chore(model): add silero-vad model config

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

.bruno/LocalAI Test Requests/Sound Generation/musicgen.bru

@@ -0,0 +1,23 @@
+meta {
+  name: musicgen
+  type: http
+  seq: 1
+}
+
+post {
+  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/v1/sound-generation
+  body: json
+  auth: none
+}
+
+headers {
+  Content-Type: application/json
+}
+
+body:json {
+  {
+      "model_id": "facebook/musicgen-small",
+      "text": "Exciting 80s Newscast Interstitial",
+      "duration_seconds": 8
+  }
+}

.bruno/LocalAI Test Requests/model gallery/model delete.bru

@@ -0,0 +1,11 @@
+meta {
+  name: model delete
+  type: http
+  seq: 7
+}
+
+post {
+  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models/galleries
+  body: none
+  auth: none
+}

.bruno/LocalAI Test Requests/transcription/transcribe.bru

@@ -0,0 +1,16 @@
+meta {
+  name: transcribe
+  type: http
+  seq: 1
+}
+
+post {
+  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/v1/audio/transcriptions
+  body: multipartForm
+  auth: none
+}
+
+body:multipart-form {
+  file: @file(transcription/gb1.ogg)
+  model: whisper-1
+}

.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!"
 }

.gitattributes

@@ -1 +1,2 @@
 *.sh text eol=lf
+backend/cpp/llama/*.hpp linguist-vendored

.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/dependabot.yml

@@ -9,6 +9,8 @@ updates:
     directory: "/"
     schedule:
       interval: "weekly"
+    ignore:
+    - dependency-name: "github.com/mudler/LocalAI/pkg/grpc/proto"
   - package-ecosystem: "github-actions"
     # Workflow files stored in the default location of `.github/workflows`. (You don't need to specify `/.github/workflows` for `directory`. You can use `directory: "/"`.)
     directory: "/"

.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

@@ -23,7 +23,7 @@ jobs:
           sudo pip install --upgrade pip
           pip install huggingface_hub
       - name: 'Setup yq'
-        uses: dcarbone/install-yq-action@v1.1.1
+        uses: dcarbone/install-yq-action@v1.3.1
         with:
           version: 'v4.44.2'
           download-compressed: true
@@ -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@v1.2.0
         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@v1.2.0
         with:
             host: ${{ secrets.EXPLORER_SSH_HOST }}
             username: ${{ secrets.EXPLORER_SSH_USERNAME }}

.github/workflows/generate_intel_image.yaml

@@ -15,7 +15,7 @@ jobs:
     strategy:
       matrix:
         include:
-          - base-image: intel/oneapi-basekit:2024.2.0-devel-ubuntu22.04
+          - base-image: intel/oneapi-basekit:2025.0.0-0-devel-ubuntu22.04
             runs-on: 'ubuntu-latest'
             platforms: 'linux/amd64'
     runs-on: ${{matrix.runs-on}}

.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/notify-models.yaml

@@ -79,7 +79,7 @@ jobs:
         args: ${{ steps.summarize.outputs.message }}
     - name: Setup tmate session if fails
       if: ${{ failure() }}
-      uses: mxschmitt/action-tmate@v3.18
+      uses: mxschmitt/action-tmate@v3.19
       with:
         detached: true
         connect-timeout-seconds: 180
@@ -161,7 +161,7 @@ jobs:
         TWITTER_ACCESS_TOKEN_SECRET: ${{ secrets.TWITTER_ACCESS_TOKEN_SECRET }}
     - name: Setup tmate session if fails
       if: ${{ failure() }}
-      uses: mxschmitt/action-tmate@v3.18
+      uses: mxschmitt/action-tmate@v3.19
       with:
         detached: true
         connect-timeout-seconds: 180

.github/workflows/release.yaml

@@ -123,7 +123,7 @@ jobs:
             release/*
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.18
+        uses: mxschmitt/action-tmate@v3.19
         with:
           detached: true
           connect-timeout-seconds: 180
@@ -232,7 +232,7 @@ jobs:
             release/*
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.18
+        uses: mxschmitt/action-tmate@v3.19
         with:
           detached: true
           connect-timeout-seconds: 180
@@ -294,7 +294,7 @@ jobs:
           export C_INCLUDE_PATH=/usr/local/include
           export CPLUS_INCLUDE_PATH=/usr/local/include
           export PATH=$PATH:$GOPATH/bin
-
+          export SKIP_GRPC_BACKEND=backend-assets/grpc/whisper
           make dist
       - uses: actions/upload-artifact@v4
         with:
@@ -308,7 +308,7 @@ jobs:
             release/*
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.18
+        uses: mxschmitt/action-tmate@v3.19
         with:
           detached: true
           connect-timeout-seconds: 180
@@ -327,7 +327,7 @@ jobs:
           cache: false
       - name: Dependencies
         run: |
-          brew install protobuf grpc
+          brew install protobuf grpc libomp llvm
           go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
           go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
       - name: Build
@@ -336,7 +336,7 @@ jobs:
           export C_INCLUDE_PATH=/usr/local/include
           export CPLUS_INCLUDE_PATH=/usr/local/include
           export PATH=$PATH:$GOPATH/bin
-
+          export CC=/opt/homebrew/opt/llvm/bin/clang
           make dist
       - uses: actions/upload-artifact@v4
         with:
@@ -350,7 +350,7 @@ jobs:
             release/*
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.18
+        uses: mxschmitt/action-tmate@v3.19
         with:
           detached: true
           connect-timeout-seconds: 180

.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-extra.yml

@@ -123,6 +123,13 @@ jobs:
         run: |
            make --jobs=5 --output-sync=target -C backend/python/parler-tts
            make --jobs=5 --output-sync=target -C backend/python/parler-tts test
+      - name: Setup tmate session if tests fail
+        if: ${{ failure() }}
+        uses: mxschmitt/action-tmate@v3.19
+        with:
+          detached: true
+          connect-timeout-seconds: 180
+          limit-access-to-actor: true
 
   tests-openvoice:
     runs-on: ubuntu-latest

.github/workflows/test.yml

@@ -133,7 +133,7 @@ jobs:
           PATH="$PATH:/root/go/bin" GO_TAGS="stablediffusion tts" make --jobs 5 --output-sync=target test
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.18
+        uses: mxschmitt/action-tmate@v3.19
         with:
           detached: true
           connect-timeout-seconds: 180
@@ -178,17 +178,26 @@ jobs:
         uses: actions/checkout@v4
         with:
           submodules: true
+      - name: Dependencies
+        run: |
+          # Install protoc
+          curl -L -s https://github.com/protocolbuffers/protobuf/releases/download/v26.1/protoc-26.1-linux-x86_64.zip -o protoc.zip && \
+          unzip -j -d /usr/local/bin protoc.zip bin/protoc && \
+          rm protoc.zip
+          go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
+          go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
+          PATH="$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: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.18
+        uses: mxschmitt/action-tmate@v3.19
         with:
           detached: true
           connect-timeout-seconds: 180
@@ -214,18 +223,19 @@ jobs:
         run: go version
       - name: Dependencies
         run: |
-          brew install protobuf grpc make protoc-gen-go protoc-gen-go-grpc
-          pip install --user --no-cache-dir grpcio-tools==1.64.1
+          brew install protobuf grpc make protoc-gen-go protoc-gen-go-grpc libomp llvm
+          pip install --user --no-cache-dir grpcio-tools
       - name: Test
         run: |
           export C_INCLUDE_PATH=/usr/local/include
           export CPLUS_INCLUDE_PATH=/usr/local/include
+          export CC=/opt/homebrew/opt/llvm/bin/clang
           # Used to run the newer GNUMake version from brew that supports --output-sync
           export PATH="/opt/homebrew/opt/make/libexec/gnubin:$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
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.18
+        uses: mxschmitt/action-tmate@v3.19
         with:
           detached: true
           connect-timeout-seconds: 180

.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

.gitignore

@@ -12,7 +12,6 @@ prepare-sources
 
 go-ggml-transformers
 go-gpt2
-go-rwkv
 whisper.cpp
 /bloomz
 go-bert

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
@@ -71,7 +85,8 @@ WORKDIR /build
 # The requirements-extras target is for any builds with IMAGE_TYPE=extras. It should not be placed in this target unless every IMAGE_TYPE=extras build will use it
 FROM requirements-core AS requirements-extras
 
-RUN curl -LsSf https://astral.sh/uv/install.sh | sh
+# Install uv as a system package
+RUN curl -LsSf https://astral.sh/uv/install.sh | UV_INSTALL_DIR=/usr/bin sh
 ENV PATH="/root/.cargo/bin:${PATH}"
 
 RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
@@ -188,6 +203,8 @@ 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 +213,24 @@ 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 +292,20 @@ 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 +320,20 @@ 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 +367,8 @@ 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 +446,6 @@ 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

@@ -8,15 +8,11 @@ DETECT_LIBS?=true
 # llama.cpp versions
 GOLLAMA_REPO?=https://github.com/go-skynet/go-llama.cpp
 GOLLAMA_VERSION?=2b57a8ae43e4699d3dc5d1496a1ccd42922993be
-CPPLLAMA_VERSION?=fc54ef0d1c138133a01933296d50a36a1ab64735
-
-# go-rwkv version
-RWKV_REPO?=https://github.com/donomii/go-rwkv.cpp
-RWKV_VERSION?=661e7ae26d442f5cfebd2a0881b44e8c55949ec6
+CPPLLAMA_VERSION?=47f931c8f9a26c072d71224bc8013cc66ea9e445
 
 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggerganov/whisper.cpp
-WHISPER_CPP_VERSION?=9e3c5345cd46ea718209db53464e426c3fe7a25e
+WHISPER_CPP_VERSION?=6266a9f9e56a5b925e9892acf650f3eb1245814d
 
 # bert.cpp version
 BERT_REPO?=https://github.com/go-skynet/go-bert.cpp
@@ -24,7 +20,7 @@ BERT_VERSION?=710044b124545415f555e4260d16b146c725a6e4
 
 # go-piper version
 PIPER_REPO?=https://github.com/mudler/go-piper
-PIPER_VERSION?=9d0100873a7dbb0824dfea40e8cec70a1b110759
+PIPER_VERSION?=e10ca041a885d4a8f3871d52924b47792d5e5aa0
 
 # stablediffusion version
 STABLEDIFFUSION_REPO?=https://github.com/mudler/go-stable-diffusion
@@ -34,6 +30,10 @@ STABLEDIFFUSION_VERSION?=4a3cd6aeae6f66ee57eae9a0075f8c58c3a6a38f
 TINYDREAM_REPO?=https://github.com/M0Rf30/go-tiny-dream
 TINYDREAM_VERSION?=c04fa463ace9d9a6464313aa5f9cd0f953b6c057
 
+ONNX_VERSION?=1.20.0
+ONNX_ARCH?=x64
+ONNX_OS?=linux
+
 export BUILD_TYPE?=
 export STABLE_BUILD_TYPE?=$(BUILD_TYPE)
 export CMAKE_ARGS?=
@@ -45,6 +45,7 @@ CGO_LDFLAGS_WHISPER+=-lggml
 CUDA_LIBPATH?=/usr/local/cuda/lib64/
 GO_TAGS?=
 BUILD_ID?=
+NATIVE?=false
 
 TEST_DIR=/tmp/test
 
@@ -83,7 +84,25 @@ ifndef UNAME_S
 UNAME_S := $(shell uname -s)
 endif
 
+# IF native is false, we add -DGGML_NATIVE=OFF to CMAKE_ARGS
+ifeq ($(NATIVE),false)
+	CMAKE_ARGS+=-DGGML_NATIVE=OFF
+endif
+
+# Detect if we are running on arm64
+ifneq (,$(findstring aarch64,$(shell uname -m)))
+	ONNX_ARCH=aarch64
+endif
+
 ifeq ($(OS),Darwin)
+	ONNX_OS=osx
+	ifneq (,$(findstring aarch64,$(shell uname -m)))
+		ONNX_ARCH=arm64
+	else ifneq (,$(findstring arm64,$(shell uname -m)))
+		ONNX_ARCH=arm64
+	else
+		ONNX_ARCH=x86_64
+	endif
 
 	ifeq ($(OSX_SIGNING_IDENTITY),)
 		OSX_SIGNING_IDENTITY := $(shell security find-identity -v -p codesigning | grep '"' | head -n 1 | sed -E 's/.*"(.*)"/\1/')
@@ -138,10 +157,10 @@ ifeq ($(BUILD_TYPE),hipblas)
 	export CC=$(ROCM_HOME)/llvm/bin/clang
 	# llama-ggml has no hipblas support, so override it here.
 	export STABLE_BUILD_TYPE=
-	export GGML_HIPBLAS=1
+	export GGML_HIP=1
 	GPU_TARGETS ?= gfx900,gfx906,gfx908,gfx940,gfx941,gfx942,gfx90a,gfx1030,gfx1031,gfx1100,gfx1101
 	AMDGPU_TARGETS ?= "$(GPU_TARGETS)"
-	CMAKE_ARGS+=-DGGML_HIPBLAS=ON -DAMDGPU_TARGETS="$(AMDGPU_TARGETS)" -DGPU_TARGETS="$(GPU_TARGETS)"
+	CMAKE_ARGS+=-DGGML_HIP=ON -DAMDGPU_TARGETS="$(AMDGPU_TARGETS)" -DGPU_TARGETS="$(GPU_TARGETS)"
 	CGO_LDFLAGS += -O3 --rtlib=compiler-rt -unwindlib=libgcc -lhipblas -lrocblas --hip-link -L${ROCM_HOME}/lib/llvm/lib
 endif
 
@@ -186,9 +205,9 @@ ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-fallback
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-ggml
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-grpc
 ALL_GRPC_BACKENDS+=backend-assets/util/llama-cpp-rpc-server
-ALL_GRPC_BACKENDS+=backend-assets/grpc/rwkv
 ALL_GRPC_BACKENDS+=backend-assets/grpc/whisper
 ALL_GRPC_BACKENDS+=backend-assets/grpc/local-store
+ALL_GRPC_BACKENDS+=backend-assets/grpc/silero-vad
 ALL_GRPC_BACKENDS+=$(OPTIONAL_GRPC)
 # Use filter-out to remove the specified backends
 ALL_GRPC_BACKENDS := $(filter-out $(SKIP_GRPC_BACKEND),$(ALL_GRPC_BACKENDS))
@@ -248,20 +267,6 @@ sources/go-piper:
 sources/go-piper/libpiper_binding.a: sources/go-piper
 	$(MAKE) -C sources/go-piper libpiper_binding.a example/main piper.o
 
-
-## RWKV
-sources/go-rwkv.cpp:
-	mkdir -p sources/go-rwkv.cpp
-	cd sources/go-rwkv.cpp && \
-	git init && \
-	git remote add origin $(RWKV_REPO) && \
-	git fetch origin && \
-	git checkout $(RWKV_VERSION) && \
-	git submodule update --init --recursive --depth 1 --single-branch
-
-sources/go-rwkv.cpp/librwkv.a: sources/go-rwkv.cpp
-	cd sources/go-rwkv.cpp && cd rwkv.cpp &&	cmake . -DRWKV_BUILD_SHARED_LIBRARY=OFF &&	cmake --build . && 	cp librwkv.a ..
-
 ## stable diffusion
 sources/go-stable-diffusion:
 	mkdir -p sources/go-stable-diffusion
@@ -275,6 +280,20 @@ sources/go-stable-diffusion:
 sources/go-stable-diffusion/libstablediffusion.a: sources/go-stable-diffusion
 	CPATH="$(CPATH):/usr/include/opencv4" $(MAKE) -C sources/go-stable-diffusion libstablediffusion.a
 
+sources/onnxruntime:
+	mkdir -p sources/onnxruntime
+	curl -L https://github.com/microsoft/onnxruntime/releases/download/v$(ONNX_VERSION)/onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz -o sources/onnxruntime/onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz
+	cd sources/onnxruntime && tar -xvf onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz && rm onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz
+	cd sources/onnxruntime && mv onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION)/* ./
+
+backend-assets/lib/libonnxruntime.so.1: backend-assets/lib sources/onnxruntime
+	cp -rfv sources/onnxruntime/lib/* backend-assets/lib/
+ifeq ($(OS),Darwin)
+	mv backend-assets/lib/libonnxruntime.$(ONNX_VERSION).dylib backend-assets/lib/libonnxruntime.dylib
+else
+	mv backend-assets/lib/libonnxruntime.so.$(ONNX_VERSION) backend-assets/lib/libonnxruntime.so.1
+endif
+
 ## tiny-dream
 sources/go-tiny-dream:
 	mkdir -p sources/go-tiny-dream
@@ -301,10 +320,9 @@ sources/whisper.cpp:
 sources/whisper.cpp/libwhisper.a: sources/whisper.cpp
 	cd sources/whisper.cpp && $(MAKE) libwhisper.a libggml.a
 
-get-sources: sources/go-llama.cpp sources/go-piper sources/go-rwkv.cpp sources/whisper.cpp sources/go-bert.cpp sources/go-stable-diffusion sources/go-tiny-dream backend/cpp/llama/llama.cpp
+get-sources: sources/go-llama.cpp sources/go-piper sources/whisper.cpp sources/go-bert.cpp sources/go-stable-diffusion sources/go-tiny-dream backend/cpp/llama/llama.cpp
 
 replace:
-	$(GOCMD) mod edit -replace github.com/donomii/go-rwkv.cpp=$(CURDIR)/sources/go-rwkv.cpp
 	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp=$(CURDIR)/sources/whisper.cpp
 	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp/bindings/go=$(CURDIR)/sources/whisper.cpp/bindings/go
 	$(GOCMD) mod edit -replace github.com/go-skynet/go-bert.cpp=$(CURDIR)/sources/go-bert.cpp
@@ -314,7 +332,6 @@ replace:
 	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(CURDIR)/sources/go-llama.cpp
 
 dropreplace:
-	$(GOCMD) mod edit -dropreplace github.com/donomii/go-rwkv.cpp
 	$(GOCMD) mod edit -dropreplace github.com/ggerganov/whisper.cpp
 	$(GOCMD) mod edit -dropreplace github.com/ggerganov/whisper.cpp/bindings/go
 	$(GOCMD) mod edit -dropreplace github.com/go-skynet/go-bert.cpp
@@ -330,7 +347,6 @@ prepare-sources: get-sources replace
 rebuild: ## Rebuilds the project
 	$(GOCMD) clean -cache
 	$(MAKE) -C sources/go-llama.cpp clean
-	$(MAKE) -C sources/go-rwkv.cpp clean
 	$(MAKE) -C sources/whisper.cpp clean
 	$(MAKE) -C sources/go-stable-diffusion clean
 	$(MAKE) -C sources/go-bert.cpp clean
@@ -359,6 +375,9 @@ 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})
@@ -436,8 +455,6 @@ test-models/testmodel.ggml:
 	wget -q https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin -O test-models/whisper-en
 	wget -q https://huggingface.co/mudler/all-MiniLM-L6-v2/resolve/main/ggml-model-q4_0.bin -O test-models/bert
 	wget -q https://cdn.openai.com/whisper/draft-20220913a/micro-machines.wav -O test-dir/audio.wav
-	wget -q https://huggingface.co/mudler/rwkv-4-raven-1.5B-ggml/resolve/main/RWKV-4-Raven-1B5-v11-Eng99%2525-Other1%2525-20230425-ctx4096_Q4_0.bin -O test-models/rwkv
-	wget -q https://raw.githubusercontent.com/saharNooby/rwkv.cpp/5eb8f09c146ea8124633ab041d9ea0b1f1db4459/rwkv/20B_tokenizer.json -O test-models/rwkv.tokenizer.json
 	cp tests/models_fixtures/* test-models
 
 prepare-test: grpcs
@@ -465,15 +482,15 @@ 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 +498,24 @@ 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 +551,10 @@ 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 +588,6 @@ 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 +684,6 @@ 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
@@ -767,7 +775,7 @@ backend-assets/grpc/llama-cpp-fallback: backend-assets/grpc backend/cpp/llama/ll
 	cp -rfv backend/cpp/llama-fallback/grpc-server backend-assets/grpc/llama-cpp-fallback
 # TODO: every binary should have its own folder instead, so can have different metal implementations
 ifeq ($(BUILD_TYPE),metal)
-	cp backend/cpp/llama-fallback/llama.cpp/build/bin/default.metallib backend-assets/grpc/
+	cp backend/cpp/llama-fallback/llama.cpp/build/bin/ggml-metal.metal backend-assets/grpc/
 endif
 
 backend-assets/grpc/llama-cpp-cuda: backend-assets/grpc backend/cpp/llama/llama.cpp
@@ -781,7 +789,7 @@ backend-assets/grpc/llama-cpp-hipblas: backend-assets/grpc backend/cpp/llama/lla
 	cp -rf backend/cpp/llama backend/cpp/llama-hipblas
 	$(MAKE) -C backend/cpp/llama-hipblas purge
 	$(info ${GREEN}I llama-cpp build info:hipblas${RESET})
-	BUILD_TYPE="hipblas" $(MAKE) VARIANT="llama-hipblas" build-llama-cpp-grpc-server
+	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off" BUILD_TYPE="hipblas" $(MAKE) VARIANT="llama-hipblas" build-llama-cpp-grpc-server
 	cp -rfv backend/cpp/llama-hipblas/grpc-server backend-assets/grpc/llama-cpp-hipblas
 
 backend-assets/grpc/llama-cpp-sycl_f16: backend-assets/grpc backend/cpp/llama/llama.cpp
@@ -823,13 +831,6 @@ ifneq ($(UPX),)
 	$(UPX) backend-assets/grpc/piper
 endif
 
-backend-assets/grpc/rwkv: sources/go-rwkv.cpp sources/go-rwkv.cpp/librwkv.a backend-assets/grpc
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-rwkv.cpp LIBRARY_PATH=$(CURDIR)/sources/go-rwkv.cpp \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/rwkv ./backend/go/llm/rwkv
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/rwkv
-endif
-
 backend-assets/grpc/stablediffusion: sources/go-stable-diffusion sources/go-stable-diffusion/libstablediffusion.a backend-assets/grpc
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" CPATH="$(CPATH):$(CURDIR)/sources/go-stable-diffusion/:/usr/include/opencv4" LIBRARY_PATH=$(CURDIR)/sources/go-stable-diffusion/ \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/stablediffusion ./backend/go/image/stablediffusion
@@ -837,6 +838,13 @@ ifneq ($(UPX),)
 	$(UPX) backend-assets/grpc/stablediffusion
 endif
 
+backend-assets/grpc/silero-vad: backend-assets/grpc backend-assets/lib/libonnxruntime.so.1
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" CPATH="$(CPATH):$(CURDIR)/sources/onnxruntime/include/" LIBRARY_PATH=$(CURDIR)/backend-assets/lib \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/silero-vad ./backend/go/vad/silero
+ifneq ($(UPX),)
+	$(UPX) backend-assets/grpc/silero-vad
+endif
+
 backend-assets/grpc/tinydream: sources/go-tiny-dream sources/go-tiny-dream/libtinydream.a backend-assets/grpc
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" LIBRARY_PATH=$(CURDIR)/go-tiny-dream \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/tinydream ./backend/go/image/tinydream
@@ -846,7 +854,7 @@ 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
@@ -897,7 +905,7 @@ docker-aio-all:
 
 docker-image-intel:
 	docker build \
-		--build-arg BASE_IMAGE=intel/oneapi-basekit:2024.2.0-devel-ubuntu22.04 \
+		--build-arg BASE_IMAGE=intel/oneapi-basekit:2025.0.0-0-devel-ubuntu22.04 \
 		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
 		--build-arg GO_TAGS="none" \
 		--build-arg MAKEFLAGS="$(DOCKER_MAKEFLAGS)" \
@@ -905,7 +913,7 @@ docker-image-intel:
 
 docker-image-intel-xpu:
 	docker build \
-		--build-arg BASE_IMAGE=intel/oneapi-basekit:2024.2.0-devel-ubuntu22.04 \
+		--build-arg BASE_IMAGE=intel/oneapi-basekit:2025.0.0-0-devel-ubuntu22.04 \
 		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
 		--build-arg GO_TAGS="none" \
 		--build-arg MAKEFLAGS="$(DOCKER_MAKEFLAGS)" \

README.md

@@ -38,9 +38,13 @@
 </a>
 </p>
 
+<p align="center">
+<a href="https://trendshift.io/repositories/1484" target="_blank"><img src="https://trendshift.io/api/badge/repositories/1484" alt="go-skynet%2FLocalAI | Trendshift" style="width: 250px; height: 55px;" width="250" height="55"/></a>
+</p>
+
 > :bulb: Get help - [❓FAQ](https://localai.io/faq/) [💭Discussions](https://github.com/go-skynet/LocalAI/discussions) [:speech_balloon: Discord](https://discord.gg/uJAeKSAGDy) [:book: Documentation website](https://localai.io/)
 >
-> [💻 Quickstart](https://localai.io/basics/getting_started/) [📣 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/mudler/LocalAI-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)
 
@@ -56,22 +60,41 @@ curl https://localai.io/install.sh | sh
 
 Or run with docker:
 ```bash
+# CPU only image:
+docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-cpu
+
+# Nvidia GPU:
+docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-gpu-nvidia-cuda-12
+
+# CPU and GPU image (bigger size):
+docker run -ti --name local-ai -p 8080:8080 localai/localai:latest
+
+# AIO images (it will pre-download a set of models ready for use, see https://localai.io/basics/container/)
 docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-aio-cpu
-# Alternative images:
-# - if you have an Nvidia GPU:
-# docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-aio-gpu-nvidia-cuda-12
-# - without preconfigured models
-# docker run -ti --name local-ai -p 8080:8080 localai/localai:latest
-# - without preconfigured models for Nvidia GPUs
-# docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-gpu-nvidia-cuda-12 
+```
+
+To load models:
+
+```bash
+# From the model gallery (see available models with `local-ai models list`, in the WebUI from the model tab, or visiting https://models.localai.io)
+local-ai run llama-3.2-1b-instruct:q4_k_m
+# Start LocalAI with the phi-2 model directly from huggingface
+local-ai run huggingface://TheBloke/phi-2-GGUF/phi-2.Q8_0.gguf
+# Install and run a model from the Ollama OCI registry
+local-ai run ollama://gemma:2b
+# Run a model from a configuration file
+local-ai run https://gist.githubusercontent.com/.../phi-2.yaml
+# Install and run a model from a standard OCI registry (e.g., Docker Hub)
+local-ai run oci://localai/phi-2:latest
 ```
 
 [💻 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
 
+- Nov 2024: Voice activity detection models (**VAD**) added to the API: https://github.com/mudler/LocalAI/pull/4204
+- Oct 2024: examples moved to [LocalAI-examples](https://github.com/mudler/LocalAI-examples)
+- Aug 2024:  🆕 FLUX-1, [P2P Explorer](https://explorer.localai.io)
 - July 2024: 🔥🔥 🆕 P2P Dashboard, LocalAI Federated mode and AI Swarms: https://github.com/mudler/LocalAI/pull/2723
 - June 2024: 🆕 You can browse now the model gallery without LocalAI! Check out https://models.localai.io
 - June 2024: Support for models from OCI registries: https://github.com/mudler/LocalAI/pull/2628
@@ -82,8 +105,12 @@ 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
@@ -137,6 +164,9 @@ Other:
 - Slack bot https://github.com/mudler/LocalAGI/tree/main/examples/slack
 - Shell-Pilot(Interact with LLM using LocalAI models via pure shell scripts on your Linux or MacOS system) https://github.com/reid41/shell-pilot
 - Telegram bot https://github.com/mudler/LocalAI/tree/master/examples/telegram-bot
+- Another Telegram Bot https://github.com/JackBekket/Hellper
+- Auto-documentation https://github.com/JackBekket/Reflexia
+- Github bot which answer on issues, with code and documentation as context https://github.com/JackBekket/GitHelper
 - Github Actions: https://github.com/marketplace/actions/start-localai
 - Examples: https://github.com/mudler/LocalAI/tree/master/examples/
   
@@ -211,7 +241,6 @@ LocalAI couldn't have been built without the help of great software already avai
 - https://github.com/antimatter15/alpaca.cpp
 - https://github.com/EdVince/Stable-Diffusion-NCNN
 - https://github.com/ggerganov/whisper.cpp
-- https://github.com/saharNooby/rwkv.cpp
 - https://github.com/rhasspy/piper
 
 ## 🤗 Contributors

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 +26,21 @@ service Backend {
   rpc StoresFind(StoresFindOptions) returns (StoresFindResult) {}
 
   rpc Rerank(RerankRequest) returns (RerankResult) {}
+
+  rpc GetMetrics(MetricsRequest) returns (MetricsResponse);
+
+  rpc VAD(VADRequest) returns (VADResponse) {}
+}
+
+// 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 +149,9 @@ 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
@@ -202,6 +221,7 @@ message ModelOptions {
   int32  SwapSpace = 53;
   int32  MaxModelLen = 54;
   int32  TensorParallelSize = 55;
+  string LoadFormat = 58;
 
   string MMProj = 41;
 
@@ -215,6 +235,11 @@ message ModelOptions {
 
   bool FlashAttention = 56;
   bool NoKVOffload = 57;
+
+  string ModelPath = 59;
+
+  repeated string LoraAdapters = 60;
+  repeated float LoraScales = 61;
 }
 
 message Result {
@@ -270,6 +295,30 @@ message TTSRequest {
   optional string language = 5;
 }
 
+message VADRequest {
+  repeated float audio = 1;
+}
+
+message VADSegment {
+  float start = 1;
+  float end = 2;
+}
+
+message VADResponse {
+  repeated VADSegment segments = 1;
+}
+
+message SoundGenerationRequest {
+  string text = 1;
+  string model = 2;
+  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/Makefile

@@ -22,7 +22,7 @@ else ifeq ($(BUILD_TYPE),clblas)
 	CMAKE_ARGS+=-DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
 # If it's hipblas we do have also to set CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++ 
 else ifeq ($(BUILD_TYPE),hipblas)
-	CMAKE_ARGS+=-DGGML_HIPBLAS=ON
+	CMAKE_ARGS+=-DGGML_HIP=ON
 # If it's OSX, DO NOT embed the metal library - -DGGML_METAL_EMBED_LIBRARY=ON requires further investigation
 # But if it's OSX without metal, disable it here
 else ifeq ($(OS),Darwin)

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_params_sampling 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
@@ -391,6 +391,39 @@ struct llama_metrics {
     }
 };
 
+struct llava_embd_batch {
+    std::vector<llama_pos>      pos;
+    std::vector<int32_t>        n_seq_id;
+    std::vector<llama_seq_id>   seq_id_0;
+    std::vector<llama_seq_id *> seq_ids;
+    std::vector<int8_t>         logits;
+    llama_batch batch;
+    llava_embd_batch(float * embd, int32_t n_tokens, llama_pos pos_0, llama_seq_id seq_id) {
+        pos     .resize(n_tokens);
+        n_seq_id.resize(n_tokens);
+        seq_ids .resize(n_tokens + 1);
+        logits  .resize(n_tokens);
+        seq_id_0.resize(1);
+        seq_id_0[0] = seq_id;
+        seq_ids [n_tokens] = nullptr;
+        batch = {
+            /*n_tokens       =*/ n_tokens,
+            /*tokens         =*/ nullptr,
+            /*embd           =*/ embd,
+            /*pos            =*/ pos.data(),
+            /*n_seq_id       =*/ n_seq_id.data(),
+            /*seq_id         =*/ seq_ids.data(),
+            /*logits         =*/ logits.data(),
+        };
+        for (int i = 0; i < n_tokens; i++) {
+            batch.pos     [i] = pos_0 + i;
+            batch.n_seq_id[i] = 1;
+            batch.seq_id  [i] = seq_id_0.data();
+            batch.logits  [i] = false;
+        }
+    }
+};
+
 struct llama_server_context
 {
     llama_model *model = nullptr;
@@ -398,7 +431,7 @@ struct llama_server_context
 
     clip_ctx *clp_ctx = nullptr;
 
-    gpt_params params;
+    common_params params;
 
     llama_batch batch;
 
@@ -441,7 +474,7 @@ 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 +482,7 @@ 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 +491,12 @@ 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 +504,7 @@ 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 +523,21 @@ 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 +611,12 @@ 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 +633,7 @@ 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 +662,7 @@ 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_params_sampling default_sparams;
  
         slot->params.stream             = json_value(data, "stream",            false);
         slot->params.cache_prompt       = json_value(data, "cache_prompt",      false);
@@ -627,8 +670,7 @@ struct llama_server_context
         slot->sparams.top_k             = json_value(data, "top_k",             default_sparams.top_k);
         slot->sparams.top_p             = json_value(data, "top_p",             default_sparams.top_p);
         slot->sparams.min_p             = json_value(data, "min_p",             default_sparams.min_p);
-        slot->sparams.tfs_z             = json_value(data, "tfs_z",             default_sparams.tfs_z);
-        slot->sparams.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 +683,7 @@ 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 +707,7 @@ struct llama_server_context
             slot->params.input_prefix = "";
         }
 
+
         if (data.count("input_suffix") != 0)
         {
             slot->params.input_suffix = data["input_suffix"];
@@ -683,6 +726,10 @@ 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 +765,10 @@ 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 +796,21 @@ 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 +824,22 @@ 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 +855,11 @@ 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 +897,12 @@ 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 +917,10 @@ 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 +930,7 @@ 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 +946,13 @@ 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)
@@ -924,11 +966,10 @@ struct llama_server_context
                     batch.n_seq_id + i,
                     batch.seq_id   + i,
                     batch.logits   + i,
-                    0, 0, 0, // unused
                 };
                 if (llama_decode(ctx, batch_view) != 0)
                 {
-                    LOG_TEE("%s: llama_decode() failed\n", __func__);
+                    LOG("%s: llama_decode() failed\n", __func__);
                     return;
                 }
             }
@@ -940,7 +981,7 @@ struct llama_server_context
             }
         }
 
-        LOG_TEE("system prompt updated\n");
+        LOG("system prompt updated\n");
         system_need_update = false;
     }
 
@@ -999,18 +1040,20 @@ 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 +1162,8 @@ 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 +1175,7 @@ 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 +1187,11 @@ 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 {
@@ -1164,14 +1205,11 @@ struct llama_server_context
             {"top_k",             slot.sparams.top_k},
             {"top_p",             slot.sparams.top_p},
             {"min_p",             slot.sparams.min_p},
-            {"tfs_z",             slot.sparams.tfs_z},
-            {"typical_p",         slot.sparams.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 +1217,13 @@ 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 +1246,7 @@ 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 +1298,7 @@ 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
@@ -1371,11 +1409,10 @@ struct llama_server_context
                     batch.n_seq_id + i,
                     batch.seq_id   + i,
                     batch.logits   + i,
-                    0, 0, 0, // unused
                 };
                 if (llama_decode(ctx, batch_view))
                 {
-                    LOG_TEE("%s : failed to eval\n", __func__);
+                    LOG("%s : failed to eval\n", __func__);
                     return false;
                 }
             }
@@ -1390,17 +1427,18 @@ struct llama_server_context
                 }
 
                 const int n_embd = llama_n_embd(model);
-                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))
+                float * embd = img.image_embedding + i * n_embd;
+                llava_embd_batch llava_batch = llava_embd_batch(embd, n_eval, slot.n_past, 0);
+                if (llama_decode(ctx, llava_batch.batch))
                 {
-                    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 +1448,7 @@ 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 +1580,7 @@ struct llama_server_context
             update_system_prompt();
         }
 
-        llama_batch_clear(batch);
+        common_batch_clear(batch);
 
         if (all_slots_are_idle)
         {
@@ -1576,7 +1614,7 @@ 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 +1658,7 @@ 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 +1752,7 @@ 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 +1764,7 @@ 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 +1856,17 @@ 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 +1906,10 @@ 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 +1919,7 @@ 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;
                 }
@@ -1895,7 +1934,6 @@ struct llama_server_context
                 batch.n_seq_id + i,
                 batch.seq_id   + i,
                 batch.logits   + i,
-                0, 0, 0, // unused
             };
 
             const int ret = llama_decode(ctx, batch_view);
@@ -1905,11 +1943,11 @@ 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 +1972,9 @@ 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 +1984,14 @@ 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 +2038,7 @@ 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); }
 };
 
@@ -2070,7 +2103,6 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
     //     slot->params.n_predict        = json_value(data, "n_predict",         default_params.n_predict);
     //     slot->sparams.top_k           = json_value(data, "top_k",             default_sparams.top_k);
     //     slot->sparams.top_p           = json_value(data, "top_p",             default_sparams.top_p);
-    //     slot->sparams.tfs_z           = json_value(data, "tfs_z",             default_sparams.tfs_z);
     //     slot->sparams.typical_p       = json_value(data, "typical_p",         default_sparams.typical_p);
     //     slot->sparams.temp            = json_value(data, "temperature",       default_sparams.temp);
     //     slot->sparams.penalty_last_n  = json_value(data, "repeat_last_n",     default_sparams.penalty_last_n);
@@ -2094,7 +2126,6 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
     data["n_predict"] = predict->tokens() == 0 ? -1 : predict->tokens();
     data["top_k"] = predict->topk();
     data["top_p"] = predict->topp();
-    data["tfs_z"] = predict->tailfreesamplingz();
     data["typical_p"] = predict->typicalp();
     data["temperature"] = predict->temperature();
     data["repeat_last_n"] = predict->repeat();
@@ -2112,6 +2143,9 @@ 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++) {
@@ -2138,7 +2172,6 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
 //     llama.params.n_predict = predict->tokens() == 0 ? -1 : predict->tokens();
 //     llama.params.sparams.top_k = predict->topk();
 //     llama.params.sparams.top_p = predict->topp();
-//     llama.params.sparams.tfs_z = predict->tailfreesamplingz();
 //     llama.params.sparams.typical_p = predict->typicalp();
 //     llama.params.sparams.penalty_last_n = predict->repeat();
 //     llama.params.sparams.temp = predict->temperature();
@@ -2196,7 +2229,7 @@ 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 +2243,7 @@ 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
@@ -2266,6 +2299,7 @@ static void params_parse(const backend::ModelOptions* request,
     params.use_mmap = request->mmap();
     params.flash_attn = request->flashattention();
     params.no_kv_offload = request->nokvoffload();
+    params.ctx_shift = false; // We control context-shifting in any case (and we disable it as it could just lead to infinite loops)
 
     params.embedding = request->embeddings();
 
@@ -2304,7 +2338,7 @@ 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 +2384,11 @@ 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 +2412,12 @@ 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 +2457,31 @@ 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/llm/rwkv/rwkv.go

@@ -1,95 +0,0 @@
-package main
-
-// This is a wrapper to statisfy the GRPC service interface
-// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
-import (
-	"fmt"
-	"path/filepath"
-
-	"github.com/donomii/go-rwkv.cpp"
-	"github.com/mudler/LocalAI/pkg/grpc/base"
-	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
-)
-
-const tokenizerSuffix = ".tokenizer.json"
-
-type LLM struct {
-	base.SingleThread
-
-	rwkv *rwkv.RwkvState
-}
-
-func (llm *LLM) Load(opts *pb.ModelOptions) error {
-	tokenizerFile := opts.Tokenizer
-	if tokenizerFile == "" {
-		modelFile := filepath.Base(opts.ModelFile)
-		tokenizerFile = modelFile + tokenizerSuffix
-	}
-	modelPath := filepath.Dir(opts.ModelFile)
-	tokenizerPath := filepath.Join(modelPath, tokenizerFile)
-
-	model := rwkv.LoadFiles(opts.ModelFile, tokenizerPath, uint32(opts.GetThreads()))
-
-	if model == nil {
-		return fmt.Errorf("rwkv could not load model")
-	}
-	llm.rwkv = model
-	return nil
-}
-
-func (llm *LLM) Predict(opts *pb.PredictOptions) (string, error) {
-	stopWord := "\n"
-	if len(opts.StopPrompts) > 0 {
-		stopWord = opts.StopPrompts[0]
-	}
-
-	if err := llm.rwkv.ProcessInput(opts.Prompt); err != nil {
-		return "", err
-	}
-
-	response := llm.rwkv.GenerateResponse(int(opts.Tokens), stopWord, float32(opts.Temperature), float32(opts.TopP), nil)
-
-	return response, nil
-}
-
-func (llm *LLM) PredictStream(opts *pb.PredictOptions, results chan string) error {
-	go func() {
-
-		stopWord := "\n"
-		if len(opts.StopPrompts) > 0 {
-			stopWord = opts.StopPrompts[0]
-		}
-
-		if err := llm.rwkv.ProcessInput(opts.Prompt); err != nil {
-			fmt.Println("Error processing input: ", err)
-			return
-		}
-
-		llm.rwkv.GenerateResponse(int(opts.Tokens), stopWord, float32(opts.Temperature), float32(opts.TopP), func(s string) bool {
-			results <- s
-			return true
-		})
-		close(results)
-	}()
-
-	return nil
-}
-
-func (llm *LLM) TokenizeString(opts *pb.PredictOptions) (pb.TokenizationResponse, error) {
-	tokens, err := llm.rwkv.Tokenizer.Encode(opts.Prompt)
-	if err != nil {
-		return pb.TokenizationResponse{}, err
-	}
-
-	l := len(tokens)
-	i32Tokens := make([]int32, l)
-
-	for i, t := range tokens {
-		i32Tokens[i] = int32(t.ID)
-	}
-
-	return pb.TokenizationResponse{
-		Length: int32(l),
-		Tokens: i32Tokens,
-	}, nil
-}

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/go/vad/silero/main.go

@@ -15,7 +15,7 @@ var (
 func main() {
 	flag.Parse()
 
-	if err := grpc.StartServer(*addr, &LLM{}); err != nil {
+	if err := grpc.StartServer(*addr, &VAD{}); err != nil {
 		panic(err)
 	}
 }

backend/go/vad/silero/vad.go

@@ -0,0 +1,54 @@
+package main
+
+// This is a wrapper to statisfy the GRPC service interface
+// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
+import (
+	"fmt"
+
+	"github.com/mudler/LocalAI/pkg/grpc/base"
+	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
+	"github.com/streamer45/silero-vad-go/speech"
+)
+
+type VAD struct {
+	base.SingleThread
+	detector *speech.Detector
+}
+
+func (vad *VAD) Load(opts *pb.ModelOptions) error {
+	v, err := speech.NewDetector(speech.DetectorConfig{
+		ModelPath:  opts.ModelFile,
+		SampleRate: 16000,
+		//WindowSize:           1024,
+		Threshold:            0.5,
+		MinSilenceDurationMs: 0,
+		SpeechPadMs:          0,
+	})
+	if err != nil {
+		return fmt.Errorf("create silero detector: %w", err)
+	}
+
+	vad.detector = v
+	return err
+}
+
+func (vad *VAD) VAD(req *pb.VADRequest) (pb.VADResponse, error) {
+	audio := req.Audio
+
+	segments, err := vad.detector.Detect(audio)
+	if err != nil {
+		return pb.VADResponse{}, fmt.Errorf("detect: %w", err)
+	}
+
+	vadSegments := []*pb.VADSegment{}
+	for _, s := range segments {
+		vadSegments = append(vadSegments, &pb.VADSegment{
+			Start: float32(s.SpeechStartAt),
+			End:   float32(s.SpeechEndAt),
+		})
+	}
+
+	return pb.VADResponse{
+		Segments: vadSegments,
+	}, nil
+}

backend/python/autogptq/requirements-cublas11.txt

@@ -1,2 +1,2 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch
+torch==2.4.1+cu118

backend/python/autogptq/requirements-cublas12.txt

@@ -1 +1 @@
-torch
+torch==2.4.1

backend/python/autogptq/requirements-hipblas.txt

@@ -1,2 +1,2 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch
+torch==2.4.1+rocm6.0

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==1.66.0
+grpcio==1.68.0
 protobuf
 certifi
 transformers

backend/python/bark/requirements-cpu.txt

@@ -1,4 +1,4 @@
 transformers
 accelerate
-torch
-torchaudio
+torch==2.4.1
+torchaudio==2.4.1

backend/python/bark/requirements-cublas11.txt

@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch
-torchaudio
+torch==2.4.1+cu118
+torchaudio==2.4.1+cu118
 transformers
 accelerate

backend/python/bark/requirements-cublas12.txt

@@ -1,4 +1,4 @@
-torch
-torchaudio
+torch==2.4.1
+torchaudio==2.4.1
 transformers
 accelerate

backend/python/bark/requirements-hipblas.txt

@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch
-torchaudio
+torch==2.4.1+rocm6.0
+torchaudio==2.4.1+rocm6.0
 transformers
 accelerate

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==1.66.0
+grpcio==1.68.0
 protobuf
 certifi

backend/python/common/template/Makefile

@@ -1,8 +1,9 @@
 .DEFAULT_GOAL := install
 
 .PHONY: install
-install: protogen
+install:
 	bash install.sh
+	$(MAKE) protogen
 
 .PHONY: protogen
 protogen: backend_pb2_grpc.py backend_pb2.py
@@ -12,7 +13,7 @@ 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
+	bash protogen.sh
 
 .PHONY: clean
 clean: protogen-clean

backend/python/common/template/protogen.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+set -e
+
+source $(dirname $0)/../common/libbackend.sh
+
+python3 -m grpc_tools.protoc -I../.. --python_out=. --grpc_python_out=. backend.proto

backend/python/common/template/requirements.txt

@@ -1,2 +1,3 @@
-grpcio==1.66.0
-protobuf
+grpcio==1.68.0
+protobuf
+grpcio-tools

backend/python/coqui/requirements-cpu.txt

@@ -1,3 +1,4 @@
 transformers
 accelerate
-torch
+torch==2.4.1
+coqui-tts

backend/python/coqui/requirements-cublas11.txt

@@ -1,5 +1,6 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch
-torchaudio
+torch==2.4.1+cu118
+torchaudio==2.4.1+cu118
 transformers
-accelerate
+accelerate
+coqui-tts

backend/python/coqui/requirements-cublas12.txt

@@ -1,4 +1,5 @@
-torch
-torchaudio
+torch==2.4.1
+torchaudio==2.4.1
 transformers
-accelerate
+accelerate
+coqui-tts

backend/python/coqui/requirements-hipblas.txt

@@ -1,5 +1,6 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch
-torchaudio
+torch==2.4.1+rocm6.0
+torchaudio==2.4.1+rocm6.0
 transformers
-accelerate
+accelerate
+coqui-tts

backend/python/coqui/requirements-intel.txt

@@ -3,6 +3,7 @@ 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
+accelerate
+coqui-tts

backend/python/coqui/requirements.txt

@@ -1,4 +1,4 @@
-TTS==0.22.0
-grpcio==1.66.0
+grpcio==1.68.0
 protobuf
-certifi
+certifi
+packaging==24.1

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"
@@ -247,11 +247,16 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                         use_safetensors=True,
                         variant=variant)
             elif request.PipelineType == "FluxPipeline":
+                if fromSingleFile:
+                    self.pipe = FluxPipeline.from_single_file(modelFile,
+                                                              torch_dtype=torchType,
+                                                              use_safetensors=True)
+                else:
                     self.pipe = FluxPipeline.from_pretrained(
                         request.Model,
                         torch_dtype=torch.bfloat16)
-                    if request.LowVRAM:
-                        self.pipe.enable_model_cpu_offload()
+                if request.LowVRAM:
+                    self.pipe.enable_model_cpu_offload()
             elif request.PipelineType == "FluxTransformer2DModel":
                     dtype = torch.bfloat16
                     # specify from environment or default to "ChuckMcSneed/FLUX.1-dev"
@@ -296,22 +301,34 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                 self.pipe.controlnet = self.controlnet
             else:
                 self.controlnet = None
-            # Assume directory from request.ModelFile.
-            # Only if request.LoraAdapter it's not an absolute path
-            if request.LoraAdapter and request.ModelFile != "" and not os.path.isabs(request.LoraAdapter) and request.LoraAdapter:
-                # get base path of modelFile
-                modelFileBase = os.path.dirname(request.ModelFile)
+
+            if request.LoraAdapter and not os.path.isabs(request.LoraAdapter):
                 # modify LoraAdapter to be relative to modelFileBase
-                request.LoraAdapter = os.path.join(modelFileBase, request.LoraAdapter)
+                request.LoraAdapter = os.path.join(request.ModelPath, request.LoraAdapter)
+
             device = "cpu" if not request.CUDA else "cuda"
             self.device = device
             if request.LoraAdapter:
                 # Check if its a local file and not a directory ( we load lora differently for a safetensor file )
                 if os.path.exists(request.LoraAdapter) and not os.path.isdir(request.LoraAdapter):
-                    # self.load_lora_weights(request.LoraAdapter, 1, device, torchType)
                     self.pipe.load_lora_weights(request.LoraAdapter)
                 else:
                     self.pipe.unet.load_attn_procs(request.LoraAdapter)
+            if len(request.LoraAdapters) > 0:
+                i = 0
+                adapters_name = []
+                adapters_weights = []
+                for adapter in request.LoraAdapters:
+                    if not os.path.isabs(adapter):
+                        adapter = os.path.join(request.ModelPath, adapter)
+                    self.pipe.load_lora_weights(adapter, adapter_name=f"adapter_{i}")
+                    adapters_name.append(f"adapter_{i}")
+                    i += 1
+
+                for adapters_weight in request.LoraScales:
+                    adapters_weights.append(adapters_weight)
+
+                self.pipe.set_adapters(adapters_name, adapter_weights=adapters_weights)
 
             if request.CUDA:
                 self.pipe.to('cuda')
@@ -392,8 +409,6 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         # create a dictionary of values for the parameters
         options = {
             "negative_prompt": request.negative_prompt,
-            "width": request.width,
-            "height": request.height,
             "num_inference_steps": steps,
         }
 
@@ -411,13 +426,13 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         keys = options.keys()
 
         if request.EnableParameters != "":
-            keys = request.EnableParameters.split(",")
+            keys = [key.strip() for key in request.EnableParameters.split(",")]
 
         if request.EnableParameters == "none":
             keys = []
 
         # create a dictionary of parameters by using the keys from EnableParameters and the values from defaults
-        kwargs = {key: options[key] for key in keys}
+        kwargs = {key: options.get(key) for key in keys if key in options}
 
         # Set seed
         if request.seed > 0:
@@ -428,6 +443,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         if self.PipelineType == "FluxPipeline":
             kwargs["max_sequence_length"] = 256
 
+        if request.width:
+            kwargs["width"] = request.width
+
+        if request.height:
+            kwargs["height"] = request.height
+
         if self.PipelineType == "FluxTransformer2DModel":
             kwargs["output_type"] = "pil"
             kwargs["generator"] = torch.Generator("cpu").manual_seed(0)
@@ -447,6 +468,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             export_to_video(video_frames, request.dst)
             return backend_pb2.Result(message="Media generated successfully", success=True)
 
+        print(f"Generating image with {kwargs=}", file=sys.stderr)
         image = {}
         if COMPEL:
             conditioning, pooled = self.compel.build_conditioning_tensor(prompt)

backend/python/diffusers/requirements-cpu.txt

@@ -5,5 +5,5 @@ accelerate
 compel
 peft
 sentencepiece
-torch
+torch==2.4.1
 optimum-quanto

backend/python/diffusers/requirements-cublas11.txt

@@ -1,5 +1,5 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
-torch
+torch==2.4.1+cu118
 diffusers
 opencv-python
 transformers

backend/python/diffusers/requirements-cublas12.txt

@@ -1,4 +1,4 @@
-torch
+torch==2.4.1
 diffusers
 opencv-python
 transformers

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==1.66.0
+grpcio==1.68.0
 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==1.66.0
-protobuf
-certifi
-setuptools