getSeed(c config.BackendConfig) int32 {
-	seed := int32(*c.Seed)
+	var seed int32 = config.RAND_SEED
+
+	if c.Seed != nil {
+		seed = int32(*c.Seed)
+	}
+
 	if seed == config.RAND_SEED {
 		seed = rand.Int31()
 	}
@@ -44,11 +77,47 @@ func getSeed(c config.BackendConfig) int32 {
 	return seed
 }
 
-func gRPCModelOpts(c config.BackendConfig) *pb.ModelOptions {
+func grpcModelOpts(c config.BackendConfig) *pb.ModelOptions {
 	b := 512
 	if c.Batch != 0 {
 		b = c.Batch
 	}
+
+	f16 := false
+	if c.F16 != nil {
+		f16 = *c.F16
+	}
+
+	embeddings := false
+	if c.Embeddings != nil {
+		embeddings = *c.Embeddings
+	}
+
+	lowVRAM := false
+	if c.LowVRAM != nil {
+		lowVRAM = *c.LowVRAM
+	}
+
+	mmap := false
+	if c.MMap != nil {
+		mmap = *c.MMap
+	}
+
+	ctxSize := 1024
+	if c.ContextSize != nil {
+		ctxSize = *c.ContextSize
+	}
+
+	mmlock := false
+	if c.MMlock != nil {
+		mmlock = *c.MMlock
+	}
+
+	nGPULayers := 9999999
+	if c.NGPULayers != nil {
+		nGPULayers = *c.NGPULayers
+	}
+
 	return &pb.ModelOptions{
 		CUDA:                 c.CUDA || c.Diffusers.CUDA,
 		SchedulerType:        c.Diffusers.SchedulerType,
@@ -56,14 +125,14 @@ func gRPCModelOpts(c config.BackendConfig) *pb.ModelOptions {
 		CFGScale:             c.Diffusers.CFGScale,
 		LoraAdapter:          c.LoraAdapter,
 		LoraScale:            c.LoraScale,
-		F16Memory:            *c.F16,
+		F16Memory:            f16,
 		LoraBase:             c.LoraBase,
 		IMG2IMG:              c.Diffusers.IMG2IMG,
 		CLIPModel:            c.Diffusers.ClipModel,
 		CLIPSubfolder:        c.Diffusers.ClipSubFolder,
 		CLIPSkip:             int32(c.Diffusers.ClipSkip),
 		ControlNet:           c.Diffusers.ControlNet,
-		ContextSize:          int32(*c.ContextSize),
+		ContextSize:          int32(ctxSize),
 		Seed:                 getSeed(c),
 		NBatch:               int32(b),
 		NoMulMatQ:            c.NoMulMatQ,
@@ -85,16 +154,16 @@ func gRPCModelOpts(c config.BackendConfig) *pb.ModelOptions {
 		YarnBetaSlow:         c.YarnBetaSlow,
 		NGQA:                 c.NGQA,
 		RMSNormEps:           c.RMSNormEps,
-		MLock:                *c.MMlock,
+		MLock:                mmlock,
 		RopeFreqBase:         c.RopeFreqBase,
 		RopeScaling:          c.RopeScaling,
 		Type:                 c.ModelType,
 		RopeFreqScale:        c.RopeFreqScale,
 		NUMA:                 c.NUMA,
-		Embeddings:           *c.Embeddings,
-		LowVRAM:              *c.LowVRAM,
-		NGPULayers:           int32(*c.NGPULayers),
-		MMap:                 *c.MMap,
+		Embeddings:           embeddings,
+		LowVRAM:              lowVRAM,
+		NGPULayers:           int32(nGPULayers),
+		MMap:                 mmap,
 		MainGPU:              c.MainGPU,
 		Threads:              int32(*c.Threads),
 		TensorSplit:          c.TensorSplit,

core/backend/rerank.go

@@ -9,21 +9,9 @@ import (
 	model "github.com/mudler/LocalAI/pkg/model"
 )
 
-func Rerank(backend, modelFile string, request *proto.RerankRequest, loader *model.ModelLoader, appConfig *config.ApplicationConfig, backendConfig config.BackendConfig) (*proto.RerankResult, error) {
-	bb := backend
-	if bb == "" {
-		return nil, fmt.Errorf("backend is required")
-	}
+func Rerank(modelFile string, request *proto.RerankRequest, loader *model.ModelLoader, appConfig *config.ApplicationConfig, backendConfig config.BackendConfig) (*proto.RerankResult, error) {
 
-	grpcOpts := gRPCModelOpts(backendConfig)
-
-	opts := modelOpts(config.BackendConfig{}, appConfig, []model.Option{
-		model.WithBackendString(bb),
-		model.WithModel(modelFile),
-		model.WithContext(appConfig.Context),
-		model.WithAssetDir(appConfig.AssetsDestination),
-		model.WithLoadGRPCLoadModelOpts(grpcOpts),
-	})
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{model.WithModel(modelFile)})
 	rerankModel, err := loader.BackendLoader(opts...)
 	if err != nil {
 		return nil, err

core/backend/soundgeneration.go

@@ -0,0 +1,63 @@
+package backend
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"path/filepath"
+
+	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/pkg/grpc/proto"
+	"github.com/mudler/LocalAI/pkg/model"
+	"github.com/mudler/LocalAI/pkg/utils"
+)
+
+func SoundGeneration(
+	modelFile string,
+	text string,
+	duration *float32,
+	temperature *float32,
+	doSample *bool,
+	sourceFile *string,
+	sourceDivisor *int32,
+	loader *model.ModelLoader,
+	appConfig *config.ApplicationConfig,
+	backendConfig config.BackendConfig,
+) (string, *proto.Result, error) {
+
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{model.WithModel(modelFile)})
+
+	soundGenModel, err := loader.BackendLoader(opts...)
+	if err != nil {
+		return "", nil, err
+	}
+
+	if soundGenModel == nil {
+		return "", nil, fmt.Errorf("could not load sound generation model")
+	}
+
+	if err := os.MkdirAll(appConfig.AudioDir, 0750); err != nil {
+		return "", nil, fmt.Errorf("failed creating audio directory: %s", err)
+	}
+
+	fileName := utils.GenerateUniqueFileName(appConfig.AudioDir, "sound_generation", ".wav")
+	filePath := filepath.Join(appConfig.AudioDir, fileName)
+
+	res, err := soundGenModel.SoundGeneration(context.Background(), &proto.SoundGenerationRequest{
+		Text:        text,
+		Model:       modelFile,
+		Dst:         filePath,
+		Sample:      doSample,
+		Duration:    duration,
+		Temperature: temperature,
+		Src:         sourceFile,
+		SrcDivisor:  sourceDivisor,
+	})
+
+	// return RPC error if any
+	if !res.Success {
+		return "", nil, fmt.Errorf(res.Message)
+	}
+
+	return filePath, res, err
+}

core/backend/token_metrics.go

@@ -0,0 +1,33 @@
+package backend
+
+import (
+	"context"
+	"fmt"
+
+	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/pkg/grpc/proto"
+	model "github.com/mudler/LocalAI/pkg/model"
+)
+
+func TokenMetrics(
+	modelFile string,
+	loader *model.ModelLoader,
+	appConfig *config.ApplicationConfig,
+	backendConfig config.BackendConfig) (*proto.MetricsResponse, error) {
+
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{
+		model.WithModel(modelFile),
+	})
+	model, err := loader.BackendLoader(opts...)
+	if err != nil {
+		return nil, err
+	}
+
+	if model == nil {
+		return nil, fmt.Errorf("could not loadmodel model")
+	}
+
+	res, err := model.GetTokenMetrics(context.Background(), &proto.MetricsRequest{})
+
+	return res, err
+}

core/backend/tokenize.go

@@ -0,0 +1,44 @@
+package backend
+
+import (
+	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/core/schema"
+	"github.com/mudler/LocalAI/pkg/grpc"
+	model "github.com/mudler/LocalAI/pkg/model"
+)
+
+func ModelTokenize(s string, loader *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (schema.TokenizeResponse, error) {
+
+	modelFile := backendConfig.Model
+
+	var inferenceModel grpc.Backend
+	var err error
+
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{
+		model.WithModel(modelFile),
+	})
+
+	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 schema.TokenizeResponse{}, err
+	}
+
+	predictOptions := gRPCPredictOpts(backendConfig, loader.ModelPath)
+	predictOptions.Prompt = s
+
+	// tokenize the string
+	resp, err := inferenceModel.TokenizeString(appConfig.Context, predictOptions)
+	if err != nil {
+		return schema.TokenizeResponse{}, err
+	}
+
+	return schema.TokenizeResponse{
+		Tokens: resp.Tokens,
+	}, nil
+
+}

core/backend/transcript.go

@@ -3,37 +3,57 @@ package backend
 import (
 	"context"
 	"fmt"
+	"time"
 
 	"github.com/mudler/LocalAI/core/config"
 	"github.com/mudler/LocalAI/core/schema"
 
 	"github.com/mudler/LocalAI/pkg/grpc/proto"
-	model "github.com/mudler/LocalAI/pkg/model"
+	"github.com/mudler/LocalAI/pkg/model"
 )
 
 func ModelTranscription(audio, language string, translate bool, ml *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (*schema.TranscriptionResult, error) {
 
-	opts := modelOpts(backendConfig, appConfig, []model.Option{
-		model.WithBackendString(model.WhisperBackend),
-		model.WithModel(backendConfig.Model),
-		model.WithContext(appConfig.Context),
-		model.WithThreads(uint32(*backendConfig.Threads)),
-		model.WithAssetDir(appConfig.AssetsDestination),
-	})
+	if backendConfig.Backend == "" {
+		backendConfig.Backend = model.WhisperBackend
+	}
 
-	whisperModel, err := ml.BackendLoader(opts...)
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{})
+
+	transcriptionModel, err := ml.BackendLoader(opts...)
 	if err != nil {
 		return nil, err
 	}
 
-	if whisperModel == nil {
-		return nil, fmt.Errorf("could not load whisper model")
+	if transcriptionModel == nil {
+		return nil, fmt.Errorf("could not load transcription model")
 	}
 
-	return whisperModel.AudioTranscription(context.Background(), &proto.TranscriptRequest{
+	r, err := transcriptionModel.AudioTranscription(context.Background(), &proto.TranscriptRequest{
 		Dst:       audio,
 		Language:  language,
 		Translate: translate,
 		Threads:   uint32(*backendConfig.Threads),
 	})
+	if err != nil {
+		return nil, err
+	}
+	tr := &schema.TranscriptionResult{
+		Text: r.Text,
+	}
+	for _, s := range r.Segments {
+		var tks []int
+		for _, t := range s.Tokens {
+			tks = append(tks, int(t))
+		}
+		tr.Segments = append(tr.Segments,
+			schema.Segment{
+				Text:   s.Text,
+				Id:     int(s.Id),
+				Start:  time.Duration(s.Start),
+				End:    time.Duration(s.End),
+				Tokens: tks,
+			})
+	}
+	return tr, err
 }

core/backend/tts.go

@@ -9,31 +9,15 @@ import (
 	"github.com/mudler/LocalAI/core/config"
 
 	"github.com/mudler/LocalAI/pkg/grpc/proto"
-	model "github.com/mudler/LocalAI/pkg/model"
+	"github.com/mudler/LocalAI/pkg/model"
 	"github.com/mudler/LocalAI/pkg/utils"
 )
 
-func generateUniqueFileName(dir, baseName, ext string) string {
-	counter := 1
-	fileName := baseName + ext
-
-	for {
-		filePath := filepath.Join(dir, fileName)
-		_, err := os.Stat(filePath)
-		if os.IsNotExist(err) {
-			return fileName
-		}
-
-		counter++
-		fileName = fmt.Sprintf("%s_%d%s", baseName, counter, ext)
-	}
-}
-
 func ModelTTS(
 	backend,
 	text,
 	modelFile,
-	voice ,
+	voice,
 	language string,
 	loader *model.ModelLoader,
 	appConfig *config.ApplicationConfig,
@@ -44,14 +28,9 @@ func ModelTTS(
 		bb = model.PiperBackend
 	}
 
-	grpcOpts := gRPCModelOpts(backendConfig)
-
-	opts := modelOpts(config.BackendConfig{}, appConfig, []model.Option{
+	opts := ModelOptions(config.BackendConfig{}, appConfig, []model.Option{
 		model.WithBackendString(bb),
 		model.WithModel(modelFile),
-		model.WithContext(appConfig.Context),
-		model.WithAssetDir(appConfig.AssetsDestination),
-		model.WithLoadGRPCLoadModelOpts(grpcOpts),
 	})
 	ttsModel, err := loader.BackendLoader(opts...)
 	if err != nil {
@@ -66,7 +45,7 @@ func ModelTTS(
 		return "", nil, fmt.Errorf("failed creating audio directory: %s", err)
 	}
 
-	fileName := generateUniqueFileName(appConfig.AudioDir, "tts", ".wav")
+	fileName := utils.GenerateUniqueFileName(appConfig.AudioDir, "tts", ".wav")
 	filePath := filepath.Join(appConfig.AudioDir, fileName)
 
 	// If the model file is not empty, we pass it joined with the model path
@@ -88,12 +67,15 @@ func ModelTTS(
 	}
 
 	res, err := ttsModel.TTS(context.Background(), &proto.TTSRequest{
-		Text:  text,
-		Model: modelPath,
-		Voice: voice,
-		Dst:   filePath,
+		Text:     text,
+		Model:    modelPath,
+		Voice:    voice,
+		Dst:      filePath,
 		Language: &language,
 	})
+	if err != nil {
+		return "", nil, err
+	}
 
 	// return RPC error if any
 	if !res.Success {

core/cli/api/p2p.go

@@ -0,0 +1,80 @@
+package cli_api
+
+import (
+	"context"
+	"fmt"
+	"net"
+	"os"
+	"strings"
+
+	"github.com/mudler/LocalAI/core/p2p"
+	"github.com/mudler/edgevpn/pkg/node"
+
+	"github.com/rs/zerolog/log"
+)
+
+func StartP2PStack(ctx context.Context, address, token, networkID string, federated bool) error {
+	var n *node.Node
+	// Here we are avoiding creating multiple nodes:
+	// - if the federated mode is enabled, we create a federated node and expose a service
+	// - exposing a service creates a node with specific options, and we don't want to create another node
+
+	// If the federated mode is enabled, we expose a service to the local instance running
+	// at r.Address
+	if federated {
+		_, port, err := net.SplitHostPort(address)
+		if err != nil {
+			return err
+		}
+
+		// Here a new node is created and started
+		// and a service is exposed by the node
+		node, err := p2p.ExposeService(ctx, "localhost", port, token, p2p.NetworkID(networkID, p2p.FederatedID))
+		if err != nil {
+			return err
+		}
+
+		if err := p2p.ServiceDiscoverer(ctx, node, token, p2p.NetworkID(networkID, p2p.FederatedID), nil, false); err != nil {
+			return err
+		}
+
+		n = node
+	}
+
+	// If the p2p mode is enabled, we start the service discovery
+	if token != "" {
+		// If a node wasn't created previously, create it
+		if n == nil {
+			node, err := p2p.NewNode(token)
+			if err != nil {
+				return err
+			}
+			err = node.Start(ctx)
+			if err != nil {
+				return fmt.Errorf("starting new node: %w", err)
+			}
+			n = node
+		}
+
+		// Attach a ServiceDiscoverer to the p2p node
+		log.Info().Msg("Starting P2P server discovery...")
+		if err := p2p.ServiceDiscoverer(ctx, n, token, p2p.NetworkID(networkID, p2p.WorkerID), func(serviceID string, node p2p.NodeData) {
+			var tunnelAddresses []string
+			for _, v := range p2p.GetAvailableNodes(p2p.NetworkID(networkID, p2p.WorkerID)) {
+				if v.IsOnline() {
+					tunnelAddresses = append(tunnelAddresses, v.TunnelAddress)
+				} else {
+					log.Info().Msgf("Node %s is offline", v.ID)
+				}
+			}
+			tunnelEnvVar := strings.Join(tunnelAddresses, ",")
+
+			os.Setenv("LLAMACPP_GRPC_SERVERS", tunnelEnvVar)
+			log.Debug().Msgf("setting LLAMACPP_GRPC_SERVERS to %s", tunnelEnvVar)
+		}, true); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}

core/cli/cli.go

@@ -8,12 +8,13 @@ import (
 var CLI struct {
 	cliContext.Context `embed:""`
 
-	Run        RunCMD        `cmd:"" help:"Run LocalAI, this the default command if no other command is specified. Run 'local-ai run --help' for more information" default:"withargs"`
-	Federated  FederatedCLI  `cmd:"" help:"Run LocalAI in federated mode"`
-	Models     ModelsCMD     `cmd:"" help:"Manage LocalAI models and definitions"`
-	TTS        TTSCMD        `cmd:"" help:"Convert text to speech"`
-	Transcript TranscriptCMD `cmd:"" help:"Convert audio to text"`
-	Worker     worker.Worker `cmd:"" help:"Run workers to distribute workload (llama.cpp-only)"`
-	Util       UtilCMD       `cmd:"" help:"Utility commands"`
-	Explorer   ExplorerCMD   `cmd:"" help:"Run p2p explorer"`
+	Run             RunCMD             `cmd:"" help:"Run LocalAI, this the default command if no other command is specified. Run 'local-ai run --help' for more information" default:"withargs"`
+	Federated       FederatedCLI       `cmd:"" help:"Run LocalAI in federated mode"`
+	Models          ModelsCMD          `cmd:"" help:"Manage LocalAI models and definitions"`
+	TTS             TTSCMD             `cmd:"" help:"Convert text to speech"`
+	SoundGeneration SoundGenerationCMD `cmd:"" help:"Generates audio files from text or audio"`
+	Transcript      TranscriptCMD      `cmd:"" help:"Convert audio to text"`
+	Worker          worker.Worker      `cmd:"" help:"Run workers to distribute workload (llama.cpp-only)"`
+	Util            UtilCMD            `cmd:"" help:"Utility commands"`
+	Explorer        ExplorerCMD        `cmd:"" help:"Run p2p explorer"`
 }