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

* initial pass: roll out bluemonday sanitization more widely

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

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

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

---------

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

.devcontainer-scripts/utils.sh

@@ -9,6 +9,7 @@
 # Param 2: email
 #
 config_user() {
+    echo "Configuring git for $1 <$2>"
     local gcn=$(git config --global user.name)
     if [ -z "${gcn}" ]; then
         echo "Setting up git user / remote"
@@ -24,6 +25,7 @@ config_user() {
 # Param 2: remote url
 #
 config_remote() {
+    echo "Adding git remote and fetching $2 as $1"
     local gr=$(git remote -v | grep $1)
     if [ -z "${gr}" ]; then
         git remote add $1 $2

.github/check_and_update.py

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

.github/ci/modelslist.go

@@ -6,6 +6,7 @@ import (
 	"io/ioutil"
 	"os"
 
+	"github.com/microcosm-cc/bluemonday"
 	"gopkg.in/yaml.v3"
 )
 
@@ -279,6 +280,12 @@ func main() {
 		return
 	}
 
+	// Ensure that all arbitrary text content is sanitized before display
+	for i, m := range models {
+		models[i].Name = bluemonday.StrictPolicy().Sanitize(m.Name)
+		models[i].Description = bluemonday.StrictPolicy().Sanitize(m.Description)
+	}
+
 	// render the template
 	data := struct {
 		Models          []*GalleryModel

.github/workflows/deploy-explorer.yaml

@@ -33,7 +33,7 @@ jobs:
         run: |
           CGO_ENABLED=0 make build-api
       - name: rm
-        uses: appleboy/ssh-action@v1.0.3
+        uses: appleboy/ssh-action@v1.1.0
         with:
             host: ${{ secrets.EXPLORER_SSH_HOST }}
             username: ${{ secrets.EXPLORER_SSH_USERNAME }}
@@ -53,7 +53,7 @@ jobs:
             rm: true
             target: ./local-ai
       - name: restarting
-        uses: appleboy/ssh-action@v1.0.3
+        uses: appleboy/ssh-action@v1.1.0
         with:
             host: ${{ secrets.EXPLORER_SSH_HOST }}
             username: ${{ secrets.EXPLORER_SSH_USERNAME }}

.github/workflows/image.yml

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

.github/workflows/secscan.yaml

@@ -18,7 +18,7 @@ jobs:
         if: ${{ github.actor != 'dependabot[bot]' }}
       - name: Run Gosec Security Scanner
         if: ${{ github.actor != 'dependabot[bot]' }}
-        uses: securego/gosec@master
+        uses: securego/gosec@v2.21.4
         with:
           # we let the report trigger content trigger a failure using the GitHub Security features.
           args: '-no-fail -fmt sarif -out results.sarif ./...'

.github/workflows/test.yml

@@ -178,13 +178,22 @@ jobs:
         uses: actions/checkout@v4
         with:
           submodules: true
+      - name: Dependencies
+        run: |
+          # Install protoc
+          curl -L -s https://github.com/protocolbuffers/protobuf/releases/download/v26.1/protoc-26.1-linux-x86_64.zip -o protoc.zip && \
+          unzip -j -d /usr/local/bin protoc.zip bin/protoc && \
+          rm protoc.zip
+          go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
+          go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
+          PATH="$PATH:$HOME/go/bin" make protogen-go
       - name: Build images
         run: |
           docker build --build-arg FFMPEG=true --build-arg IMAGE_TYPE=extras --build-arg EXTRA_BACKENDS=rerankers --build-arg MAKEFLAGS="--jobs=5 --output-sync=target" -t local-ai:tests -f Dockerfile .
           BASE_IMAGE=local-ai:tests DOCKER_AIO_IMAGE=local-ai-aio:test make docker-aio
       - name: Test
         run: |
-          LOCALAI_MODELS_DIR=$PWD/models LOCALAI_IMAGE_TAG=test LOCALAI_IMAGE=local-ai-aio \
+            PATH="$PATH:$HOME/go/bin" LOCALAI_MODELS_DIR=$PWD/models LOCALAI_IMAGE_TAG=test LOCALAI_IMAGE=local-ai-aio \
             make run-e2e-aio
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}

CONTRIBUTING.md

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

Dockerfile

@@ -9,11 +9,13 @@ FROM ${BASE_IMAGE} AS requirements-core
 USER root
 
 ARG GO_VERSION=1.22.6
+ARG CMAKE_VERSION=3.26.4
+ARG CMAKE_FROM_SOURCE=false
 ARG TARGETARCH
 ARG TARGETVARIANT
 
 ENV DEBIAN_FRONTEND=noninteractive
-ENV EXTERNAL_GRPC_BACKENDS="coqui:/build/backend/python/coqui/run.sh,huggingface-embeddings:/build/backend/python/sentencetransformers/run.sh,transformers:/build/backend/python/transformers/run.sh,sentencetransformers:/build/backend/python/sentencetransformers/run.sh,rerankers:/build/backend/python/rerankers/run.sh,autogptq:/build/backend/python/autogptq/run.sh,bark:/build/backend/python/bark/run.sh,diffusers:/build/backend/python/diffusers/run.sh,exllama:/build/backend/python/exllama/run.sh,openvoice:/build/backend/python/openvoice/run.sh,vall-e-x:/build/backend/python/vall-e-x/run.sh,vllm:/build/backend/python/vllm/run.sh,mamba:/build/backend/python/mamba/run.sh,exllama2:/build/backend/python/exllama2/run.sh,transformers-musicgen:/build/backend/python/transformers-musicgen/run.sh,parler-tts:/build/backend/python/parler-tts/run.sh"
+ENV EXTERNAL_GRPC_BACKENDS="coqui:/build/backend/python/coqui/run.sh,huggingface-embeddings:/build/backend/python/sentencetransformers/run.sh,transformers:/build/backend/python/transformers/run.sh,sentencetransformers:/build/backend/python/sentencetransformers/run.sh,rerankers:/build/backend/python/rerankers/run.sh,autogptq:/build/backend/python/autogptq/run.sh,bark:/build/backend/python/bark/run.sh,diffusers:/build/backend/python/diffusers/run.sh,openvoice:/build/backend/python/openvoice/run.sh,vall-e-x:/build/backend/python/vall-e-x/run.sh,vllm:/build/backend/python/vllm/run.sh,mamba:/build/backend/python/mamba/run.sh,exllama2:/build/backend/python/exllama2/run.sh,transformers-musicgen:/build/backend/python/transformers-musicgen/run.sh,parler-tts:/build/backend/python/parler-tts/run.sh"
 
 
 RUN apt-get update && \
@@ -21,13 +23,25 @@ RUN apt-get update && \
         build-essential \
         ccache \
         ca-certificates \
-        cmake \
-        curl \
+        curl libssl-dev \
         git \
         unzip upx-ucl && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
+# Install CMake (the version in 22.04 is too old)
+RUN <<EOT bash
+    if [ "${CMAKE_FROM_SOURCE}}" = "true" ]; then
+        curl -L -s https://github.com/Kitware/CMake/releases/download/v${CMAKE_VERSION}/cmake-${CMAKE_VERSION}.tar.gz -o cmake.tar.gz && tar xvf cmake.tar.gz && cd cmake-${CMAKE_VERSION} && ./configure && make && make install
+    else
+        apt-get update && \
+        apt-get install -y \
+            cmake && \
+        apt-get clean && \
+        rm -rf /var/lib/apt/lists/*
+    fi
+EOT
+
 # Install Go
 RUN curl -L -s https://go.dev/dl/go${GO_VERSION}.linux-${TARGETARCH}.tar.gz | tar -C /usr/local -xz
 ENV PATH=$PATH:/root/go/bin:/usr/local/go/bin
@@ -188,6 +202,8 @@ FROM ${GRPC_BASE_IMAGE} AS grpc
 # This is a bit of a hack, but it's required in order to be able to effectively cache this layer in CI
 ARG GRPC_MAKEFLAGS="-j4 -Otarget"
 ARG GRPC_VERSION=v1.65.0
+ARG CMAKE_FROM_SOURCE=false
+ARG CMAKE_VERSION=3.26.4
 
 ENV MAKEFLAGS=${GRPC_MAKEFLAGS}
 
@@ -196,12 +212,24 @@ WORKDIR /build
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
         ca-certificates \
-        build-essential \
-        cmake \
+        build-essential curl libssl-dev \
         git && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
+# Install CMake (the version in 22.04 is too old)
+RUN <<EOT bash
+    if [ "${CMAKE_FROM_SOURCE}}" = "true" ]; then
+        curl -L -s https://github.com/Kitware/CMake/releases/download/v${CMAKE_VERSION}/cmake-${CMAKE_VERSION}.tar.gz -o cmake.tar.gz && tar xvf cmake.tar.gz && cd cmake-${CMAKE_VERSION} && ./configure && make && make install
+    else
+        apt-get update && \
+        apt-get install -y \
+            cmake && \
+        apt-get clean && \
+        rm -rf /var/lib/apt/lists/*
+    fi
+EOT
+
 # We install GRPC to a different prefix here so that we can copy in only the build artifacts later
 # saves several hundred MB on the final docker image size vs copying in the entire GRPC source tree
 # and running make install in the target container
@@ -297,10 +325,10 @@ COPY .git .
 RUN make prepare
 
 ## Build the binary
-## If it's CUDA, 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 build
-## (both will use CUDA for the actual computation)
-RUN if [ "${BUILD_TYPE}" = "cublas" ]; then \
+## If it's CUDA or hipblas, we want to skip some of the llama-compat backends to save space
+## We only leave the most CPU-optimized variant and the fallback for the cublas/hipblas build
+## (both will use CUDA or hipblas for the actual computation)
+RUN if [ "${BUILD_TYPE}" = "cublas" ] || [ "${BUILD_TYPE}" = "hipblas" ]; then \
         SKIP_GRPC_BACKEND="backend-assets/grpc/llama-cpp-avx backend-assets/grpc/llama-cpp-avx2" make build; \
     else \
         make build; \
@@ -338,9 +366,8 @@ RUN if [ "${FFMPEG}" = "true" ]; then \
 
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
-        ssh less && \
-    apt-get clean && \
-    rm -rf /var/lib/apt/lists/*
+        ssh less wget
+# For the devcontainer, leave apt functional in case additional devtools are needed at runtime.
 
 RUN go install github.com/go-delve/delve/cmd/dlv@latest
 
@@ -418,9 +445,6 @@ RUN if [[ ( "${EXTRA_BACKENDS}" =~ "coqui" || -z "${EXTRA_BACKENDS}" ) && "$IMAG
     ; fi && \
     if [[ ( "${EXTRA_BACKENDS}" =~ "transformers-musicgen" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
         make -C backend/python/transformers-musicgen \
-    ; fi && \
-    if [[ ( "${EXTRA_BACKENDS}" =~ "exllama1" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
-        make -C backend/python/exllama \
     ; fi
 
 RUN if [[ ( "${EXTRA_BACKENDS}" =~ "vall-e-x" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \

Makefile

@@ -8,7 +8,7 @@ DETECT_LIBS?=true
 # llama.cpp versions
 GOLLAMA_REPO?=https://github.com/go-skynet/go-llama.cpp
 GOLLAMA_VERSION?=2b57a8ae43e4699d3dc5d1496a1ccd42922993be
-CPPLLAMA_VERSION?=e6b7801bd189d102d901d3e72035611a25456ef1
+CPPLLAMA_VERSION?=96776405a17034dcfd53d3ddf5d142d34bdbb657
 
 # go-rwkv version
 RWKV_REPO?=https://github.com/donomii/go-rwkv.cpp
@@ -16,7 +16,7 @@ RWKV_VERSION?=661e7ae26d442f5cfebd2a0881b44e8c55949ec6
 
 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggerganov/whisper.cpp
-WHISPER_CPP_VERSION?=a551933542d956ae84634937acd2942eb40efaaf
+WHISPER_CPP_VERSION?=fdbfb460ed546452a5d53611bba66d10d842e719
 
 # bert.cpp version
 BERT_REPO?=https://github.com/go-skynet/go-bert.cpp
@@ -359,6 +359,9 @@ clean-tests:
 	rm -rf test-dir
 	rm -rf core/http/backend-assets
 
+clean-dc: clean
+	cp -r /build/backend-assets /workspace/backend-assets
+
 ## Build:
 build: prepare backend-assets grpcs ## Build the project
 	$(info ${GREEN}I local-ai build info:${RESET})
@@ -465,15 +468,15 @@ run-e2e-image:
 	ls -liah $(abspath ./tests/e2e-fixtures)
 	docker run -p 5390:8080 -e MODELS_PATH=/models -e THREADS=1 -e DEBUG=true -d --rm -v $(TEST_DIR):/models --gpus all --name e2e-tests-$(RANDOM) localai-tests
 
-run-e2e-aio:
+run-e2e-aio: protogen-go
 	@echo 'Running e2e AIO tests'
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts 5 -v -r ./tests/e2e-aio
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts $(TEST_FLAKES) -v -r ./tests/e2e-aio
 
 test-e2e:
 	@echo 'Running e2e tests'
 	BUILD_TYPE=$(BUILD_TYPE) \
 	LOCALAI_API=http://$(E2E_BRIDGE_IP):5390/v1 \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts 5 -v -r ./tests/e2e
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts $(TEST_FLAKES) -v -r ./tests/e2e
 
 teardown-e2e:
 	rm -rf $(TEST_DIR) || true
@@ -481,24 +484,24 @@ teardown-e2e:
 
 test-llama: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama" --flake-attempts 5 -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
 
 test-llama-gguf: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama-gguf" --flake-attempts 5 -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama-gguf" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
 
 test-tts: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="tts" --flake-attempts 1 -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="tts" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
 
 test-stablediffusion: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stablediffusion" --flake-attempts 1 -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stablediffusion" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
 
 test-stores: backend-assets/grpc/local-store
 	mkdir -p tests/integration/backend-assets/grpc
 	cp -f backend-assets/grpc/local-store tests/integration/backend-assets/grpc/
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stores" --flake-attempts 1 -v -r tests/integration
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stores" --flake-attempts $(TEST_FLAKES) -v -r tests/integration
 
 test-container:
 	docker build --target requirements -t local-ai-test-container .
@@ -534,10 +537,10 @@ protogen-go-clean:
 	$(RM) bin/*
 
 .PHONY: protogen-python
-protogen-python: autogptq-protogen bark-protogen coqui-protogen diffusers-protogen exllama-protogen exllama2-protogen mamba-protogen rerankers-protogen sentencetransformers-protogen transformers-protogen parler-tts-protogen transformers-musicgen-protogen vall-e-x-protogen vllm-protogen openvoice-protogen
+protogen-python: autogptq-protogen bark-protogen coqui-protogen diffusers-protogen exllama2-protogen mamba-protogen rerankers-protogen sentencetransformers-protogen transformers-protogen parler-tts-protogen transformers-musicgen-protogen vall-e-x-protogen vllm-protogen openvoice-protogen
 
 .PHONY: protogen-python-clean
-protogen-python-clean: autogptq-protogen-clean bark-protogen-clean coqui-protogen-clean diffusers-protogen-clean exllama-protogen-clean exllama2-protogen-clean mamba-protogen-clean sentencetransformers-protogen-clean rerankers-protogen-clean transformers-protogen-clean transformers-musicgen-protogen-clean parler-tts-protogen-clean vall-e-x-protogen-clean vllm-protogen-clean openvoice-protogen-clean
+protogen-python-clean: autogptq-protogen-clean bark-protogen-clean coqui-protogen-clean diffusers-protogen-clean  exllama2-protogen-clean mamba-protogen-clean sentencetransformers-protogen-clean rerankers-protogen-clean transformers-protogen-clean transformers-musicgen-protogen-clean parler-tts-protogen-clean vall-e-x-protogen-clean vllm-protogen-clean openvoice-protogen-clean
 
 .PHONY: autogptq-protogen
 autogptq-protogen:
@@ -571,14 +574,6 @@ diffusers-protogen:
 diffusers-protogen-clean:
 	$(MAKE) -C backend/python/diffusers protogen-clean
 
-.PHONY: exllama-protogen
-exllama-protogen:
-	$(MAKE) -C backend/python/exllama protogen
-
-.PHONY: exllama-protogen-clean
-exllama-protogen-clean:
-	$(MAKE) -C backend/python/exllama protogen-clean
-
 .PHONY: exllama2-protogen
 exllama2-protogen:
 	$(MAKE) -C backend/python/exllama2 protogen
@@ -675,7 +670,6 @@ prepare-extra-conda-environments: protogen-python
 	$(MAKE) -C backend/python/parler-tts
 	$(MAKE) -C backend/python/vall-e-x
 	$(MAKE) -C backend/python/openvoice
-	$(MAKE) -C backend/python/exllama
 	$(MAKE) -C backend/python/exllama2
 
 prepare-test-extra: protogen-python

README.md

@@ -68,9 +68,7 @@ docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-aio-cpu
 
 [💻 Getting started](https://localai.io/basics/getting_started/index.html)
 
-## 🔥🔥 Hot topics / Roadmap
-
-[Roadmap](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap)
+## 📰 Latest project news
 
 - Aug 2024:  🆕 FLUX-1, [P2P Explorer](https://explorer.localai.io)
 - July 2024: 🔥🔥 🆕 P2P Dashboard, LocalAI Federated mode and AI Swarms: https://github.com/mudler/LocalAI/pull/2723
@@ -83,8 +81,12 @@ docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-aio-cpu
 - May 2024: Chat, TTS, and Image generation in the WebUI: https://github.com/mudler/LocalAI/pull/2222
 - April 2024: Reranker API: https://github.com/mudler/LocalAI/pull/2121
 
-Hot topics (looking for contributors):
+Roadmap items: [List of issues](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap)
 
+## 🔥🔥 Hot topics (looking for help):
+
+- Multimodal with vLLM and Video understanding: https://github.com/mudler/LocalAI/pull/3729
+- Realtime API https://github.com/mudler/LocalAI/issues/3714
 - 🔥🔥 Distributed, P2P Global community pools: https://github.com/mudler/LocalAI/issues/3113
 - WebUI improvements: https://github.com/mudler/LocalAI/issues/2156
 - Backends v2: https://github.com/mudler/LocalAI/issues/1126

aio/cpu/vision.yaml

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

aio/gpu-8g/vision.yaml

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

aio/intel/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

@@ -26,6 +26,19 @@ service Backend {
   rpc StoresFind(StoresFindOptions) returns (StoresFindResult) {}
 
   rpc Rerank(RerankRequest) returns (RerankResult) {}
+
+  rpc GetMetrics(MetricsRequest) returns (MetricsResponse);
+}
+
+// Define the empty request
+message MetricsRequest {}
+
+message MetricsResponse {
+  int32 slot_id = 1;
+  string prompt_json_for_slot = 2;  // Stores the prompt as a JSON string.
+  float tokens_per_second = 3;
+  int32 tokens_generated = 4;
+  int32 prompt_tokens_processed = 5;
 }
 
 message RerankRequest {
@@ -134,6 +147,9 @@ message PredictOptions {
   repeated string Images = 42;
   bool UseTokenizerTemplate = 43;
   repeated Message Messages = 44;
+  repeated string Videos = 45;
+  repeated string Audios = 46;
+  string CorrelationId = 47;
 }
 
 // The response message containing the result

backend/cpp/llama/grpc-server.cpp

@@ -13,6 +13,7 @@
 #include <getopt.h>
 #include "clip.h"
 #include "llava.h"
+#include "log.h"
 #include "stb_image.h"
 #include "common.h"
 #include "json.hpp"
@@ -112,7 +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;
 }
@@ -120,7 +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)
@@ -202,8 +203,8 @@ struct llama_client_slot
     std::string stopping_word;
 
     // sampling
-    struct gpt_sampler_params sparams;
-    gpt_sampler *ctx_sampling = nullptr;
+    struct common_sampler_params sparams;
+    common_sampler *ctx_sampling = nullptr;
 
     int32_t ga_i = 0;   // group-attention state
     int32_t ga_n = 1;   // group-attention factor
@@ -256,7 +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
@@ -397,7 +398,7 @@ struct llama_server_context
 
     clip_ctx *clp_ctx = nullptr;
 
-    gpt_params params;
+    common_params params;
 
     llama_batch batch;
 
@@ -440,7 +441,7 @@ struct llama_server_context
         }
     }
 
-    bool load_model(const gpt_params &params_)
+    bool load_model(const common_params &params_)
     {
         params = params_;
         if (!params.mmproj.empty()) {
@@ -448,7 +449,7 @@ struct llama_server_context
             LOG_INFO("Multi Modal Mode Enabled", {});
             clp_ctx = clip_model_load(params.mmproj.c_str(), /*verbosity=*/ 1);
             if(clp_ctx == nullptr) {
-                LOG_ERROR("unable to load clip model", {{"model", params.mmproj}});
+                LOG_ERR("unable to load clip model: %s", params.mmproj.c_str());
                 return false;
             }
 
@@ -457,12 +458,12 @@ struct llama_server_context
             }
         }
 
-        llama_init_result llama_init = llama_init_from_gpt_params(params);
-        model = llama_init.model;
-        ctx = llama_init.context;
+        common_init_result common_init = common_init_from_params(params);
+        model = common_init.model;
+        ctx = common_init.context;
         if (model == nullptr)
         {
-            LOG_ERROR("unable to load model", {{"model", params.model}});
+            LOG_ERR("unable to load model: %s", params.model.c_str());
             return false;
         }
 
@@ -470,7 +471,7 @@ struct llama_server_context
             const int n_embd_clip = clip_n_mmproj_embd(clp_ctx);
             const int n_embd_llm  = llama_n_embd(model);
             if (n_embd_clip != n_embd_llm) {
-                LOG_TEE("%s: embedding dim of the multimodal projector (%d) is not equal to that of LLaMA (%d). Make sure that you use the correct mmproj file.\n", __func__, n_embd_clip, n_embd_llm);
+                LOG("%s: embedding dim of the multimodal projector (%d) is not equal to that of LLaMA (%d). Make sure that you use the correct mmproj file.\n", __func__, n_embd_clip, n_embd_llm);
                 llama_free(ctx);
                 llama_free_model(model);
                 return false;
@@ -489,11 +490,21 @@ struct llama_server_context
         std::vector<char> buf(1);
         int res = llama_chat_apply_template(model, nullptr, chat, 1, true, buf.data(), buf.size());
         if (res < 0) {
-            LOG_ERROR("The chat template comes with this model is not yet supported, falling back to chatml. This may cause the model to output suboptimal responses", {});
+            LOG_ERR("The chat template comes with this model is not yet supported, falling back to chatml. This may cause the model to output suboptimal responses", __func__);
             sparams.chat_template = "<|im_start|>"; // llama_chat_apply_template only checks if <|im_start|> exist in the template
         }
     }
 
+    llama_client_slot* get_active_slot() {
+        for (llama_client_slot& slot : slots) {
+            // Check if the slot is currently processing
+            if (slot.is_processing()) {
+                return &slot;  // Return the active slot
+            }
+        }
+        return nullptr;  // No active slot found
+    }
+
     void initialize() {
         // create slots
         all_slots_are_idle = true;
@@ -567,12 +578,12 @@ struct llama_server_context
                     std::vector<llama_token> p;
                     if (first)
                     {
-                        p = ::llama_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
+                        p = common_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
                         first = false;
                     }
                     else
                     {
-                        p = ::llama_tokenize(ctx, s, false, TMP_FORCE_SPECIAL);
+                        p = common_tokenize(ctx, s, false, TMP_FORCE_SPECIAL);
                     }
                     prompt_tokens.insert(prompt_tokens.end(), p.begin(), p.end());
                 }
@@ -589,7 +600,7 @@ struct llama_server_context
         else
         {
             auto s = json_prompt.template get<std::string>();
-            prompt_tokens = ::llama_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
+            prompt_tokens = common_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
         }
 
         return prompt_tokens;
@@ -618,7 +629,7 @@ struct llama_server_context
 
     bool launch_slot_with_data(llama_client_slot* &slot, json data) {
         slot_params default_params;
-        gpt_sampler_params default_sparams;
+        common_sampler_params default_sparams;
  
         slot->params.stream             = json_value(data, "stream",            false);
         slot->params.cache_prompt       = json_value(data, "cache_prompt",      false);
@@ -758,7 +769,7 @@ struct llama_server_context
                     }
                     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.push_back({tok, bias});
@@ -790,7 +801,7 @@ struct llama_server_context
                         sampler_names.emplace_back(name);
                     }
                 }
-                slot->sparams.samplers = gpt_sampler_types_from_names(sampler_names, false);
+                slot->sparams.samplers = common_sampler_types_from_names(sampler_names, false);
         }
         else
         {
@@ -812,10 +823,11 @@ struct llama_server_context
                     img_sl.img_data = clip_image_u8_init();
                     if (!clip_image_load_from_bytes(image_buffer.data(), image_buffer.size(), img_sl.img_data))
                     {
-                        LOG_ERROR("failed to load image", {
-                            {"slot_id",   slot->id},
-                            {"img_sl_id", img_sl.id}
-                        });
+                        LOG_ERR("%s: failed to load image, slot_id: %d, img_sl_id: %d", 
+                             __func__,
+                             slot->id,
+                             img_sl.id
+                        );
                         return false;
                     }
                     LOG_VERBOSE("image loaded", {
@@ -853,12 +865,12 @@ struct llama_server_context
                                     }
                                 }
                                 if (!found) {
-                                    LOG_TEE("ERROR: Image with id: %i, not found.\n", img_id);
+                                    LOG("ERROR: Image with id: %i, not found.\n", img_id);
                                     slot->images.clear();
                                     return false;
                                 }
                             } catch (const std::invalid_argument& e) {
-                                LOG_TEE("Invalid image number id in prompt\n");
+                                LOG("Invalid image number id in prompt\n");
                                 slot->images.clear();
                                 return false;
                             }
@@ -873,9 +885,9 @@ struct llama_server_context
 
         if (slot->ctx_sampling != nullptr)
         {
-            gpt_sampler_free(slot->ctx_sampling);
+            common_sampler_free(slot->ctx_sampling);
         }
-        slot->ctx_sampling = gpt_sampler_init(model, slot->sparams);
+        slot->ctx_sampling = common_sampler_init(model, slot->sparams);
         //llama_set_rng_seed(ctx, slot->params.seed);
         slot->command = LOAD_PROMPT;
 
@@ -886,7 +898,7 @@ struct llama_server_context
             {"task_id", slot->task_id},
         });
 
-      //  LOG_TEE("sampling: \n%s\n", llama_sampling_print(slot->sparams).c_str());
+      //  LOG("sampling: \n%s\n", llama_sampling_print(slot->sparams).c_str());
 
         return true;
     }
@@ -902,13 +914,13 @@ struct llama_server_context
         system_tokens.clear();
 
         if (!system_prompt.empty()) {
-            system_tokens = ::llama_tokenize(ctx, system_prompt, add_bos_token);
+            system_tokens = common_tokenize(ctx, system_prompt, add_bos_token);
 
-            llama_batch_clear(batch);
+            common_batch_clear(batch);
 
             for (int i = 0; i < (int)system_tokens.size(); ++i)
             {
-                llama_batch_add(batch, system_tokens[i], i, { 0 }, false);
+                common_batch_add(batch, system_tokens[i], i, { 0 }, false);
             }
 
             for (int32_t i = 0; i < (int32_t) batch.n_tokens; i += params.n_batch)
@@ -926,7 +938,7 @@ struct llama_server_context
                 };
                 if (llama_decode(ctx, batch_view) != 0)
                 {
-                    LOG_TEE("%s: llama_decode() failed\n", __func__);
+                    LOG("%s: llama_decode() failed\n", __func__);
                     return;
                 }
             }
@@ -938,7 +950,7 @@ struct llama_server_context
             }
         }
 
-        LOG_TEE("system prompt updated\n");
+        LOG("system prompt updated\n");
         system_need_update = false;
     }
 
@@ -997,7 +1009,7 @@ struct llama_server_context
 
     bool process_token(completion_token_output &result, llama_client_slot &slot) {
         // remember which tokens were sampled - used for repetition penalties during sampling
-        const std::string token_str = 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
@@ -1120,7 +1132,7 @@ struct llama_server_context
             }
 
             if (!llava_image_embed_make_with_clip_img(clp_ctx, params.cpuparams.n_threads, img.img_data, &img.image_embedding, &img.image_tokens)) {
-                LOG_TEE("Error processing the given image");
+                LOG("Error processing the given image");
                 return false;
             }
 
@@ -1132,7 +1144,7 @@ struct llama_server_context
 
     void send_error(task_server& task, const std::string &error)
     {
-        LOG_TEE("task %i - error: %s\n", task.id, error.c_str());
+        LOG("task %i - error: %s\n", task.id, error.c_str());
         task_result res;
         res.id = task.id;
         res.multitask_id = task.multitask_id;
@@ -1148,7 +1160,7 @@ struct llama_server_context
         samplers.reserve(slot.sparams.samplers.size());
         for (const auto & sampler : slot.sparams.samplers)
         {
-            samplers.emplace_back(gpt_sampler_type_to_str(sampler));
+            samplers.emplace_back(common_sampler_type_to_str(sampler));
         }
 
         return json {
@@ -1204,7 +1216,7 @@ struct llama_server_context
         if (slot.sparams.n_probs > 0)
         {
             std::vector<completion_token_output> probs_output = {};
-            const std::vector<llama_token> to_send_toks = llama_tokenize(ctx, tkn.text_to_send, false);
+            const std::vector<llama_token> to_send_toks = common_tokenize(ctx, tkn.text_to_send, false);
             size_t probs_pos      = std::min(slot.sent_token_probs_index,                       slot.generated_token_probs.size());
             size_t probs_stop_pos = std::min(slot.sent_token_probs_index + to_send_toks.size(), slot.generated_token_probs.size());
             if (probs_pos < probs_stop_pos)
@@ -1256,7 +1268,7 @@ struct llama_server_context
             std::vector<completion_token_output> probs = {};
             if (!slot.params.stream && slot.stopped_word)
             {
-                const std::vector<llama_token> stop_word_toks = llama_tokenize(ctx, slot.stopping_word, false);
+                const std::vector<llama_token> stop_word_toks = common_tokenize(ctx, slot.stopping_word, false);
                 probs = std::vector<completion_token_output>(slot.generated_token_probs.begin(), slot.generated_token_probs.end() - stop_word_toks.size());
             }
             else
@@ -1371,7 +1383,7 @@ struct llama_server_context
                 };
                 if (llama_decode(ctx, batch_view))
                 {
-                    LOG_TEE("%s : failed to eval\n", __func__);
+                    LOG("%s : failed to eval\n", __func__);
                     return false;
                 }
             }
@@ -1389,14 +1401,14 @@ struct llama_server_context
                 llama_batch batch_img = { n_eval, nullptr, (img.image_embedding + i * n_embd), nullptr, nullptr, nullptr, nullptr, slot.n_past, 1, 0, };
                 if (llama_decode(ctx, batch_img))
                 {
-                    LOG_TEE("%s : failed to eval image\n", __func__);
+                    LOG("%s : failed to eval image\n", __func__);
                     return false;
                 }
                 slot.n_past += n_eval;
             }
             image_idx++;
 
-            llama_batch_clear(batch);
+            common_batch_clear(batch);
 
             // append prefix of next image
             const auto json_prompt = (image_idx >= (int) slot.images.size()) ?
@@ -1406,7 +1418,7 @@ struct llama_server_context
             std::vector<llama_token> append_tokens = tokenize(json_prompt, false); // has next image
             for (int i = 0; i < (int) append_tokens.size(); ++i)
             {
-                llama_batch_add(batch, append_tokens[i], system_tokens.size() + slot.n_past, { slot.id }, true);
+                common_batch_add(batch, append_tokens[i], system_tokens.size() + slot.n_past, { slot.id }, true);
                 slot.n_past += 1;
             }
         }
@@ -1538,7 +1550,7 @@ struct llama_server_context
             update_system_prompt();
         }
 
-        llama_batch_clear(batch);
+        common_batch_clear(batch);
 
         if (all_slots_are_idle)
         {
@@ -1572,7 +1584,7 @@ struct llama_server_context
                     slot.n_past = 0;
                     slot.truncated = false;
                     slot.has_next_token = true;
-                    LOG_TEE("Context exhausted. Slot %d released (%d tokens in cache)\n", slot.id, (int) slot.cache_tokens.size());
+                    LOG("Context exhausted. Slot %d released (%d tokens in cache)\n", slot.id, (int) slot.cache_tokens.size());
 
                     continue;
                     // END LOCALAI changes
@@ -1616,7 +1628,7 @@ struct llama_server_context
 
             // TODO: we always have to take into account the "system_tokens"
             //       this is not great and needs to be improved somehow
-            llama_batch_add(batch, slot.sampled, system_tokens.size() + slot_npast, { slot.id }, true);
+            common_batch_add(batch, slot.sampled, system_tokens.size() + slot_npast, { slot.id }, true);
             slot.n_past += 1;
         }
 
@@ -1710,7 +1722,7 @@ struct llama_server_context
 
                     if (!slot.params.cache_prompt)
                     {
-                        gpt_sampler_reset(slot.ctx_sampling);
+                        common_sampler_reset(slot.ctx_sampling);
 
                         slot.n_past = 0;
                         slot.n_past_se = 0;
@@ -1722,7 +1734,7 @@ struct llama_server_context
                         // push the prompt into the sampling context (do not apply grammar)
                         for (auto &token : prompt_tokens)
                         {
-                            gpt_sampler_accept(slot.ctx_sampling, token, false);
+                            common_sampler_accept(slot.ctx_sampling, token, false);
                         }
 
                         slot.n_past = common_part(slot.cache_tokens, prompt_tokens);
@@ -1814,16 +1826,17 @@ struct llama_server_context
                                 ga_i += ga_w/ga_n;
                             }
                         }
-                        llama_batch_add(batch, prefix_tokens[slot.n_past], system_tokens.size() + slot_npast, {slot.id }, false);
+                        common_batch_add(batch, prefix_tokens[slot.n_past], system_tokens.size() + slot_npast, {slot.id }, false);
                         slot_npast++;
                     }
 
                     if (has_images && !ingest_images(slot, n_batch))
                     {
-                        LOG_ERROR("failed processing images", {
-                            "slot_id", slot.id,
-                            "task_id", slot.task_id,
-                        });
+                        LOG_ERR("%s: failed processing images Slot id : %d, Task id: %d", 
+                            __func__,
+                            slot.id,
+                            slot.task_id
+                        );
                         // FIXME @phymbert: to be properly tested
                         //  early returning without changing the slot state will block the slot for ever
                         // no one at the moment is checking the return value
@@ -1863,10 +1876,10 @@ struct llama_server_context
                         const int bd = (slot.ga_w / slot.ga_n) * (slot.ga_n - 1);
                         const int dd = (slot.ga_w / slot.ga_n) - ib * bd - slot.ga_w;
 
-                        LOG_TEE("\n");
-                        LOG_TEE("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i, slot.n_past_se, ib * bd, slot.ga_i + ib * bd, slot.n_past_se + ib * bd);
-                        LOG_TEE("div:   [%6d, %6d] / %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd, slot.ga_i + ib * bd + slot.ga_w, slot.ga_n, (slot.ga_i + ib * bd) / slot.ga_n, (slot.ga_i + ib * bd + slot.ga_w) / slot.ga_n);
-                        LOG_TEE("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd + slot.ga_w, slot.n_past_se + ib * bd, dd, slot.ga_i + ib * bd + slot.ga_w + dd, slot.n_past_se + ib * bd + dd);
+                        LOG("\n");
+                        LOG("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i, slot.n_past_se, ib * bd, slot.ga_i + ib * bd, slot.n_past_se + ib * bd);
+                        LOG("div:   [%6d, %6d] / %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd, slot.ga_i + ib * bd + slot.ga_w, slot.ga_n, (slot.ga_i + ib * bd) / slot.ga_n, (slot.ga_i + ib * bd + slot.ga_w) / slot.ga_n);
+                        LOG("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd + slot.ga_w, slot.n_past_se + ib * bd, dd, slot.ga_i + ib * bd + slot.ga_w + dd, slot.n_past_se + ib * bd + dd);
 
                         llama_kv_cache_seq_add(ctx, slot.id, slot.ga_i, slot.n_past_se, ib * bd);
                         llama_kv_cache_seq_div(ctx, slot.id, slot.ga_i + ib * bd, slot.ga_i + ib * bd + slot.ga_w,slot.ga_n);
@@ -1876,7 +1889,7 @@ struct llama_server_context
 
                         slot.ga_i += slot.ga_w / slot.ga_n;
 
-                        LOG_TEE("\nn_past_old = %d, n_past = %d, ga_i = %d\n\n", slot.n_past_se + bd, slot.n_past_se, slot.ga_i);
+                        LOG("\nn_past_old = %d, n_past = %d, ga_i = %d\n\n", slot.n_past_se + bd, slot.n_past_se, slot.ga_i);
                     }
                     slot.n_past_se += n_tokens;
                 }
@@ -1901,11 +1914,11 @@ struct llama_server_context
                 if (n_batch == 1 || ret < 0)
                 {
                     // if you get here, it means the KV cache is full - try increasing it via the context size
-                    LOG_TEE("%s : failed to decode the batch, n_batch = %d, ret = %d\n", __func__, n_batch, ret);
+                    LOG("%s : failed to decode the batch, n_batch = %d, ret = %d\n", __func__, n_batch, ret);
                     return false;
                 }
 
-                LOG_TEE("%s : failed to find free space in the KV cache, retrying with smaller n_batch = %d\n", __func__, n_batch / 2);
+                LOG("%s : failed to find free space in the KV cache, retrying with smaller n_batch = %d\n", __func__, n_batch / 2);
 
                 // retry with half the batch size to try to find a free slot in the KV cache
                 n_batch /= 2;
@@ -1930,9 +1943,9 @@ struct llama_server_context
                 }
 
                 completion_token_output result;
-                const llama_token id = gpt_sampler_sample(slot.ctx_sampling, ctx, slot.i_batch - i);
+                const llama_token id = common_sampler_sample(slot.ctx_sampling, ctx, slot.i_batch - i);
 
-                gpt_sampler_accept(slot.ctx_sampling, id, true);
+                common_sampler_accept(slot.ctx_sampling, id, true);
 
                 slot.n_decoded += 1;
                 if (slot.n_decoded == 1)
@@ -1943,7 +1956,7 @@ struct llama_server_context
                 }
 
                 result.tok = id;
-                const auto * cur_p = gpt_sampler_get_candidates(slot.ctx_sampling);
+                const auto * cur_p = common_sampler_get_candidates(slot.ctx_sampling);
 
                 for (size_t i = 0; i < (size_t) slot.sparams.n_probs; ++i) {
                     result.probs.push_back({
@@ -1996,7 +2009,7 @@ static json format_partial_response(
 struct token_translator
 {
     llama_context * ctx;
-    std::string operator()(llama_token tok)                    const { return llama_token_to_piece(ctx, tok); }
+    std::string operator()(llama_token tok)                    const { return common_token_to_piece(ctx, tok); }
     std::string operator()(const completion_token_output &cto) const { return (*this)(cto.tok); }
 };
 
@@ -2103,6 +2116,9 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
     data["ignore_eos"] = predict->ignoreeos();
     data["embeddings"] = predict->embeddings();
 
+    // Add the correlationid to json data
+    data["correlation_id"] = predict->correlationid();
+
     // for each image in the request, add the image data
     //
     for (int i = 0; i < predict->images_size(); i++) {
@@ -2187,7 +2203,7 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
 // }
 
 static void params_parse(const backend::ModelOptions* request,
-                                gpt_params & params) {
+                                common_params & params) {
    
     // this is comparable to: https://github.com/ggerganov/llama.cpp/blob/d9b33fe95bd257b36c84ee5769cc048230067d6f/examples/server/server.cpp#L1809
 
@@ -2295,7 +2311,7 @@ public:
 
   grpc::Status LoadModel(ServerContext* context, const backend::ModelOptions* request, backend::Result* result) {
     // Implement LoadModel RPC
-    gpt_params params;
+    common_params params;
     params_parse(request, params);
 
     llama_backend_init();
@@ -2341,6 +2357,11 @@ public:
                 int32_t tokens_evaluated = result.result_json.value("tokens_evaluated", 0);
                 reply.set_prompt_tokens(tokens_evaluated);
 
+                // Log Request Correlation Id
+                LOG_VERBOSE("correlation:", {
+                    { "id", data["correlation_id"] }
+                });
+
                 // Send the reply
                 writer->Write(reply);
 
@@ -2364,6 +2385,12 @@ public:
         std::string completion_text;
         task_result result = llama.queue_results.recv(task_id);
         if (!result.error && result.stop) {
+            
+            // Log Request Correlation Id
+            LOG_VERBOSE("correlation:", {
+                { "id", data["correlation_id"] }
+            });
+
             completion_text = result.result_json.value("content", "");
             int32_t tokens_predicted = result.result_json.value("tokens_predicted", 0);
             int32_t tokens_evaluated = result.result_json.value("tokens_evaluated", 0);
@@ -2403,6 +2430,31 @@ public:
 
         return grpc::Status::OK;
     }
+
+    grpc::Status GetMetrics(ServerContext* context, const backend::MetricsRequest* request, backend::MetricsResponse* response) {
+        llama_client_slot* active_slot = llama.get_active_slot();
+
+        if (active_slot != nullptr) {
+            // Calculate the tokens per second using existing logic
+            double tokens_per_second = 1e3 / active_slot->t_token_generation * active_slot->n_decoded;
+
+            // Populate the response with metrics
+            response->set_slot_id(active_slot->id);
+            response->set_prompt_json_for_slot(active_slot->prompt.dump());
+            response->set_tokens_per_second(tokens_per_second);
+            response->set_tokens_generated(active_slot->n_decoded);
+            response->set_prompt_tokens_processed(active_slot->num_prompt_tokens_processed);
+        } else {
+            // Handle case when no active slot exists
+            response->set_slot_id(0);
+            response->set_prompt_json_for_slot("");
+            response->set_tokens_per_second(0);
+            response->set_tokens_generated(0);
+            response->set_prompt_tokens_processed(0);
+        }
+
+        return grpc::Status::OK;
+    } 
 };
 
 void RunServer(const std::string& server_address) {

backend/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==1.66.1
+grpcio==1.66.2
 protobuf
 certifi
 transformers

backend/python/bark/requirements-intel.txt

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

backend/python/bark/requirements.txt

@@ -1,4 +1,4 @@
 bark==0.1.5
-grpcio==1.66.1
+grpcio==1.66.2
 protobuf
 certifi

backend/python/common/template/requirements.txt

@@ -1,2 +1,2 @@
-grpcio==1.66.1
+grpcio==1.66.2
 protobuf

backend/python/coqui/requirements-intel.txt

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

backend/python/coqui/requirements.txt

@@ -1,4 +1,4 @@
-TTS==0.22.0
-grpcio==1.66.1
+coqui-tts
+grpcio==1.66.2
 protobuf
 certifi

backend/python/coqui/test.py

@@ -19,7 +19,7 @@ class TestBackendServicer(unittest.TestCase):
         This method sets up the gRPC service by starting the server
         """
         self.service = subprocess.Popen(["python3", "backend.py", "--addr", "localhost:50051"])
-        time.sleep(10)
+        time.sleep(30)
 
     def tearDown(self) -> None:
         """

backend/python/diffusers/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==1.66.1
+grpcio==1.66.2
 pillow
 protobuf
 certifi

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==1.66.1
-protobuf
-certifi
-setuptools

backend/python/exllama/run.sh

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

backend/python/exllama/test.sh

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

backend/python/exllama2/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.66.1
+grpcio==1.66.2
 protobuf
 certifi
 wheel

backend/python/mamba/requirements.txt

@@ -1,3 +1,3 @@
-grpcio==1.66.1
+grpcio==1.66.2
 protobuf
 certifi

backend/python/openvoice/requirements-intel.txt

@@ -2,7 +2,7 @@
 intel-extension-for-pytorch
 torch
 optimum[openvino]
-grpcio==1.66.1
+grpcio==1.66.2
 protobuf
 librosa==0.9.1
 faster-whisper==1.0.3
@@ -18,6 +18,6 @@ python-dotenv
 pypinyin==0.50.0
 cn2an==0.5.22
 jieba==0.42.1
-gradio==4.38.1
+gradio==4.44.1
 langid==1.1.6
 git+https://github.com/myshell-ai/MeloTTS.git

backend/python/openvoice/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.66.1
+grpcio==1.66.2
 protobuf
 librosa
 faster-whisper

backend/python/openvoice/test.py

@@ -19,7 +19,7 @@ class TestBackendServicer(unittest.TestCase):
         This method sets up the gRPC service by starting the server
         """
         self.service = subprocess.Popen(["python3", "backend.py", "--addr", "localhost:50051"])
-        time.sleep(10)
+        time.sleep(30)
 
     def tearDown(self) -> None:
         """

backend/python/parler-tts/install.sh

@@ -15,5 +15,12 @@ installRequirements
 
 # https://github.com/descriptinc/audiotools/issues/101
 # incompatible protobuf versions.
-PYDIR=$(ls ${MY_DIR}/venv/lib)
-curl -L https://raw.githubusercontent.com/protocolbuffers/protobuf/main/python/google/protobuf/internal/builder.py -o ${MY_DIR}/venv/lib/${PYDIR}/site-packages/google/protobuf/internal/builder.py
+PYDIR=python3.10
+pyenv="${MY_DIR}/venv/lib/${PYDIR}/site-packages/google/protobuf/internal/"
+
+if [ ! -d ${pyenv} ]; then
+    echo "(parler-tts/install.sh): Error: ${pyenv} does not exist"
+    exit 1
+fi
+
+curl -L https://raw.githubusercontent.com/protocolbuffers/protobuf/main/python/google/protobuf/internal/builder.py -o ${pyenv}/builder.py

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

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

backend/python/parler-tts/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.66.1
+grpcio==1.66.2
 protobuf
 certifi
 llvmlite==0.43.0

backend/python/rerankers/requirements-intel.txt

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

backend/python/rerankers/requirements.txt

@@ -1,3 +1,3 @@
-grpcio==1.66.1
+grpcio==1.66.2
 protobuf
 certifi

backend/python/sentencetransformers/backend.py

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

backend/python/sentencetransformers/requirements-cpu.txt

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

backend/python/sentencetransformers/requirements-cublas11.txt

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

backend/python/sentencetransformers/requirements-cublas12.txt

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

backend/python/sentencetransformers/requirements-hipblas.txt

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

backend/python/sentencetransformers/requirements-intel.txt

@@ -4,5 +4,5 @@ torch
 optimum[openvino]
 setuptools==69.5.1 # https://github.com/mudler/LocalAI/issues/2406
 accelerate
-sentence-transformers==3.0.1
+sentence-transformers==3.1.1
 transformers

backend/python/sentencetransformers/requirements.txt

@@ -1,3 +1,5 @@
-grpcio==1.66.1
+grpcio==1.66.2
 protobuf
-certifi
+certifi
+datasets
+einops

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

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

backend/python/transformers-musicgen/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.66.1
+grpcio==1.66.2
 protobuf
 scipy==1.14.0
 certifi

backend/python/transformers/backend.py

@@ -72,7 +72,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         Returns:
             A Result object that contains the result of the LoadModel operation.
         """
+
         model_name = request.Model
+        
+        # Check to see if the Model exists in the filesystem already.
+        if os.path.exists(request.ModelFile):
+            model_name = request.ModelFile
 
         compute = torch.float16
         if request.F16Memory == True:

backend/python/transformers/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.66.1
+grpcio==1.66.2
 protobuf
 certifi
 setuptools==69.5.1 # https://github.com/mudler/LocalAI/issues/2406

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

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

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

@@ -1,3 +1,3 @@
-grpcio==1.66.1
+grpcio==1.66.2
 protobuf
 certifi

backend/python/vllm/backend.py

@@ -5,6 +5,8 @@ import argparse
 import signal
 import sys
 import os
+from typing import List
+from PIL import Image
 
 import backend_pb2
 import backend_pb2_grpc
@@ -15,6 +17,8 @@ from vllm.engine.async_llm_engine import AsyncLLMEngine
 from vllm.sampling_params import SamplingParams
 from vllm.utils import random_uuid
 from vllm.transformers_utils.tokenizer import get_tokenizer
+from vllm.multimodal.utils import fetch_image
+from vllm.assets.video import VideoAsset
 
 _ONE_DAY_IN_SECONDS = 60 * 60 * 24
 
@@ -105,6 +109,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         try:
             self.llm = AsyncLLMEngine.from_engine_args(engine_args)
         except Exception as err:
+            print(f"Unexpected {err=}, {type(err)=}", file=sys.stderr)
             return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
 
         try:
@@ -117,7 +122,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
            )
         except Exception as err:
             return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
-
+        print("Model loaded successfully", file=sys.stderr)
         return backend_pb2.Result(message="Model loaded successfully", success=True)
 
     async def Predict(self, request, context):
@@ -196,15 +201,33 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         if request.Seed != 0:
             sampling_params.seed = request.Seed
 
+        # Extract image paths and process images
         prompt = request.Prompt
-        
-        # If tokenizer template is enabled and messages are provided instead of prompt apply the tokenizer template
+
+        image_paths = request.Images
+        image_data = [self.load_image(img_path) for img_path in image_paths]
+
+        videos_path = request.Videos
+        video_data = [self.load_video(video_path) for video_path in videos_path]
+
+        # If tokenizer template is enabled and messages are provided instead of prompt, apply the tokenizer template
         if not request.Prompt and request.UseTokenizerTemplate and request.Messages:
             prompt = self.tokenizer.apply_chat_template(request.Messages, tokenize=False, add_generation_prompt=True)
 
-        # Generate text
+        # Generate text using the LLM engine
         request_id = random_uuid()
-        outputs = self.llm.generate(prompt, sampling_params, request_id)
+        print(f"Generating text with request_id: {request_id}", file=sys.stderr)
+        outputs = self.llm.generate(
+            {
+                "prompt": prompt,
+                "multi_modal_data": {
+                    "image": image_data if image_data else None,
+                    "video": video_data if video_data else None,
+                } if image_data or video_data else None,
+            },
+            sampling_params=sampling_params,
+            request_id=request_id,
+        )
 
         # Stream the results
         generated_text = ""
@@ -227,9 +250,49 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         if streaming:
             return
 
+        # Remove the image files from /tmp folder
+        for img_path in image_paths:
+            try:
+                os.remove(img_path)
+            except Exception as e:
+                print(f"Error removing image file: {img_path}, {e}", file=sys.stderr)
+
         # Sending the final generated text
         yield backend_pb2.Reply(message=bytes(generated_text, encoding='utf-8'))
 
+    def load_image(self, image_path: str):
+        """
+        Load an image from the given file path.
+        
+        Args:
+            image_path (str): The path to the image file.
+
+        Returns:
+            Image: The loaded image.
+        """
+        try:
+            return Image.open(image_path)
+        except Exception as e:
+            print(f"Error loading image {image_path}: {e}", file=sys.stderr)
+            return self.load_video(image_path)
+
+    def load_video(self, video_path: str):
+        """
+        Load a video from the given file path.
+        
+        Args:
+            video_path (str): The path to the image file.
+
+        Returns:
+            Video: The loaded video.
+        """
+        try:
+            video = VideoAsset(name=video_path).np_ndarrays
+            return video
+        except Exception as e:
+            print(f"Error loading video {image_path}: {e}", file=sys.stderr)
+            return None
+
 async def serve(address):
     # Start asyncio gRPC server
     server = grpc.aio.server(migration_thread_pool=futures.ThreadPoolExecutor(max_workers=MAX_WORKERS))

backend/python/vllm/install.sh

@@ -13,4 +13,20 @@ if [ "x${BUILD_PROFILE}" == "xintel" ]; then
     EXTRA_PIP_INSTALL_FLAGS+=" --upgrade --index-strategy=unsafe-first-match"
 fi
 
-installRequirements
+# We don't embed this into the images as it is a large dependency and not always needed.
+# Besides, the speed inference are not actually usable in the current state for production use-cases.
+if [ "x${BUILD_TYPE}" == "x" ] && [ "x${FROM_SOURCE}" == "xtrue" ]; then
+        ensureVenv
+        # https://docs.vllm.ai/en/v0.6.1/getting_started/cpu-installation.html
+        if [ ! -d vllm ]; then
+            git clone https://github.com/vllm-project/vllm
+        fi
+        pushd vllm
+            uv pip install wheel packaging ninja "setuptools>=49.4.0" numpy typing-extensions pillow setuptools-scm grpcio==1.66.2 protobuf bitsandbytes
+            uv pip install -v -r requirements-cpu.txt --extra-index-url https://download.pytorch.org/whl/cpu
+            VLLM_TARGET_DEVICE=cpu python setup.py install
+        popd
+        rm -rf vllm
+    else
+        installRequirements
+fi

backend/python/vllm/requirements-cublas11.txt

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

backend/python/vllm/requirements-cublas12.txt

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

backend/python/vllm/requirements-hipblas.txt

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

backend/python/vllm/requirements-intel.txt

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

backend/python/vllm/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.66.1
+grpcio==1.66.2
 protobuf
 certifi
 setuptools

core/backend/embeddings.go

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

core/backend/image.go

@@ -8,19 +8,8 @@ import (
 )
 
 func ImageGeneration(height, width, mode, step, seed int, positive_prompt, negative_prompt, src, dst string, loader *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (func() error, error) {
-	threads := backendConfig.Threads
-	if *threads == 0 && appConfig.Threads != 0 {
-		threads = &appConfig.Threads
-	}
-	gRPCOpts := gRPCModelOpts(backendConfig)
-	opts := modelOpts(backendConfig, appConfig, []model.Option{
-		model.WithBackendString(backendConfig.Backend),
-		model.WithAssetDir(appConfig.AssetsDestination),
-		model.WithThreads(uint32(*threads)),
-		model.WithContext(appConfig.Context),
-		model.WithModel(backendConfig.Model),
-		model.WithLoadGRPCLoadModelOpts(gRPCOpts),
-	})
+
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{})
 
 	inferenceModel, err := loader.BackendLoader(
 		opts...,

core/backend/llm.go

@@ -31,24 +31,13 @@ type TokenUsage struct {
 	Completion int
 }
 
-func ModelInference(ctx context.Context, s string, messages []schema.Message, images []string, loader *model.ModelLoader, c config.BackendConfig, o *config.ApplicationConfig, tokenCallback func(string, TokenUsage) bool) (func() (LLMResponse, error), error) {
+func ModelInference(ctx context.Context, s string, messages []schema.Message, images, videos, audios []string, loader *model.ModelLoader, c config.BackendConfig, o *config.ApplicationConfig, tokenCallback func(string, TokenUsage) bool) (func() (LLMResponse, error), error) {
 	modelFile := c.Model
-	threads := c.Threads
-	if *threads == 0 && o.Threads != 0 {
-		threads = &o.Threads
-	}
-	grpcOpts := gRPCModelOpts(c)
 
 	var inferenceModel grpc.Backend
 	var err error
 
-	opts := modelOpts(c, o, []model.Option{
-		model.WithLoadGRPCLoadModelOpts(grpcOpts),
-		model.WithThreads(uint32(*threads)), // some models uses this to allocate threads during startup
-		model.WithAssetDir(o.AssetsDestination),
-		model.WithModel(modelFile),
-		model.WithContext(o.Context),
-	})
+	opts := ModelOptions(c, o, []model.Option{})
 
 	if c.Backend != "" {
 		opts = append(opts, model.WithBackendString(c.Backend))
@@ -101,6 +90,8 @@ func ModelInference(ctx context.Context, s string, messages []schema.Message, im
 		opts.Messages = protoMessages
 		opts.UseTokenizerTemplate = c.TemplateConfig.UseTokenizerTemplate
 		opts.Images = images
+		opts.Videos = videos
+		opts.Audios = audios
 
 		tokenUsage := TokenUsage{}
 

core/backend/options.go

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

core/backend/rerank.go

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

core/backend/soundgeneration.go

@@ -13,7 +13,6 @@ import (
 )
 
 func SoundGeneration(
-	backend string,
 	modelFile string,
 	text string,
 	duration *float32,
@@ -25,18 +24,8 @@ func SoundGeneration(
 	appConfig *config.ApplicationConfig,
 	backendConfig config.BackendConfig,
 ) (string, *proto.Result, error) {
-	if backend == "" {
-		return "", nil, fmt.Errorf("backend is a required parameter")
-	}
 
-	grpcOpts := gRPCModelOpts(backendConfig)
-	opts := modelOpts(config.BackendConfig{}, appConfig, []model.Option{
-		model.WithBackendString(backend),
-		model.WithModel(modelFile),
-		model.WithContext(appConfig.Context),
-		model.WithAssetDir(appConfig.AssetsDestination),
-		model.WithLoadGRPCLoadModelOpts(grpcOpts),
-	})
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{model.WithModel(modelFile)})
 
 	soundGenModel, err := loader.BackendLoader(opts...)
 	if err != nil {

core/backend/token_metrics.go

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

core/backend/tokenize.go

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

core/backend/transcript.go

@@ -14,13 +14,11 @@ import (
 
 func ModelTranscription(audio, language string, translate bool, ml *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (*schema.TranscriptionResult, error) {
 
-	opts := modelOpts(backendConfig, appConfig, []model.Option{
-		model.WithBackendString(model.WhisperBackend),
-		model.WithModel(backendConfig.Model),
-		model.WithContext(appConfig.Context),
-		model.WithThreads(uint32(*backendConfig.Threads)),
-		model.WithAssetDir(appConfig.AssetsDestination),
-	})
+	if backendConfig.Backend == "" {
+		backendConfig.Backend = model.WhisperBackend
+	}
+
+	opts := ModelOptions(backendConfig, appConfig, []model.Option{})
 
 	transcriptionModel, err := ml.BackendLoader(opts...)
 	if err != nil {

core/backend/tts.go

@@ -28,14 +28,9 @@ func ModelTTS(
 		bb = model.PiperBackend
 	}
 
-	grpcOpts := gRPCModelOpts(backendConfig)
-
-	opts := modelOpts(config.BackendConfig{}, appConfig, []model.Option{
+	opts := ModelOptions(config.BackendConfig{}, appConfig, []model.Option{
 		model.WithBackendString(bb),
 		model.WithModel(modelFile),
-		model.WithContext(appConfig.Context),
-		model.WithAssetDir(appConfig.AssetsDestination),
-		model.WithLoadGRPCLoadModelOpts(grpcOpts),
 	})
 	ttsModel, err := loader.BackendLoader(opts...)
 	if err != nil {

core/cli/run.go

@@ -41,31 +41,35 @@ type RunCMD struct {
 	Threads     int  `env:"LOCALAI_THREADS,THREADS" short:"t" help:"Number of threads used for parallel computation. Usage of the number of physical cores in the system is suggested" group:"performance"`
 	ContextSize int  `env:"LOCALAI_CONTEXT_SIZE,CONTEXT_SIZE" default:"512" help:"Default context size for models" group:"performance"`
 
-	Address                string   `env:"LOCALAI_ADDRESS,ADDRESS" default:":8080" help:"Bind address for the API server" group:"api"`
-	CORS                   bool     `env:"LOCALAI_CORS,CORS" help:"" group:"api"`
-	CORSAllowOrigins       string   `env:"LOCALAI_CORS_ALLOW_ORIGINS,CORS_ALLOW_ORIGINS" group:"api"`
-	LibraryPath            string   `env:"LOCALAI_LIBRARY_PATH,LIBRARY_PATH" help:"Path to the library directory (for e.g. external libraries used by backends)" default:"/usr/share/local-ai/libs" group:"backends"`
-	CSRF                   bool     `env:"LOCALAI_CSRF" help:"Enables fiber CSRF middleware" group:"api"`
-	UploadLimit            int      `env:"LOCALAI_UPLOAD_LIMIT,UPLOAD_LIMIT" default:"15" help:"Default upload-limit in MB" group:"api"`
-	APIKeys                []string `env:"LOCALAI_API_KEY,API_KEY" help:"List of API Keys to enable API authentication. When this is set, all the requests must be authenticated with one of these API keys" group:"api"`
-	DisableWebUI           bool     `env:"LOCALAI_DISABLE_WEBUI,DISABLE_WEBUI" default:"false" help:"Disable webui" group:"api"`
-	DisablePredownloadScan bool     `env:"LOCALAI_DISABLE_PREDOWNLOAD_SCAN" help:"If true, disables the best-effort security scanner before downloading any files." group:"hardening" default:"false"`
-	OpaqueErrors           bool     `env:"LOCALAI_OPAQUE_ERRORS" default:"false" help:"If true, all error responses are replaced with blank 500 errors. This is intended only for hardening against information leaks and is normally not recommended." group:"hardening"`
-	Peer2Peer              bool     `env:"LOCALAI_P2P,P2P" name:"p2p" default:"false" help:"Enable P2P mode" group:"p2p"`
-	Peer2PeerDHTInterval   int      `env:"LOCALAI_P2P_DHT_INTERVAL,P2P_DHT_INTERVAL" default:"360" name:"p2p-dht-interval" help:"Interval for DHT refresh (used during token generation)" group:"p2p"`
-	Peer2PeerOTPInterval   int      `env:"LOCALAI_P2P_OTP_INTERVAL,P2P_OTP_INTERVAL" default:"9000" name:"p2p-otp-interval" help:"Interval for OTP refresh (used during token generation)" group:"p2p"`
-	Peer2PeerToken         string   `env:"LOCALAI_P2P_TOKEN,P2P_TOKEN,TOKEN" name:"p2ptoken" help:"Token for P2P mode (optional)" group:"p2p"`
-	Peer2PeerNetworkID     string   `env:"LOCALAI_P2P_NETWORK_ID,P2P_NETWORK_ID" help:"Network ID for P2P mode, can be set arbitrarly by the user for grouping a set of instances" group:"p2p"`
-	ParallelRequests       bool     `env:"LOCALAI_PARALLEL_REQUESTS,PARALLEL_REQUESTS" help:"Enable backends to handle multiple requests in parallel if they support it (e.g.: llama.cpp or vllm)" group:"backends"`
-	SingleActiveBackend    bool     `env:"LOCALAI_SINGLE_ACTIVE_BACKEND,SINGLE_ACTIVE_BACKEND" help:"Allow only one backend to be run at a time" group:"backends"`
-	PreloadBackendOnly     bool     `env:"LOCALAI_PRELOAD_BACKEND_ONLY,PRELOAD_BACKEND_ONLY" default:"false" help:"Do not launch the API services, only the preloaded models / backends are started (useful for multi-node setups)" group:"backends"`
-	ExternalGRPCBackends   []string `env:"LOCALAI_EXTERNAL_GRPC_BACKENDS,EXTERNAL_GRPC_BACKENDS" help:"A list of external grpc backends" group:"backends"`
-	EnableWatchdogIdle     bool     `env:"LOCALAI_WATCHDOG_IDLE,WATCHDOG_IDLE" default:"false" help:"Enable watchdog for stopping backends that are idle longer than the watchdog-idle-timeout" group:"backends"`
-	WatchdogIdleTimeout    string   `env:"LOCALAI_WATCHDOG_IDLE_TIMEOUT,WATCHDOG_IDLE_TIMEOUT" default:"15m" help:"Threshold beyond which an idle backend should be stopped" group:"backends"`
-	EnableWatchdogBusy     bool     `env:"LOCALAI_WATCHDOG_BUSY,WATCHDOG_BUSY" default:"false" help:"Enable watchdog for stopping backends that are busy longer than the watchdog-busy-timeout" group:"backends"`
-	WatchdogBusyTimeout    string   `env:"LOCALAI_WATCHDOG_BUSY_TIMEOUT,WATCHDOG_BUSY_TIMEOUT" default:"5m" help:"Threshold beyond which a busy backend should be stopped" group:"backends"`
-	Federated              bool     `env:"LOCALAI_FEDERATED,FEDERATED" help:"Enable federated instance" group:"federated"`
-	DisableGalleryEndpoint bool     `env:"LOCALAI_DISABLE_GALLERY_ENDPOINT,DISABLE_GALLERY_ENDPOINT" help:"Disable the gallery endpoints" group:"api"`
+	Address                            string   `env:"LOCALAI_ADDRESS,ADDRESS" default:":8080" help:"Bind address for the API server" group:"api"`
+	CORS                               bool     `env:"LOCALAI_CORS,CORS" help:"" group:"api"`
+	CORSAllowOrigins                   string   `env:"LOCALAI_CORS_ALLOW_ORIGINS,CORS_ALLOW_ORIGINS" group:"api"`
+	LibraryPath                        string   `env:"LOCALAI_LIBRARY_PATH,LIBRARY_PATH" help:"Path to the library directory (for e.g. external libraries used by backends)" default:"/usr/share/local-ai/libs" group:"backends"`
+	CSRF                               bool     `env:"LOCALAI_CSRF" help:"Enables fiber CSRF middleware" group:"api"`
+	UploadLimit                        int      `env:"LOCALAI_UPLOAD_LIMIT,UPLOAD_LIMIT" default:"15" help:"Default upload-limit in MB" group:"api"`
+	APIKeys                            []string `env:"LOCALAI_API_KEY,API_KEY" help:"List of API Keys to enable API authentication. When this is set, all the requests must be authenticated with one of these API keys" group:"api"`
+	DisableWebUI                       bool     `env:"LOCALAI_DISABLE_WEBUI,DISABLE_WEBUI" default:"false" help:"Disable webui" group:"api"`
+	DisablePredownloadScan             bool     `env:"LOCALAI_DISABLE_PREDOWNLOAD_SCAN" help:"If true, disables the best-effort security scanner before downloading any files." group:"hardening" default:"false"`
+	OpaqueErrors                       bool     `env:"LOCALAI_OPAQUE_ERRORS" default:"false" help:"If true, all error responses are replaced with blank 500 errors. This is intended only for hardening against information leaks and is normally not recommended." group:"hardening"`
+	UseSubtleKeyComparison             bool     `env:"LOCALAI_SUBTLE_KEY_COMPARISON" default:"false" help:"If true, API Key validation comparisons will be performed using constant-time comparisons rather than simple equality. This trades off performance on each request for resiliancy against timing attacks." group:"hardening"`
+	DisableApiKeyRequirementForHttpGet bool     `env:"LOCALAI_DISABLE_API_KEY_REQUIREMENT_FOR_HTTP_GET" default:"false" help:"If true, a valid API key is not required to issue GET requests to portions of the web ui. This should only be enabled in secure testing environments" group:"hardening"`
+	HttpGetExemptedEndpoints           []string `env:"LOCALAI_HTTP_GET_EXEMPTED_ENDPOINTS" default:"^/$,^/browse/?$,^/talk/?$,^/p2p/?$,^/chat/?$,^/text2image/?$,^/tts/?$,^/static/.*$,^/swagger.*$" help:"If LOCALAI_DISABLE_API_KEY_REQUIREMENT_FOR_HTTP_GET is overriden to true, this is the list of endpoints to exempt. Only adjust this in case of a security incident or as a result of a personal security posture review" group:"hardening"`
+	Peer2Peer                          bool     `env:"LOCALAI_P2P,P2P" name:"p2p" default:"false" help:"Enable P2P mode" group:"p2p"`
+	Peer2PeerDHTInterval               int      `env:"LOCALAI_P2P_DHT_INTERVAL,P2P_DHT_INTERVAL" default:"360" name:"p2p-dht-interval" help:"Interval for DHT refresh (used during token generation)" group:"p2p"`
+	Peer2PeerOTPInterval               int      `env:"LOCALAI_P2P_OTP_INTERVAL,P2P_OTP_INTERVAL" default:"9000" name:"p2p-otp-interval" help:"Interval for OTP refresh (used during token generation)" group:"p2p"`
+	Peer2PeerToken                     string   `env:"LOCALAI_P2P_TOKEN,P2P_TOKEN,TOKEN" name:"p2ptoken" help:"Token for P2P mode (optional)" group:"p2p"`
+	Peer2PeerNetworkID                 string   `env:"LOCALAI_P2P_NETWORK_ID,P2P_NETWORK_ID" help:"Network ID for P2P mode, can be set arbitrarly by the user for grouping a set of instances" group:"p2p"`
+	ParallelRequests                   bool     `env:"LOCALAI_PARALLEL_REQUESTS,PARALLEL_REQUESTS" help:"Enable backends to handle multiple requests in parallel if they support it (e.g.: llama.cpp or vllm)" group:"backends"`
+	SingleActiveBackend                bool     `env:"LOCALAI_SINGLE_ACTIVE_BACKEND,SINGLE_ACTIVE_BACKEND" help:"Allow only one backend to be run at a time" group:"backends"`
+	PreloadBackendOnly                 bool     `env:"LOCALAI_PRELOAD_BACKEND_ONLY,PRELOAD_BACKEND_ONLY" default:"false" help:"Do not launch the API services, only the preloaded models / backends are started (useful for multi-node setups)" group:"backends"`
+	ExternalGRPCBackends               []string `env:"LOCALAI_EXTERNAL_GRPC_BACKENDS,EXTERNAL_GRPC_BACKENDS" help:"A list of external grpc backends" group:"backends"`
+	EnableWatchdogIdle                 bool     `env:"LOCALAI_WATCHDOG_IDLE,WATCHDOG_IDLE" default:"false" help:"Enable watchdog for stopping backends that are idle longer than the watchdog-idle-timeout" group:"backends"`
+	WatchdogIdleTimeout                string   `env:"LOCALAI_WATCHDOG_IDLE_TIMEOUT,WATCHDOG_IDLE_TIMEOUT" default:"15m" help:"Threshold beyond which an idle backend should be stopped" group:"backends"`
+	EnableWatchdogBusy                 bool     `env:"LOCALAI_WATCHDOG_BUSY,WATCHDOG_BUSY" default:"false" help:"Enable watchdog for stopping backends that are busy longer than the watchdog-busy-timeout" group:"backends"`
+	WatchdogBusyTimeout                string   `env:"LOCALAI_WATCHDOG_BUSY_TIMEOUT,WATCHDOG_BUSY_TIMEOUT" default:"5m" help:"Threshold beyond which a busy backend should be stopped" group:"backends"`
+	Federated                          bool     `env:"LOCALAI_FEDERATED,FEDERATED" help:"Enable federated instance" group:"federated"`
+	DisableGalleryEndpoint             bool     `env:"LOCALAI_DISABLE_GALLERY_ENDPOINT,DISABLE_GALLERY_ENDPOINT" help:"Disable the gallery endpoints" group:"api"`
+	LoadToMemory                       []string `env:"LOCALAI_LOAD_TO_MEMORY,LOAD_TO_MEMORY" help:"A list of models to load into memory at startup" group:"models"`
 }
 
 func (r *RunCMD) Run(ctx *cliContext.Context) error {
@@ -97,7 +101,11 @@ func (r *RunCMD) Run(ctx *cliContext.Context) error {
 		config.WithModelsURL(append(r.Models, r.ModelArgs...)...),
 		config.WithOpaqueErrors(r.OpaqueErrors),
 		config.WithEnforcedPredownloadScans(!r.DisablePredownloadScan),
+		config.WithSubtleKeyComparison(r.UseSubtleKeyComparison),
+		config.WithDisableApiKeyRequirementForHttpGet(r.DisableApiKeyRequirementForHttpGet),
+		config.WithHttpGetExemptedEndpoints(r.HttpGetExemptedEndpoints),
 		config.WithP2PNetworkID(r.Peer2PeerNetworkID),
+		config.WithLoadToMemory(r.LoadToMemory),
 	}
 
 	token := ""

core/cli/soundgeneration.go

@@ -85,13 +85,14 @@ func (t *SoundGenerationCMD) Run(ctx *cliContext.Context) error {
 
 	options := config.BackendConfig{}
 	options.SetDefaults()
+	options.Backend = t.Backend
 
 	var inputFile *string
 	if t.InputFile != "" {
 		inputFile = &t.InputFile
 	}
 
-	filePath, _, err := backend.SoundGeneration(t.Backend, t.Model, text,
+	filePath, _, err := backend.SoundGeneration(t.Model, text,
 		parseToFloat32Ptr(t.Duration), parseToFloat32Ptr(t.Temperature), &t.DoSample,
 		inputFile, parseToInt32Ptr(t.InputFileSampleDivisor), ml, opts, options)
 

core/cli/util.go

@@ -15,8 +15,9 @@ import (
 )
 
 type UtilCMD struct {
-	GGUFInfo GGUFInfoCMD `cmd:"" name:"gguf-info" help:"Get information about a GGUF file"`
-	HFScan   HFScanCMD   `cmd:"" name:"hf-scan" help:"Checks installed models for known security issues. WARNING: this is a best-effort feature and may not catch everything!"`
+	GGUFInfo         GGUFInfoCMD         `cmd:"" name:"gguf-info" help:"Get information about a GGUF file"`
+	HFScan           HFScanCMD           `cmd:"" name:"hf-scan" help:"Checks installed models for known security issues. WARNING: this is a best-effort feature and may not catch everything!"`
+	UsecaseHeuristic UsecaseHeuristicCMD `cmd:"" name:"usecase-heuristic" help:"Checks a specific model config and prints what usecase LocalAI will offer for it."`
 }
 
 type GGUFInfoCMD struct {
@@ -30,6 +31,11 @@ type HFScanCMD struct {
 	ToScan     []string `arg:""`
 }
 
+type UsecaseHeuristicCMD struct {
+	ConfigName string `name:"The config file to check"`
+	ModelsPath string `env:"LOCALAI_MODELS_PATH,MODELS_PATH" type:"path" default:"${basepath}/models" help:"Path containing models used for inferencing" group:"storage"`
+}
+
 func (u *GGUFInfoCMD) Run(ctx *cliContext.Context) error {
 	if u.Args == nil || len(u.Args) == 0 {
 		return fmt.Errorf("no GGUF file provided")
@@ -99,3 +105,31 @@ func (hfscmd *HFScanCMD) Run(ctx *cliContext.Context) error {
 		return nil
 	}
 }
+
+func (uhcmd *UsecaseHeuristicCMD) Run(ctx *cliContext.Context) error {
+	if len(uhcmd.ConfigName) == 0 {
+		log.Error().Msg("ConfigName is a required parameter")
+		return fmt.Errorf("config name is a required parameter")
+	}
+	if len(uhcmd.ModelsPath) == 0 {
+		log.Error().Msg("ModelsPath is a required parameter")
+		return fmt.Errorf("model path is a required parameter")
+	}
+	bcl := config.NewBackendConfigLoader(uhcmd.ModelsPath)
+	err := bcl.LoadBackendConfig(uhcmd.ConfigName)
+	if err != nil {
+		log.Error().Err(err).Str("ConfigName", uhcmd.ConfigName).Msg("error while loading backend")
+		return err
+	}
+	bc, exists := bcl.GetBackendConfig(uhcmd.ConfigName)
+	if !exists {
+		log.Error().Str("ConfigName", uhcmd.ConfigName).Msg("ConfigName not found")
+	}
+	for name, uc := range config.GetAllBackendConfigUsecases() {
+		if bc.HasUsecases(uc) {
+			log.Info().Str("Usecase", name)
+		}
+	}
+	log.Info().Msg("---")
+	return nil
+}

core/config/application_config.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"embed"
 	"encoding/json"
+	"regexp"
 	"time"
 
 	"github.com/mudler/LocalAI/pkg/xsysinfo"
@@ -16,7 +17,6 @@ type ApplicationConfig struct {
 	ModelPath                           string
 	LibPath                             string
 	UploadLimitMB, Threads, ContextSize int
-	DisableWebUI                        bool
 	F16                                 bool
 	Debug                               bool
 	ImageDir                            string
@@ -31,11 +31,18 @@ type ApplicationConfig struct {
 	PreloadModelsFromPath               string
 	CORSAllowOrigins                    string
 	ApiKeys                             []string
-	EnforcePredownloadScans             bool
-	OpaqueErrors                        bool
 	P2PToken                            string
 	P2PNetworkID                        string
 
+	DisableWebUI                       bool
+	EnforcePredownloadScans            bool
+	OpaqueErrors                       bool
+	UseSubtleKeyComparison             bool
+	DisableApiKeyRequirementForHttpGet bool
+	HttpGetExemptedEndpoints           []*regexp.Regexp
+	DisableGalleryEndpoint             bool
+	LoadToMemory                       []string
+
 	ModelLibraryURL string
 
 	Galleries []Gallery
@@ -57,8 +64,6 @@ type ApplicationConfig struct {
 	ModelsURL []string
 
 	WatchDogBusyTimeout, WatchDogIdleTimeout time.Duration
-
-	DisableGalleryEndpoint bool
 }
 
 type AppOption func(*ApplicationConfig)
@@ -327,6 +332,38 @@ func WithOpaqueErrors(opaque bool) AppOption {
 	}
 }
 
+func WithLoadToMemory(models []string) AppOption {
+	return func(o *ApplicationConfig) {
+		o.LoadToMemory = models
+	}
+}
+
+func WithSubtleKeyComparison(subtle bool) AppOption {
+	return func(o *ApplicationConfig) {
+		o.UseSubtleKeyComparison = subtle
+	}
+}
+
+func WithDisableApiKeyRequirementForHttpGet(required bool) AppOption {
+	return func(o *ApplicationConfig) {
+		o.DisableApiKeyRequirementForHttpGet = required
+	}
+}
+
+func WithHttpGetExemptedEndpoints(endpoints []string) AppOption {
+	return func(o *ApplicationConfig) {
+		o.HttpGetExemptedEndpoints = []*regexp.Regexp{}
+		for _, epr := range endpoints {
+			r, err := regexp.Compile(epr)
+			if err == nil && r != nil {
+				o.HttpGetExemptedEndpoints = append(o.HttpGetExemptedEndpoints, r)
+			} else {
+				log.Warn().Err(err).Str("regex", epr).Msg("Error while compiling HTTP Get Exemption regex, skipping this entry.")
+			}
+		}
+	}
+}
+
 // ToConfigLoaderOptions returns a slice of ConfigLoader Option.
 // Some options defined at the application level are going to be passed as defaults for
 // all the configuration for the models.

core/config/backend_config.go

@@ -3,11 +3,13 @@ package config
 import (
 	"os"
 	"regexp"
+	"slices"
 	"strings"
 
 	"github.com/mudler/LocalAI/core/schema"
 	"github.com/mudler/LocalAI/pkg/downloader"
 	"github.com/mudler/LocalAI/pkg/functions"
+	"gopkg.in/yaml.v3"
 )
 
 const (
@@ -27,13 +29,15 @@ type BackendConfig struct {
 	schema.PredictionOptions `yaml:"parameters"`
 	Name                     string `yaml:"name"`
 
-	F16            *bool             `yaml:"f16"`
-	Threads        *int              `yaml:"threads"`
-	Debug          *bool             `yaml:"debug"`
-	Roles          map[string]string `yaml:"roles"`
-	Embeddings     *bool             `yaml:"embeddings"`
-	Backend        string            `yaml:"backend"`
-	TemplateConfig TemplateConfig    `yaml:"template"`
+	F16                 *bool                  `yaml:"f16"`
+	Threads             *int                   `yaml:"threads"`
+	Debug               *bool                  `yaml:"debug"`
+	Roles               map[string]string      `yaml:"roles"`
+	Embeddings          *bool                  `yaml:"embeddings"`
+	Backend             string                 `yaml:"backend"`
+	TemplateConfig      TemplateConfig         `yaml:"template"`
+	KnownUsecaseStrings []string               `yaml:"known_usecases"`
+	KnownUsecases       *BackendConfigUsecases `yaml:"-"`
 
 	PromptStrings, InputStrings                []string               `yaml:"-"`
 	InputToken                                 [][]int                `yaml:"-"`
@@ -192,6 +196,21 @@ type TemplateConfig struct {
 	// JoinChatMessagesByCharacter is a string that will be used to join chat messages together.
 	// It defaults to \n
 	JoinChatMessagesByCharacter *string `yaml:"join_chat_messages_by_character"`
+
+	Video string `yaml:"video"`
+	Image string `yaml:"image"`
+	Audio string `yaml:"audio"`
+}
+
+func (c *BackendConfig) UnmarshalYAML(value *yaml.Node) error {
+	type BCAlias BackendConfig
+	var aux BCAlias
+	if err := value.Decode(&aux); err != nil {
+		return err
+	}
+	*c = BackendConfig(aux)
+	c.KnownUsecases = GetUsecasesFromYAML(c.KnownUsecaseStrings)
+	return nil
 }
 
 func (c *BackendConfig) SetFunctionCallString(s string) {
@@ -410,3 +429,121 @@ func (c *BackendConfig) Validate() bool {
 func (c *BackendConfig) HasTemplate() bool {
 	return c.TemplateConfig.Completion != "" || c.TemplateConfig.Edit != "" || c.TemplateConfig.Chat != "" || c.TemplateConfig.ChatMessage != ""
 }
+
+type BackendConfigUsecases int
+
+const (
+	FLAG_ANY              BackendConfigUsecases = 0b000000000
+	FLAG_CHAT             BackendConfigUsecases = 0b000000001
+	FLAG_COMPLETION       BackendConfigUsecases = 0b000000010
+	FLAG_EDIT             BackendConfigUsecases = 0b000000100
+	FLAG_EMBEDDINGS       BackendConfigUsecases = 0b000001000
+	FLAG_RERANK           BackendConfigUsecases = 0b000010000
+	FLAG_IMAGE            BackendConfigUsecases = 0b000100000
+	FLAG_TRANSCRIPT       BackendConfigUsecases = 0b001000000
+	FLAG_TTS              BackendConfigUsecases = 0b010000000
+	FLAG_SOUND_GENERATION BackendConfigUsecases = 0b100000000
+
+	// Common Subsets
+	FLAG_LLM BackendConfigUsecases = FLAG_CHAT & FLAG_COMPLETION & FLAG_EDIT
+)
+
+func GetAllBackendConfigUsecases() map[string]BackendConfigUsecases {
+	return map[string]BackendConfigUsecases{
+		"FLAG_ANY":              FLAG_ANY,
+		"FLAG_CHAT":             FLAG_CHAT,
+		"FLAG_COMPLETION":       FLAG_COMPLETION,
+		"FLAG_EDIT":             FLAG_EDIT,
+		"FLAG_EMBEDDINGS":       FLAG_EMBEDDINGS,
+		"FLAG_RERANK":           FLAG_RERANK,
+		"FLAG_IMAGE":            FLAG_IMAGE,
+		"FLAG_TRANSCRIPT":       FLAG_TRANSCRIPT,
+		"FLAG_TTS":              FLAG_TTS,
+		"FLAG_SOUND_GENERATION": FLAG_SOUND_GENERATION,
+		"FLAG_LLM":              FLAG_LLM,
+	}
+}
+
+func GetUsecasesFromYAML(input []string) *BackendConfigUsecases {
+	if len(input) == 0 {
+		return nil
+	}
+	result := FLAG_ANY
+	flags := GetAllBackendConfigUsecases()
+	for _, str := range input {
+		flag, exists := flags["FLAG_"+strings.ToUpper(str)]
+		if exists {
+			result |= flag
+		}
+	}
+	return &result
+}
+
+// HasUsecases examines a BackendConfig and determines which endpoints have a chance of success.
+func (c *BackendConfig) HasUsecases(u BackendConfigUsecases) bool {
+	if (c.KnownUsecases != nil) && ((u & *c.KnownUsecases) == u) {
+		return true
+	}
+	return c.GuessUsecases(u)
+}
+
+// GuessUsecases is a **heuristic based** function, as the backend in question may not be loaded yet, and the config may not record what it's useful at.
+// In its current state, this function should ideally check for properties of the config like templates, rather than the direct backend name checks for the lower half.
+// This avoids the maintenance burden of updating this list for each new backend - but unfortunately, that's the best option for some services currently.
+func (c *BackendConfig) GuessUsecases(u BackendConfigUsecases) bool {
+	if (u & FLAG_CHAT) == FLAG_CHAT {
+		if c.TemplateConfig.Chat == "" && c.TemplateConfig.ChatMessage == "" {
+			return false
+		}
+	}
+	if (u & FLAG_COMPLETION) == FLAG_COMPLETION {
+		if c.TemplateConfig.Completion == "" {
+			return false
+		}
+	}
+	if (u & FLAG_EDIT) == FLAG_EDIT {
+		if c.TemplateConfig.Edit == "" {
+			return false
+		}
+	}
+	if (u & FLAG_EMBEDDINGS) == FLAG_EMBEDDINGS {
+		if c.Embeddings == nil || !*c.Embeddings {
+			return false
+		}
+	}
+	if (u & FLAG_IMAGE) == FLAG_IMAGE {
+		imageBackends := []string{"diffusers", "tinydream", "stablediffusion"}
+		if !slices.Contains(imageBackends, c.Backend) {
+			return false
+		}
+
+		if c.Backend == "diffusers" && c.Diffusers.PipelineType == "" {
+			return false
+		}
+
+	}
+	if (u & FLAG_RERANK) == FLAG_RERANK {
+		if c.Backend != "rerankers" {
+			return false
+		}
+	}
+	if (u & FLAG_TRANSCRIPT) == FLAG_TRANSCRIPT {
+		if c.Backend != "whisper" {
+			return false
+		}
+	}
+	if (u & FLAG_TTS) == FLAG_TTS {
+		ttsBackends := []string{"piper", "transformers-musicgen", "parler-tts"}
+		if !slices.Contains(ttsBackends, c.Backend) {
+			return false
+		}
+	}
+
+	if (u & FLAG_SOUND_GENERATION) == FLAG_SOUND_GENERATION {
+		if c.Backend != "transformers-musicgen" {
+			return false
+		}
+	}
+
+	return true
+}

core/config/backend_config_filter.go

@@ -0,0 +1,35 @@
+package config
+
+import "regexp"
+
+type BackendConfigFilterFn func(string, *BackendConfig) bool
+
+func NoFilterFn(_ string, _ *BackendConfig) bool { return true }
+
+func BuildNameFilterFn(filter string) (BackendConfigFilterFn, error) {
+	if filter == "" {
+		return NoFilterFn, nil
+	}
+	rxp, err := regexp.Compile(filter)
+	if err != nil {
+		return nil, err
+	}
+	return func(name string, config *BackendConfig) bool {
+		if config != nil {
+			return rxp.MatchString(config.Name)
+		}
+		return rxp.MatchString(name)
+	}, nil
+}
+
+func BuildUsecaseFilterFn(usecases BackendConfigUsecases) BackendConfigFilterFn {
+	if usecases == FLAG_ANY {
+		return NoFilterFn
+	}
+	return func(name string, config *BackendConfig) bool {
+		if config == nil {
+			return false // TODO: Potentially make this a param, for now, no known usecase to include
+		}
+		return config.HasUsecases(usecases)
+	}
+}

core/config/backend_config_loader.go

@@ -201,6 +201,26 @@ func (bcl *BackendConfigLoader) GetAllBackendConfigs() []BackendConfig {
 	return res
 }
 
+func (bcl *BackendConfigLoader) GetBackendConfigsByFilter(filter BackendConfigFilterFn) []BackendConfig {
+	bcl.Lock()
+	defer bcl.Unlock()
+	var res []BackendConfig
+
+	if filter == nil {
+		filter = NoFilterFn
+	}
+
+	for n, v := range bcl.configs {
+		if filter(n, &v) {
+			res = append(res, v)
+		}
+	}
+
+	// TODO: I don't think this one needs to Sort on name... but we'll see what breaks.
+
+	return res
+}
+
 func (bcl *BackendConfigLoader) RemoveBackendConfig(m string) {
 	bcl.Lock()
 	defer bcl.Unlock()

core/config/backend_config_test.go

@@ -19,12 +19,17 @@ var _ = Describe("Test cases for config related functions", func() {
 				`backend: "../foo-bar"
 name: "foo"
 parameters:
-  model: "foo-bar"`)
+  model: "foo-bar"
+known_usecases:
+- chat
+- COMPLETION
+`)
 			Expect(err).ToNot(HaveOccurred())
 			config, err := readBackendConfigFromFile(tmp.Name())
 			Expect(err).To(BeNil())
 			Expect(config).ToNot(BeNil())
 			Expect(config.Validate()).To(BeFalse())
+			Expect(config.KnownUsecases).ToNot(BeNil())
 		})
 		It("Test Validate", func() {
 			tmp, err := os.CreateTemp("", "config.yaml")
@@ -61,4 +66,99 @@ parameters:
 			Expect(config.Validate()).To(BeTrue())
 		})
 	})
+	It("Properly handles backend usecase matching", func() {
+
+		a := BackendConfig{
+			Name: "a",
+		}
+		Expect(a.HasUsecases(FLAG_ANY)).To(BeTrue()) // FLAG_ANY just means the config _exists_ essentially.
+
+		b := BackendConfig{
+			Name:    "b",
+			Backend: "stablediffusion",
+		}
+		Expect(b.HasUsecases(FLAG_ANY)).To(BeTrue())
+		Expect(b.HasUsecases(FLAG_IMAGE)).To(BeTrue())
+		Expect(b.HasUsecases(FLAG_CHAT)).To(BeFalse())
+
+		c := BackendConfig{
+			Name:    "c",
+			Backend: "llama-cpp",
+			TemplateConfig: TemplateConfig{
+				Chat: "chat",
+			},
+		}
+		Expect(c.HasUsecases(FLAG_ANY)).To(BeTrue())
+		Expect(c.HasUsecases(FLAG_IMAGE)).To(BeFalse())
+		Expect(c.HasUsecases(FLAG_COMPLETION)).To(BeFalse())
+		Expect(c.HasUsecases(FLAG_CHAT)).To(BeTrue())
+
+		d := BackendConfig{
+			Name:    "d",
+			Backend: "llama-cpp",
+			TemplateConfig: TemplateConfig{
+				Chat:       "chat",
+				Completion: "completion",
+			},
+		}
+		Expect(d.HasUsecases(FLAG_ANY)).To(BeTrue())
+		Expect(d.HasUsecases(FLAG_IMAGE)).To(BeFalse())
+		Expect(d.HasUsecases(FLAG_COMPLETION)).To(BeTrue())
+		Expect(d.HasUsecases(FLAG_CHAT)).To(BeTrue())
+
+		trueValue := true
+		e := BackendConfig{
+			Name:    "e",
+			Backend: "llama-cpp",
+			TemplateConfig: TemplateConfig{
+				Completion: "completion",
+			},
+			Embeddings: &trueValue,
+		}
+
+		Expect(e.HasUsecases(FLAG_ANY)).To(BeTrue())
+		Expect(e.HasUsecases(FLAG_IMAGE)).To(BeFalse())
+		Expect(e.HasUsecases(FLAG_COMPLETION)).To(BeTrue())
+		Expect(e.HasUsecases(FLAG_CHAT)).To(BeFalse())
+		Expect(e.HasUsecases(FLAG_EMBEDDINGS)).To(BeTrue())
+
+		f := BackendConfig{
+			Name:    "f",
+			Backend: "piper",
+		}
+		Expect(f.HasUsecases(FLAG_ANY)).To(BeTrue())
+		Expect(f.HasUsecases(FLAG_TTS)).To(BeTrue())
+		Expect(f.HasUsecases(FLAG_CHAT)).To(BeFalse())
+
+		g := BackendConfig{
+			Name:    "g",
+			Backend: "whisper",
+		}
+		Expect(g.HasUsecases(FLAG_ANY)).To(BeTrue())
+		Expect(g.HasUsecases(FLAG_TRANSCRIPT)).To(BeTrue())
+		Expect(g.HasUsecases(FLAG_TTS)).To(BeFalse())
+
+		h := BackendConfig{
+			Name:    "h",
+			Backend: "transformers-musicgen",
+		}
+		Expect(h.HasUsecases(FLAG_ANY)).To(BeTrue())
+		Expect(h.HasUsecases(FLAG_TRANSCRIPT)).To(BeFalse())
+		Expect(h.HasUsecases(FLAG_TTS)).To(BeTrue())
+		Expect(h.HasUsecases(FLAG_SOUND_GENERATION)).To(BeTrue())
+
+		knownUsecases := FLAG_CHAT | FLAG_COMPLETION
+		i := BackendConfig{
+			Name:    "i",
+			Backend: "whisper",
+			// Earlier test checks parsing, this just needs to set final values
+			KnownUsecases: &knownUsecases,
+		}
+		Expect(i.HasUsecases(FLAG_ANY)).To(BeTrue())
+		Expect(i.HasUsecases(FLAG_TRANSCRIPT)).To(BeTrue())
+		Expect(i.HasUsecases(FLAG_TTS)).To(BeFalse())
+		Expect(i.HasUsecases(FLAG_COMPLETION)).To(BeTrue())
+		Expect(i.HasUsecases(FLAG_CHAT)).To(BeTrue())
+
+	})
 })

core/gallery/gallery.go

@@ -132,7 +132,7 @@ func AvailableGalleryModels(galleries []config.Gallery, basePath string) ([]*Gal
 func findGalleryURLFromReferenceURL(url string, basePath string) (string, error) {
 	var refFile string
 	uri := downloader.URI(url)
-	err := uri.DownloadAndUnmarshal(basePath, func(url string, d []byte) error {
+	err := uri.DownloadWithCallback(basePath, func(url string, d []byte) error {
 		refFile = string(d)
 		if len(refFile) == 0 {
 			return fmt.Errorf("invalid reference file at url %s: %s", url, d)
@@ -156,7 +156,7 @@ func getGalleryModels(gallery config.Gallery, basePath string) ([]*GalleryModel,
 	}
 	uri := downloader.URI(gallery.URL)
 
-	err := uri.DownloadAndUnmarshal(basePath, func(url string, d []byte) error {
+	err := uri.DownloadWithCallback(basePath, func(url string, d []byte) error {
 		return yaml.Unmarshal(d, &models)
 	})
 	if err != nil {

core/gallery/models.go

@@ -69,7 +69,7 @@ type PromptTemplate struct {
 func GetGalleryConfigFromURL(url string, basePath string) (Config, error) {
 	var config Config
 	uri := downloader.URI(url)
-	err := uri.DownloadAndUnmarshal(basePath, func(url string, d []byte) error {
+	err := uri.DownloadWithCallback(basePath, func(url string, d []byte) error {
 		return yaml.Unmarshal(d, &config)
 	})
 	if err != nil {

core/http/app.go

@@ -3,13 +3,15 @@ package http
 import (
 	"embed"
 	"errors"
+	"fmt"
 	"net/http"
-	"strings"
 
+	"github.com/dave-gray101/v2keyauth"
 	"github.com/mudler/LocalAI/pkg/utils"
 
 	"github.com/mudler/LocalAI/core/http/endpoints/localai"
 	"github.com/mudler/LocalAI/core/http/endpoints/openai"
+	"github.com/mudler/LocalAI/core/http/middleware"
 	"github.com/mudler/LocalAI/core/http/routes"
 
 	"github.com/mudler/LocalAI/core/config"
@@ -29,24 +31,6 @@ import (
 	"github.com/rs/zerolog/log"
 )
 
-func readAuthHeader(c *fiber.Ctx) string {
-	authHeader := c.Get("Authorization")
-
-	// elevenlabs
-	xApiKey := c.Get("xi-api-key")
-	if xApiKey != "" {
-		authHeader = "Bearer " + xApiKey
-	}
-
-	// anthropic
-	xApiKey = c.Get("x-api-key")
-	if xApiKey != "" {
-		authHeader = "Bearer " + xApiKey
-	}
-
-	return authHeader
-}
-
 // Embed a directory
 //
 //go:embed static/*
@@ -137,37 +121,17 @@ func App(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfig *confi
 		})
 	}
 
-	// Auth middleware checking if API key is valid. If no API key is set, no auth is required.
-	auth := func(c *fiber.Ctx) error {
-		if len(appConfig.ApiKeys) == 0 {
-			return c.Next()
-		}
+ // Health Checks should always be exempt from auth, so register these first
+	routes.HealthRoutes(app)
 
-		if len(appConfig.ApiKeys) == 0 {
-			return c.Next()
-		}
-
-		authHeader := readAuthHeader(c)
-		if authHeader == "" {
-			return c.Status(fiber.StatusUnauthorized).JSON(fiber.Map{"message": "Authorization header missing"})
-		}
-
-		// If it's a bearer token
-		authHeaderParts := strings.Split(authHeader, " ")
-		if len(authHeaderParts) != 2 || authHeaderParts[0] != "Bearer" {
-			return c.Status(fiber.StatusUnauthorized).JSON(fiber.Map{"message": "Invalid Authorization header format"})
-		}
-
-		apiKey := authHeaderParts[1]
-		for _, key := range appConfig.ApiKeys {
-			if apiKey == key {
-				return c.Next()
-			}
-		}
-
-		return c.Status(fiber.StatusUnauthorized).JSON(fiber.Map{"message": "Invalid API key"})
+	kaConfig, err := middleware.GetKeyAuthConfig(appConfig)
+	if err != nil || kaConfig == nil {
+		return nil, fmt.Errorf("failed to create key auth config: %w", err)
 	}
 
+	// Auth is applied to _all_ endpoints. No exceptions. Filtering out endpoints to bypass is the role of the Filter property of the KeyAuth Configuration
+	app.Use(v2keyauth.New(*kaConfig))
+
 	if appConfig.CORS {
 		var c func(ctx *fiber.Ctx) error
 		if appConfig.CORSAllowOrigins == "" {
@@ -192,13 +156,13 @@ func App(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfig *confi
 	galleryService := services.NewGalleryService(appConfig)
 	galleryService.Start(appConfig.Context, cl)
 
-	routes.RegisterElevenLabsRoutes(app, cl, ml, appConfig, auth)
-	routes.RegisterLocalAIRoutes(app, cl, ml, appConfig, galleryService, auth)
-	routes.RegisterOpenAIRoutes(app, cl, ml, appConfig, auth)
+	routes.RegisterElevenLabsRoutes(app, cl, ml, appConfig)
+	routes.RegisterLocalAIRoutes(app, cl, ml, appConfig, galleryService)
+	routes.RegisterOpenAIRoutes(app, cl, ml, appConfig)
 	if !appConfig.DisableWebUI {
-		routes.RegisterUIRoutes(app, cl, ml, appConfig, galleryService, auth)
+		routes.RegisterUIRoutes(app, cl, ml, appConfig, galleryService)
 	}
-	routes.RegisterJINARoutes(app, cl, ml, appConfig, auth)
+	routes.RegisterJINARoutes(app, cl, ml, appConfig)
 
 	httpFS := http.FS(embedDirStatic)
 

core/http/app_test.go

@@ -12,6 +12,7 @@ import (
 	"os"
 	"path/filepath"
 	"runtime"
+	"strings"
 
 	"github.com/mudler/LocalAI/core/config"
 	. "github.com/mudler/LocalAI/core/http"
@@ -31,6 +32,9 @@ import (
 	"github.com/sashabaranov/go-openai/jsonschema"
 )
 
+const apiKey = "joshua"
+const bearerKey = "Bearer " + apiKey
+
 const testPrompt = `### System:
 You are an AI assistant that follows instruction extremely well. Help as much as you can.
 
@@ -50,11 +54,19 @@ type modelApplyRequest struct {
 
 func getModelStatus(url string) (response map[string]interface{}) {
 	// Create the HTTP request
-	resp, err := http.Get(url)
+	req, err := http.NewRequest("GET", url, nil)
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", bearerKey)
 	if err != nil {
 		fmt.Println("Error creating request:", err)
 		return
 	}
+	client := &http.Client{}
+	resp, err := client.Do(req)
+	if err != nil {
+		fmt.Println("Error sending request:", err)
+		return
+	}
 	defer resp.Body.Close()
 
 	body, err := io.ReadAll(resp.Body)
@@ -72,14 +84,15 @@ func getModelStatus(url string) (response map[string]interface{}) {
 	return
 }
 
-func getModels(url string) (response []gallery.GalleryModel) {
+func getModels(url string) ([]gallery.GalleryModel, error) {
+	response := []gallery.GalleryModel{}
 	uri := downloader.URI(url)
 	// TODO: No tests currently seem to exercise file:// urls. Fix?
-	uri.DownloadAndUnmarshal("", func(url string, i []byte) error {
+	err := uri.DownloadWithAuthorizationAndCallback("", bearerKey, func(url string, i []byte) error {
 		// Unmarshal YAML data into a struct
 		return json.Unmarshal(i, &response)
 	})
-	return
+	return response, err
 }
 
 func postModelApplyRequest(url string, request modelApplyRequest) (response map[string]interface{}) {
@@ -101,6 +114,7 @@ func postModelApplyRequest(url string, request modelApplyRequest) (response map[
 		return
 	}
 	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", bearerKey)
 
 	// Make the request
 	client := &http.Client{}
@@ -140,6 +154,7 @@ func postRequestJSON[B any](url string, bodyJson *B) error {
 	}
 
 	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", bearerKey)
 
 	client := &http.Client{}
 	resp, err := client.Do(req)
@@ -175,6 +190,7 @@ func postRequestResponseJSON[B1 any, B2 any](url string, reqJson *B1, respJson *
 	}
 
 	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", bearerKey)
 
 	client := &http.Client{}
 	resp, err := client.Do(req)
@@ -195,6 +211,35 @@ func postRequestResponseJSON[B1 any, B2 any](url string, reqJson *B1, respJson *
 	return json.Unmarshal(body, respJson)
 }
 
+func postInvalidRequest(url string) (error, int) {
+
+	req, err := http.NewRequest("POST", url, bytes.NewBufferString("invalid request"))
+	if err != nil {
+		return err, -1
+	}
+
+	req.Header.Set("Content-Type", "application/json")
+
+	client := &http.Client{}
+	resp, err := client.Do(req)
+	if err != nil {
+		return err, -1
+	}
+
+	defer resp.Body.Close()
+
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return err, -1
+	}
+
+	if resp.StatusCode < 200 || resp.StatusCode >= 400 {
+		return fmt.Errorf("unexpected status code: %d, body: %s", resp.StatusCode, string(body)), resp.StatusCode
+	}
+
+	return nil, resp.StatusCode
+}
+
 //go:embed backend-assets/*
 var backendAssets embed.FS
 
@@ -260,6 +305,7 @@ var _ = Describe("API test", func() {
 					config.WithContext(c),
 					config.WithGalleries(galleries),
 					config.WithModelPath(modelDir),
+					config.WithApiKeys([]string{apiKey}),
 					config.WithBackendAssets(backendAssets),
 					config.WithBackendAssetsOutput(backendAssetsDir))...)
 			Expect(err).ToNot(HaveOccurred())
@@ -269,7 +315,7 @@ var _ = Describe("API test", func() {
 
 			go app.Listen("127.0.0.1:9090")
 
-			defaultConfig := openai.DefaultConfig("")
+			defaultConfig := openai.DefaultConfig(apiKey)
 			defaultConfig.BaseURL = "http://127.0.0.1:9090/v1"
 
 			client2 = openaigo.NewClient("")
@@ -295,10 +341,19 @@ var _ = Describe("API test", func() {
 			Expect(err).To(HaveOccurred())
 		})
 
+		Context("Auth Tests", func() {
+			It("Should fail if the api key is missing", func() {
+				err, sc := postInvalidRequest("http://127.0.0.1:9090/models/available")
+				Expect(err).ToNot(BeNil())
+				Expect(sc).To(Equal(403))
+			})
+		})
+
 		Context("Applying models", func() {
 
 			It("applies models from a gallery", func() {
-				models := getModels("http://127.0.0.1:9090/models/available")
+				models, err := getModels("http://127.0.0.1:9090/models/available")
+				Expect(err).To(BeNil())
 				Expect(len(models)).To(Equal(2), fmt.Sprint(models))
 				Expect(models[0].Installed).To(BeFalse(), fmt.Sprint(models))
 				Expect(models[1].Installed).To(BeFalse(), fmt.Sprint(models))
@@ -331,7 +386,8 @@ var _ = Describe("API test", func() {
 				Expect(content["backend"]).To(Equal("bert-embeddings"))
 				Expect(content["foo"]).To(Equal("bar"))
 
-				models = getModels("http://127.0.0.1:9090/models/available")
+				models, err = getModels("http://127.0.0.1:9090/models/available")
+				Expect(err).To(BeNil())
 				Expect(len(models)).To(Equal(2), fmt.Sprint(models))
 				Expect(models[0].Name).To(Or(Equal("bert"), Equal("bert2")))
 				Expect(models[1].Name).To(Or(Equal("bert"), Equal("bert2")))
@@ -895,7 +951,7 @@ var _ = Describe("API test", func() {
 					openai.ChatCompletionRequest{Model: "rwkv_test", Messages: []openai.ChatCompletionMessage{{Content: "Can you count up to five?", Role: "user"}}})
 				Expect(err).ToNot(HaveOccurred())
 				Expect(len(resp.Choices) > 0).To(BeTrue())
-				Expect(resp.Choices[0].Message.Content).To(Or(ContainSubstring("Sure"), ContainSubstring("five")))
+				Expect(strings.ToLower(resp.Choices[0].Message.Content)).To(Or(ContainSubstring("sure"), ContainSubstring("five")))
 
 				stream, err := client.CreateChatCompletionStream(context.TODO(), openai.ChatCompletionRequest{Model: "rwkv_test", Messages: []openai.ChatCompletionMessage{{Content: "Can you count up to five?", Role: "user"}}})
 				Expect(err).ToNot(HaveOccurred())
@@ -914,7 +970,7 @@ var _ = Describe("API test", func() {
 					tokens++
 				}
 				Expect(text).ToNot(BeEmpty())
-				Expect(text).To(Or(ContainSubstring("Sure"), ContainSubstring("five")))
+				Expect(strings.ToLower(text)).To(Or(ContainSubstring("sure"), ContainSubstring("five")))
 
 				Expect(tokens).ToNot(Or(Equal(1), Equal(0)))
 			})

core/http/ctx/fiber.go

@@ -19,14 +19,16 @@ func ModelFromContext(ctx *fiber.Ctx, cl *config.BackendConfigLoader, loader *mo
 	if ctx.Params("model") != "" {
 		modelInput = ctx.Params("model")
 	}
-
+	if ctx.Query("model") != "" {
+		modelInput = ctx.Query("model")
+	}
 	// Set model from bearer token, if available
-	bearer := strings.TrimLeft(ctx.Get("authorization"), "Bearer ")
+	bearer := strings.TrimLeft(ctx.Get("authorization"), "Bear ") // Reduced duplicate characters of Bearer
 	bearerExists := bearer != "" && loader.ExistsInModelPath(bearer)
 
 	// If no model was specified, take the first available
 	if modelInput == "" && !bearerExists && firstModel {
-		models, _ := services.ListModels(cl, loader, "", true)
+		models, _ := services.ListModels(cl, loader, config.NoFilterFn, services.SKIP_IF_CONFIGURED)
 		if len(models) > 0 {
 			modelInput = models[0]
 			log.Debug().Msgf("No model specified, using: %s", modelInput)

core/http/elements/gallery.go

@@ -6,6 +6,7 @@ import (
 
 	"github.com/chasefleming/elem-go"
 	"github.com/chasefleming/elem-go/attrs"
+	"github.com/microcosm-cc/bluemonday"
 	"github.com/mudler/LocalAI/core/gallery"
 	"github.com/mudler/LocalAI/core/p2p"
 	"github.com/mudler/LocalAI/core/services"
@@ -41,7 +42,7 @@ func DoneProgress(galleryID, text string, showDelete bool) string {
 				"tabindex":  "-1",
 				"autofocus": "",
 			},
-			elem.Text(text),
+			elem.Text(bluemonday.StrictPolicy().Sanitize(text)),
 		),
 		elem.If(showDelete, deleteButton(galleryID, modelName), reInstallButton(galleryID)),
 	).Render()
@@ -57,7 +58,7 @@ func ErrorProgress(err, galleryName string) string {
 				"tabindex":  "-1",
 				"autofocus": "",
 			},
-			elem.Text("Error "+err),
+			elem.Text("Error "+bluemonday.StrictPolicy().Sanitize(err)),
 		),
 		installButton(galleryName),
 	).Render()
@@ -170,7 +171,7 @@ func P2PNodeBoxes(nodes []p2p.NodeData) string {
 						attrs.Props{
 							"class": "text-gray-200 font-semibold ml-2 mr-1",
 						},
-						elem.Text(n.ID),
+						elem.Text(bluemonday.StrictPolicy().Sanitize(n.ID)),
 					),
 					elem.Text("Status: "),
 					elem.If(
@@ -227,7 +228,7 @@ func StartProgressBar(uid, progress, text string) string {
 				"tabindex":  "-1",
 				"autofocus": "",
 			},
-			elem.Text(text),
+			elem.Text(bluemonday.StrictPolicy().Sanitize(text)), //Perhaps overly defensive
 			elem.Div(attrs.Props{
 				"hx-get":     "/browse/job/progress/" + uid,
 				"hx-trigger": "every 600ms",
@@ -249,9 +250,7 @@ func cardSpan(text, icon string) elem.Node {
 			"class": icon + " pr-2",
 		}),
 
-		elem.Text(text),
-
-		//elem.Text(text),
+		elem.Text(bluemonday.StrictPolicy().Sanitize(text)),
 	)
 }
 
@@ -285,11 +284,9 @@ func searchableElement(text, icon string) elem.Node {
 				elem.I(attrs.Props{
 					"class": icon + " pr-2",
 				}),
-				elem.Text(text),
+				elem.Text(bluemonday.StrictPolicy().Sanitize(text)),
 			),
 		),
-
-		//elem.Text(text),
 	)
 }
 
@@ -303,7 +300,7 @@ func link(text, url string) elem.Node {
 		elem.I(attrs.Props{
 			"class": "fas fa-link pr-2",
 		}),
-		elem.Text(text),
+		elem.Text(bluemonday.StrictPolicy().Sanitize(text)),
 	)
 }
 func installButton(galleryName string) elem.Node {
@@ -387,13 +384,13 @@ func ListModels(models []*gallery.GalleryModel, processTracker ProcessTracker, g
 				attrs.Props{
 					"class": "mb-2 text-xl font-bold leading-tight",
 				},
-				elem.Text(m.Name),
+				elem.Text(bluemonday.StrictPolicy().Sanitize(m.Name)),
 			),
 			elem.P(
 				attrs.Props{
 					"class": "mb-4 text-sm [&:not(:hover)]:truncate text-base",
 				},
-				elem.Text(m.Description),
+				elem.Text(bluemonday.StrictPolicy().Sanitize(m.Description)),
 			),
 		)
 	}

core/http/endpoints/elevenlabs/soundgeneration.go

@@ -55,7 +55,7 @@ func SoundGenerationEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoad
 		}
 
 		// TODO: Support uploading files?
-		filePath, _, err := backend.SoundGeneration(cfg.Backend, modelFile, input.Text, input.Duration, input.Temperature, input.DoSample, nil, nil, ml, appConfig, *cfg)
+		filePath, _, err := backend.SoundGeneration(modelFile, input.Text, input.Duration, input.Temperature, input.DoSample, nil, nil, ml, appConfig, *cfg)
 		if err != nil {
 			return err
 		}

core/http/endpoints/jina/rerank.go

@@ -45,13 +45,13 @@ func JINARerankEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, a
 			config.LoadOptionContextSize(appConfig.ContextSize),
 			config.LoadOptionF16(appConfig.F16),
 		)
-
 		if err != nil {
 			modelFile = input.Model
 			log.Warn().Msgf("Model not found in context: %s", input.Model)
 		} else {
 			modelFile = cfg.Model
 		}
+
 		log.Debug().Msgf("Request for model: %s", modelFile)
 
 		if input.Backend != "" {
@@ -64,7 +64,7 @@ func JINARerankEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, a
 			Documents: req.Documents,
 		}
 
-		results, err := backend.Rerank(cfg.Backend, modelFile, request, ml, appConfig, *cfg)
+		results, err := backend.Rerank(modelFile, request, ml, appConfig, *cfg)
 		if err != nil {
 			return err
 		}

core/http/endpoints/localai/get_token_metrics.go

@@ -0,0 +1,60 @@
+package localai
+
+import (
+	"github.com/gofiber/fiber/v2"
+	"github.com/mudler/LocalAI/core/backend"
+	"github.com/mudler/LocalAI/core/config"
+	fiberContext "github.com/mudler/LocalAI/core/http/ctx"
+	"github.com/mudler/LocalAI/core/schema"
+	"github.com/rs/zerolog/log"
+
+	"github.com/mudler/LocalAI/pkg/model"
+)
+
+// TokenMetricsEndpoint is an endpoint to get TokensProcessed Per Second for Active SlotID
+//
+//	@Summary	Get TokenMetrics for Active Slot.
+//	@Accept json
+//	@Produce audio/x-wav
+//	@Success	200		{string}	binary				"generated audio/wav file"
+//	@Router		/v1/tokenMetrics [get]
+//	@Router		/tokenMetrics [get]
+func TokenMetricsEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfig *config.ApplicationConfig) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+
+		input := new(schema.TokenMetricsRequest)
+
+		// Get input data from the request body
+		if err := c.BodyParser(input); err != nil {
+			return err
+		}
+
+		modelFile, err := fiberContext.ModelFromContext(c, cl, ml, input.Model, false)
+		if err != nil {
+			modelFile = input.Model
+			log.Warn().Msgf("Model not found in context: %s", input.Model)
+		}
+
+		cfg, err := cl.LoadBackendConfigFileByName(modelFile, appConfig.ModelPath,
+			config.LoadOptionDebug(appConfig.Debug),
+			config.LoadOptionThreads(appConfig.Threads),
+			config.LoadOptionContextSize(appConfig.ContextSize),
+			config.LoadOptionF16(appConfig.F16),
+		)
+
+		if err != nil {
+			log.Err(err)
+			modelFile = input.Model
+			log.Warn().Msgf("Model not found in context: %s", input.Model)
+		} else {
+			modelFile = cfg.Model
+		}
+		log.Debug().Msgf("Token Metrics for model: %s", modelFile)
+
+		response, err := backend.TokenMetrics(modelFile, ml, appConfig, *cfg)
+		if err != nil {
+			return err
+		}
+		return c.JSON(response)
+	}
+}

core/http/endpoints/localai/system.go

@@ -17,12 +17,14 @@ func SystemInformations(ml *model.ModelLoader, appConfig *config.ApplicationConf
 		if err != nil {
 			return err
 		}
+		loadedModels := ml.ListModels()
 		for b := range appConfig.ExternalGRPCBackends {
 			availableBackends = append(availableBackends, b)
 		}
 		return c.JSON(
 			schema.SystemInformationResponse{
 				Backends: availableBackends,
+				Models:   loadedModels,
 			},
 		)
 	}

core/http/endpoints/localai/tokenize.go

@@ -0,0 +1,58 @@
+package localai
+
+import (
+	"github.com/gofiber/fiber/v2"
+	"github.com/mudler/LocalAI/core/backend"
+	"github.com/mudler/LocalAI/core/config"
+	fiberContext "github.com/mudler/LocalAI/core/http/ctx"
+	"github.com/mudler/LocalAI/core/schema"
+	"github.com/mudler/LocalAI/pkg/model"
+	"github.com/rs/zerolog/log"
+)
+
+// TokenizeEndpoint exposes a REST API to tokenize the content
+// @Summary Tokenize the input.
+// @Success 200 {object} schema.TokenizeResponse "Response"
+// @Router /v1/tokenize [post]
+func TokenizeEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfig *config.ApplicationConfig) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+
+		input := new(schema.TokenizeRequest)
+
+		// Get input data from the request body
+		if err := c.BodyParser(input); err != nil {
+			return err
+		}
+
+		modelFile, err := fiberContext.ModelFromContext(c, cl, ml, input.Model, false)
+		if err != nil {
+			modelFile = input.Model
+			log.Warn().Msgf("Model not found in context: %s", input.Model)
+		}
+
+		cfg, err := cl.LoadBackendConfigFileByName(modelFile, appConfig.ModelPath,
+			config.LoadOptionDebug(appConfig.Debug),
+			config.LoadOptionThreads(appConfig.Threads),
+			config.LoadOptionContextSize(appConfig.ContextSize),
+			config.LoadOptionF16(appConfig.F16),
+		)
+
+		if err != nil {
+			log.Err(err)
+			modelFile = input.Model
+			log.Warn().Msgf("Model not found in context: %s", input.Model)
+		} else {
+			modelFile = cfg.Model
+		}
+		log.Debug().Msgf("Request for model: %s", modelFile)
+
+		tokenResponse, err := backend.ModelTokenize(input.Content, ml, *cfg, appConfig)
+		if err != nil {
+			return err
+		}
+
+		c.JSON(tokenResponse)
+		return nil
+
+	}
+}

core/http/endpoints/localai/welcome.go

@@ -13,15 +13,10 @@ import (
 func WelcomeEndpoint(appConfig *config.ApplicationConfig,
 	cl *config.BackendConfigLoader, ml *model.ModelLoader, modelStatus func() (map[string]string, map[string]string)) func(*fiber.Ctx) error {
 	return func(c *fiber.Ctx) error {
-		models, _ := services.ListModels(cl, ml, "", true)
 		backendConfigs := cl.GetAllBackendConfigs()
-
 		galleryConfigs := map[string]*gallery.Config{}
-		modelsWithBackendConfig := map[string]interface{}{}
 
 		for _, m := range backendConfigs {
-			modelsWithBackendConfig[m.Name] = nil
-
 			cfg, err := gallery.GetLocalModelConfiguration(ml.ModelPath, m.Name)
 			if err != nil {
 				continue
@@ -29,17 +24,11 @@ func WelcomeEndpoint(appConfig *config.ApplicationConfig,
 			galleryConfigs[m.Name] = cfg
 		}
 
+		modelsWithoutConfig, _ := services.ListModels(cl, ml, config.NoFilterFn, services.LOOSE_ONLY)
+
 		// Get model statuses to display in the UI the operation in progress
 		processingModels, taskTypes := modelStatus()
 
-		modelsWithoutConfig := []string{}
-
-		for _, m := range models {
-			if _, ok := modelsWithBackendConfig[m]; !ok {
-				modelsWithoutConfig = append(modelsWithoutConfig, m)
-			}
-		}
-
 		summary := fiber.Map{
 			"Title":             "LocalAI API - " + internal.PrintableVersion(),
 			"Version":           internal.PrintableVersion(),

core/http/endpoints/openai/assistant.go

@@ -10,6 +10,7 @@ import (
 	"time"
 
 	"github.com/gofiber/fiber/v2"
+	"github.com/microcosm-cc/bluemonday"
 	"github.com/mudler/LocalAI/core/config"
 	"github.com/mudler/LocalAI/core/schema"
 	"github.com/mudler/LocalAI/core/services"
@@ -83,7 +84,7 @@ func CreateAssistantEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoad
 
 		if !modelExists(cl, ml, request.Model) {
 			log.Warn().Msgf("Model: %s was not found in list of models.", request.Model)
-			return c.Status(fiber.StatusBadRequest).SendString("Model " + request.Model + " not found")
+			return c.Status(fiber.StatusBadRequest).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Model %q not found", request.Model)))
 		}
 
 		if request.Tools == nil {
@@ -147,7 +148,7 @@ func ListAssistantsEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoade
 		// Convert string limit to integer
 		limit, err := strconv.Atoi(limitQuery)
 		if err != nil {
-			return c.Status(http.StatusBadRequest).SendString(fmt.Sprintf("Invalid limit query value: %s", limitQuery))
+			return c.Status(http.StatusBadRequest).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Invalid limit query value: %s", limitQuery)))
 		}
 
 		// Sort assistants
@@ -225,7 +226,7 @@ func filterAssistantsAfterID(assistants []Assistant, id string) []Assistant {
 
 func modelExists(cl *config.BackendConfigLoader, ml *model.ModelLoader, modelName string) (found bool) {
 	found = false
-	models, err := services.ListModels(cl, ml, "", true)
+	models, err := services.ListModels(cl, ml, config.NoFilterFn, services.SKIP_IF_CONFIGURED)
 	if err != nil {
 		return
 	}
@@ -288,7 +289,7 @@ func GetAssistantEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader,
 			}
 		}
 
-		return c.Status(fiber.StatusNotFound).SendString(fmt.Sprintf("Unable to find assistant with id: %s", assistantID))
+		return c.Status(fiber.StatusNotFound).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Unable to find assistant with id: %s", assistantID)))
 	}
 }
 
@@ -337,11 +338,11 @@ func CreateAssistantFileEndpoint(cl *config.BackendConfigLoader, ml *model.Model
 					}
 				}
 
-				return c.Status(fiber.StatusNotFound).SendString(fmt.Sprintf("Unable to find file_id: %s", request.FileID))
+				return c.Status(fiber.StatusNotFound).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Unable to find file_id: %s", request.FileID)))
 			}
 		}
 
-		return c.Status(fiber.StatusNotFound).SendString(fmt.Sprintf("Unable to find %q", assistantID))
+		return c.Status(fiber.StatusNotFound).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Unable to find %q", assistantID)))
 	}
 }
 
@@ -442,7 +443,7 @@ func ModifyAssistantEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoad
 				return c.Status(fiber.StatusOK).JSON(newAssistant)
 			}
 		}
-		return c.Status(fiber.StatusNotFound).SendString(fmt.Sprintf("Unable to find assistant with id: %s", assistantID))
+		return c.Status(fiber.StatusNotFound).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Unable to find assistant with id: %s", assistantID)))
 	}
 }
 
@@ -513,9 +514,9 @@ func GetAssistantFileEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoa
 				if assistantFile.ID == fileId {
 					return c.Status(fiber.StatusOK).JSON(assistantFile)
 				}
-				return c.Status(fiber.StatusNotFound).SendString(fmt.Sprintf("Unable to find assistant file with file_id: %s", fileId))
+				return c.Status(fiber.StatusNotFound).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Unable to find assistant file with file_id: %s", fileId)))
 			}
 		}
-		return c.Status(fiber.StatusNotFound).SendString(fmt.Sprintf("Unable to find assistant file with assistant_id: %s", assistantID))
+		return c.Status(fiber.StatusNotFound).SendString(bluemonday.StrictPolicy().Sanitize(fmt.Sprintf("Unable to find assistant file with assistant_id: %s", assistantID)))
 	}
 }

core/http/endpoints/openai/chat.go

@@ -161,6 +161,12 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, startup
 		textContentToReturn = ""
 		id = uuid.New().String()
 		created = int(time.Now().Unix())
+		// Set CorrelationID
+		correlationID := c.Get("X-Correlation-ID")
+		if len(strings.TrimSpace(correlationID)) == 0 {
+			correlationID = id
+		}
+		c.Set("X-Correlation-ID", correlationID)
 
 		modelFile, input, err := readRequest(c, cl, ml, startupOptions, true)
 		if err != nil {
@@ -444,6 +450,7 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, startup
 			c.Set("Cache-Control", "no-cache")
 			c.Set("Connection", "keep-alive")
 			c.Set("Transfer-Encoding", "chunked")
+			c.Set("X-Correlation-ID", id)
 
 			responses := make(chan schema.OpenAIResponse)
 
@@ -640,8 +647,16 @@ func handleQuestion(config *config.BackendConfig, input *schema.OpenAIRequest, m
 	for _, m := range input.Messages {
 		images = append(images, m.StringImages...)
 	}
+	videos := []string{}
+	for _, m := range input.Messages {
+		videos = append(videos, m.StringVideos...)
+	}
+	audios := []string{}
+	for _, m := range input.Messages {
+		audios = append(audios, m.StringAudios...)
+	}
 
-	predFunc, err := backend.ModelInference(input.Context, prompt, input.Messages, images, ml, *config, o, nil)
+	predFunc, err := backend.ModelInference(input.Context, prompt, input.Messages, images, videos, audios, ml, *config, o, nil)
 	if err != nil {
 		log.Error().Err(err).Msg("model inference failed")
 		return "", err

core/http/endpoints/openai/completion.go

@@ -57,6 +57,8 @@ func CompletionEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, a
 	}
 
 	return func(c *fiber.Ctx) error {
+		// Add Correlation
+		c.Set("X-Correlation-ID", id)
 		modelFile, input, err := readRequest(c, cl, ml, appConfig, true)
 		if err != nil {
 			return fmt.Errorf("failed reading parameters from request:%w", err)