wire to grpc

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

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

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

.bruno/LocalAI Test Requests/backend monitor/backend monitor.bru

@@ -1,17 +0,0 @@
-meta {
-  name: backend monitor
-  type: http
-  seq: 4
-}
-
-get {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/backend/monitor
-  body: json
-  auth: none
-}
-
-body:json {
-  {
-    "model": "{{DEFAULT_MODEL}}"
-  }
-}

.bruno/LocalAI Test Requests/backend monitor/backend-shutdown.bru

@@ -1,21 +0,0 @@
-meta {
-  name: backend-shutdown
-  type: http
-  seq: 3
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/backend/shutdown
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "model": "{{DEFAULT_MODEL}}"
-  }
-}

.bruno/LocalAI Test Requests/bruno.json

@@ -1,5 +0,0 @@
-{
-  "version": "1",
-  "name": "LocalAI Test Requests",
-  "type": "collection"
-}

.bruno/LocalAI Test Requests/environments/localhost.bru

@@ -1,6 +0,0 @@
-vars {
-  HOST: localhost
-  PORT: 8080
-  DEFAULT_MODEL: gpt-3.5-turbo
-  PROTOCOL: http://
-}

.bruno/LocalAI Test Requests/get models list.bru

@@ -1,11 +0,0 @@
-meta {
-  name: get models list
-  type: http
-  seq: 2
-}
-
-get {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models
-  body: none
-  auth: none
-}

.bruno/LocalAI Test Requests/image generation/Generate image.bru

@@ -1,25 +0,0 @@
-meta {
-  name: Generate image
-  type: http
-  seq: 1
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/v1/images/generations
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-    "prompt": "<positive prompt>|<negative prompt>",
-    "model": "model-name",
-    "step": 51,
-    "size": "1024x1024",
-    "image": ""
-  }
-}

.bruno/LocalAI Test Requests/llm text/-completions.bru

@@ -1,24 +0,0 @@
-meta {
-  name: -completions
-  type: http
-  seq: 4
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/completions
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "model": "{{DEFAULT_MODEL}}",
-      "prompt": "function downloadFile(string url, string outputPath) {",
-      "max_tokens": 256,
-      "temperature": 0.5
-  }
-}

.bruno/LocalAI Test Requests/llm text/-edits.bru

@@ -1,23 +0,0 @@
-meta {
-  name: -edits
-  type: http
-  seq: 5
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/edits
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "model": "{{DEFAULT_MODEL}}",
-      "input": "What day of the wek is it?",
-      "instruction": "Fix the spelling mistakes"
-  }
-}

.bruno/LocalAI Test Requests/llm text/-embeddings.bru

@@ -1,22 +0,0 @@
-meta {
-  name: -embeddings
-  type: http
-  seq: 6
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/embeddings
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "model": "{{DEFAULT_MODEL}}",
-      "input": "A STRANGE GAME.\nTHE ONLY WINNING MOVE IS NOT TO PLAY.\n\nHOW ABOUT A NICE GAME OF CHESS?"
-  }
-}

.bruno/LocalAI Test Requests/llm text/chat/chat completion -simple- 1 message-.bru

@@ -1,30 +0,0 @@
-meta {
-  name: chat completion -simple- 1 message-
-  type: http
-  seq: 4
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/chat/completions
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-    "model": "{{DEFAULT_MODEL}}",
-    "messages": [
-      {
-        "role": "user",
-        "content": "How could one use friction to cook an egg?"
-      }
-    ],
-    "max_tokens": 256,
-    "temperature": 0.2,
-    "grammar": ""
-  }
-}

.bruno/LocalAI Test Requests/llm text/chat/chat-completions -long-.bru

@@ -1,29 +0,0 @@
-meta {
-  name: chat-completions -long-
-  type: http
-  seq: 5
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/chat/completions
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-       "model": "{{DEFAULT_MODEL}}",
-       "messages": [{"role": "system", "content": "You are a helpful, intelligent pirate assistant. Always answer as helpfully as possible, while being a pirate and using pirate language. If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information."},
-           {"role": "user", "content": "How could one use electricity to cook an egg?"},
-           {"role": "assistant",
-                  "content": "Shiver me timbers! Using electricity to cook an egg? Well, matey, I reckon that be a right curious idea! *adjusts spectacles* Now, I ain't sure if this be possible, but I'll do me best to help ye out.\n\nFirst things first, ye gotta understand that electricity be a powerful force, and it can't just cook an egg on its own. Ye see, electricity be like a mighty wind that can make things happen, but it needs somethin' to work with. So, if ye want to use electricity to cook an egg, ye gotta find a way to harness that power and make it do the cookin'.\n\nNow, I know what ye might be thinkin': \"How do I harness the power of electricity to cook an egg?\" Well, matey, there be a few ways to do it. One way be to use a special device called an \"electric frying pan.\" This be a pan that has a built-in heating element that gets hot when ye plug it into a wall socket. When the element gets hot, ye can crack an egg into the pan and watch as it cook"
-              },
-              {"role": "user", "content": "I don't have one of those, just a raw wire and plenty of power! How do we get it done?"}],
-       "max_tokens": 1024,
-       "temperature": 0.5
-  }
-}

.bruno/LocalAI Test Requests/llm text/chat/chat-completions -stream-.bru

@@ -1,25 +0,0 @@
-meta {
-  name: chat-completions -stream-
-  type: http
-  seq: 6
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/chat/completions
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-       "model": "{{DEFAULT_MODEL}}",
-       "messages": [{"role": "user", "content": "Explain how I can set sail on the ocean using only power generated by seagulls?"}],
-       "max_tokens": 256,
-       "temperature": 0.9,
-       "stream": true
-  }
-}

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

@@ -1,22 +0,0 @@
-meta {
-  name: add model gallery
-  type: http
-  seq: 10
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models/galleries
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "url": "file:///home/dave/projects/model-gallery/huggingface/TheBloke__CodeLlama-7B-Instruct-GGML.yaml",
-      "name": "test"
-  }
-}

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

@@ -1,21 +0,0 @@
-meta {
-  name: delete model gallery
-  type: http
-  seq: 11
-}
-
-delete {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models/galleries
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "name": "test"
-  }
-}

.bruno/LocalAI Test Requests/model gallery/list MODELS in galleries.bru

@@ -1,11 +0,0 @@
-meta {
-  name: list MODELS in galleries
-  type: http
-  seq: 7
-}
-
-get {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models/available
-  body: none
-  auth: none
-}

.bruno/LocalAI Test Requests/model gallery/list model GALLERIES.bru

@@ -1,11 +0,0 @@
-meta {
-  name: list model GALLERIES
-  type: http
-  seq: 8
-}
-
-get {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models/galleries
-  body: none
-  auth: none
-}

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

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

.bruno/LocalAI Test Requests/model gallery/model gallery apply -gist-.bru

@@ -1,21 +0,0 @@
-meta {
-  name: model gallery apply -gist-
-  type: http
-  seq: 12
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models/apply
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "id": "TheBloke__CodeLlama-7B-Instruct-GGML__codellama-7b-instruct.ggmlv3.Q2_K.bin"
-  }
-}

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

@@ -1,22 +0,0 @@
-meta {
-  name: model gallery apply
-  type: http
-  seq: 9
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/models/apply
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "id": "dave@TheBloke__CodeLlama-7B-Instruct-GGML__codellama-7b-instruct.ggmlv3.Q3_K_S.bin",
-      "name": "codellama7b"
-  }
-}

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

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

.bruno/LocalAI Test Requests/tts/-tts.bru

@@ -1,22 +0,0 @@
-meta {
-  name: -tts
-  type: http
-  seq: 2
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/tts
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "model": "{{DEFAULT_MODEL}}",
-      "input": "A STRANGE GAME.\nTHE ONLY WINNING MOVE IS NOT TO PLAY.\n\nHOW ABOUT A NICE GAME OF CHESS?"
-  }
-}

.bruno/LocalAI Test Requests/tts/musicgen.bru

@@ -1,23 +0,0 @@
-meta {
-  name: musicgen
-  type: http
-  seq: 2
-}
-
-post {
-  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/tts
-  body: json
-  auth: none
-}
-
-headers {
-  Content-Type: application/json
-}
-
-body:json {
-  {
-      "backend": "transformers",
-      "model": "facebook/musicgen-small",
-      "input": "80s Synths playing Jazz"
-  }
-}

.env

@@ -29,6 +29,9 @@
 ## Enable/Disable single backend (useful if only one GPU is available)
 # LOCALAI_SINGLE_ACTIVE_BACKEND=true
 
+# Forces shutdown of the backends if busy (only if LOCALAI_SINGLE_ACTIVE_BACKEND is set)
+# LOCALAI_FORCE_BACKEND_SHUTDOWN=true
+
 ## Specify a build type. Available: cublas, openblas, clblas.
 ## cuBLAS: This is a GPU-accelerated version of the complete standard BLAS (Basic Linear Algebra Subprograms) library. It's provided by Nvidia and is part of their CUDA toolkit.
 ## OpenBLAS: This is an open-source implementation of the BLAS library that aims to provide highly optimized code for various platforms. It includes support for multi-threading and can be compiled to use hardware-specific features for additional performance. OpenBLAS can run on many kinds of hardware, including CPUs from Intel, AMD, and ARM.

.github/dependabot.yml

@@ -29,10 +29,6 @@ updates:
     schedule:
       # Check for updates to GitHub Actions every weekday
       interval: "weekly"
-  - package-ecosystem: "pip"
-    directory: "/backend/python/autogptq"
-    schedule:
-      interval: "weekly"
   - package-ecosystem: "pip"
     directory: "/backend/python/bark"
     schedule:

.github/labeler.yml

@@ -1,4 +1,4 @@
-enhancements:
+enhancement:
  - head-branch: ['^feature', 'feature']
 
 dependencies:

.github/workflows/bump_deps.yaml

@@ -9,7 +9,7 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - repository: "ggerganov/llama.cpp"
+          - repository: "ggml-org/llama.cpp"
             variable: "CPPLLAMA_VERSION"
             branch: "master"
           - repository: "ggerganov/whisper.cpp"

.github/workflows/dependabot_auto.yml

@@ -14,7 +14,7 @@ jobs:
     steps:
       - name: Dependabot metadata
         id: metadata
-        uses: dependabot/fetch-metadata@v2.2.0
+        uses: dependabot/fetch-metadata@v2.3.0
         with:
           github-token: "${{ secrets.GITHUB_TOKEN }}"
           skip-commit-verification: true

.github/workflows/deploy-explorer.yaml

@@ -33,7 +33,7 @@ jobs:
         run: |
           CGO_ENABLED=0 make build-api
       - name: rm
-        uses: appleboy/ssh-action@v1.2.0
+        uses: appleboy/ssh-action@v1.2.2
         with:
             host: ${{ secrets.EXPLORER_SSH_HOST }}
             username: ${{ secrets.EXPLORER_SSH_USERNAME }}
@@ -53,7 +53,7 @@ jobs:
             rm: true
             target: ./local-ai
       - name: restarting
-        uses: appleboy/ssh-action@v1.2.0
+        uses: appleboy/ssh-action@v1.2.2
         with:
             host: ${{ secrets.EXPLORER_SSH_HOST }}
             username: ${{ secrets.EXPLORER_SSH_USERNAME }}

.github/workflows/generate_grpc_cache.yaml

@@ -2,9 +2,10 @@ name: 'generate and publish GRPC docker caches'
 
 on:
   workflow_dispatch:
-  push:
-    branches:
-      - master
+
+  schedule:
+    # daily at midnight
+    - cron: '0 0 * * *'
 
 concurrency:
   group: grpc-cache-${{ github.head_ref || github.ref }}-${{ github.repository }}
@@ -16,7 +17,7 @@ jobs:
       matrix:
         include:
           - grpc-base-image: ubuntu:22.04
-            runs-on: 'ubuntu-latest'
+            runs-on: 'arc-runner-set'
             platforms: 'linux/amd64,linux/arm64'
     runs-on: ${{matrix.runs-on}}
     steps:

.github/workflows/generate_intel_image.yaml

@@ -15,7 +15,7 @@ jobs:
     strategy:
       matrix:
         include:
-          - base-image: intel/oneapi-basekit:2025.0.0-0-devel-ubuntu22.04
+          - base-image: intel/oneapi-basekit:2025.1.0-0-devel-ubuntu22.04
             runs-on: 'ubuntu-latest'
             platforms: 'linux/amd64'
     runs-on: ${{matrix.runs-on}}

.github/workflows/image.yml

@@ -75,6 +75,7 @@ jobs:
             grpc-base-image: "ubuntu:22.04"
             runs-on: 'arc-runner-set'
             makeflags: "--jobs=3 --output-sync=target"
+            latest-image: 'latest-gpu-hipblas-core'
           - build-type: 'hipblas'
             platforms: 'linux/amd64'
             tag-latest: 'false'
@@ -251,6 +252,7 @@ jobs:
             image-type: 'core'
             runs-on: 'arc-runner-set'
             makeflags: "--jobs=3 --output-sync=target"
+            latest-image: 'latest-gpu-intel-f16-core'
           - build-type: 'sycl_f32'
             platforms: 'linux/amd64'
             tag-latest: 'false'
@@ -261,6 +263,7 @@ jobs:
             image-type: 'core'
             runs-on: 'arc-runner-set'
             makeflags: "--jobs=3 --output-sync=target"
+            latest-image: 'latest-gpu-intel-f32-core'
 
   core-image-build:
     uses: ./.github/workflows/image_build.yml
@@ -339,6 +342,7 @@ jobs:
             base-image: "ubuntu:22.04"
             makeflags: "--jobs=4 --output-sync=target"
             skip-drivers: 'false'
+            latest-image: 'latest-gpu-nvidia-cuda-12-core'
           - build-type: 'cublas'
             cuda-major-version: "12"
             cuda-minor-version: "0"
@@ -351,17 +355,18 @@ jobs:
             base-image: "ubuntu:22.04"
             skip-drivers: 'false'
             makeflags: "--jobs=4 --output-sync=target"
+            latest-image: 'latest-gpu-nvidia-cuda-12-core'
           - build-type: 'vulkan'
             platforms: 'linux/amd64'
             tag-latest: 'false'
             tag-suffix: '-vulkan-ffmpeg-core'
-            latest-image: 'latest-vulkan-ffmpeg-core'
             ffmpeg: 'true'
             image-type: 'core'
             runs-on: 'arc-runner-set'
             base-image: "ubuntu:22.04"
             skip-drivers: 'false'
             makeflags: "--jobs=4 --output-sync=target"
+            latest-image: 'latest-gpu-vulkan-core'
   gh-runner:
     uses: ./.github/workflows/image_build.yml
     with:

.github/workflows/image_build.yml

@@ -310,6 +310,11 @@ jobs:
           tags: ${{ steps.meta_aio_dockerhub.outputs.tags }}
           labels: ${{ steps.meta_aio_dockerhub.outputs.labels }}
 
+      - name: Cleanup
+        run: |
+          docker builder prune -f
+          docker system prune --force --volumes --all
+
       - name: Latest tag
         # run this on branches, when it is a tag and there is a latest-image defined
         if: github.event_name != 'pull_request' && inputs.latest-image != ''  && github.ref_type == 'tag'

.github/workflows/notify-models.yaml

@@ -8,7 +8,7 @@ jobs:
   notify-discord:
     if: ${{ (github.event.pull_request.merged == true) && (contains(github.event.pull_request.labels.*.name, 'area/ai-model')) }}
     env:
-        MODEL_NAME: hermes-2-theta-llama-3-8b
+        MODEL_NAME: gemma-3-12b-it
     runs-on: ubuntu-latest
     steps:
     - uses: actions/checkout@v4
@@ -16,9 +16,9 @@ jobs:
         fetch-depth: 0 # needed to checkout all branches for this Action to work
     - uses: mudler/localai-github-action@v1
       with:
-        model: 'hermes-2-theta-llama-3-8b' # Any from models.localai.io, or from huggingface.com with: "huggingface://<repository>/file"
+        model: 'gemma-3-12b-it' # Any from models.localai.io, or from huggingface.com with: "huggingface://<repository>/file"
         # Check the PR diff using the current branch and the base branch of the PR
-    - uses: GrantBirki/git-diff-action@v2.7.0
+    - uses: GrantBirki/git-diff-action@v2.8.0
       id: git-diff-action
       with:
             json_diff_file_output: diff.json
@@ -87,7 +87,7 @@ jobs:
   notify-twitter:
     if: ${{ (github.event.pull_request.merged == true) && (contains(github.event.pull_request.labels.*.name, 'area/ai-model')) }}
     env:
-        MODEL_NAME: hermes-2-theta-llama-3-8b
+        MODEL_NAME: gemma-3-12b-it
     runs-on: ubuntu-latest
     steps:
     - uses: actions/checkout@v4
@@ -99,7 +99,7 @@ jobs:
         docker run -e -ti -d --name local-ai -p 8080:8080 localai/localai:master-ffmpeg-core run --debug $MODEL_NAME
         until [ "`docker inspect -f {{.State.Health.Status}} local-ai`" == "healthy" ]; do echo "Waiting for container to be ready";  docker logs --tail 10 local-ai; sleep 2; done
       # Check the PR diff using the current branch and the base branch of the PR
-    - uses: GrantBirki/git-diff-action@v2.7.0
+    - uses: GrantBirki/git-diff-action@v2.8.0
       id: git-diff-action
       with:
             json_diff_file_output: diff.json

.github/workflows/notify-releases.yaml

@@ -14,7 +14,7 @@ jobs:
     steps:
     - uses: mudler/localai-github-action@v1
       with:
-        model: 'hermes-2-theta-llama-3-8b' # Any from models.localai.io, or from huggingface.com with: "huggingface://<repository>/file"
+        model: 'gemma-3-12b-it' # Any from models.localai.io, or from huggingface.com with: "huggingface://<repository>/file"
     - name: Summarize
       id: summarize
       run: |
@@ -60,4 +60,4 @@ jobs:
         DISCORD_AVATAR: "https://avatars.githubusercontent.com/u/139863280?v=4"
       uses: Ilshidur/action-discord@master
       with:
-        args: ${{ steps.summarize.outputs.message }}
+        args: ${{ steps.summarize.outputs.message }}

.github/workflows/secscan.yaml

@@ -18,7 +18,7 @@ jobs:
         if: ${{ github.actor != 'dependabot[bot]' }}
       - name: Run Gosec Security Scanner
         if: ${{ github.actor != 'dependabot[bot]' }}
-        uses: securego/gosec@v2.22.0
+        uses: securego/gosec@v2.22.3
         with:
           # we let the report trigger content trigger a failure using the GitHub Security features.
           args: '-no-fail -fmt sarif -out results.sarif ./...'

Dockerfile

@@ -15,7 +15,7 @@ ARG TARGETARCH
 ARG TARGETVARIANT
 
 ENV DEBIAN_FRONTEND=noninteractive
-ENV EXTERNAL_GRPC_BACKENDS="coqui:/build/backend/python/coqui/run.sh,transformers:/build/backend/python/transformers/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,faster-whisper:/build/backend/python/faster-whisper/run.sh,kokoro:/build/backend/python/kokoro/run.sh,vllm:/build/backend/python/vllm/run.sh,exllama2:/build/backend/python/exllama2/run.sh"
+ENV EXTERNAL_GRPC_BACKENDS="coqui:/build/backend/python/coqui/run.sh,transformers:/build/backend/python/transformers/run.sh,rerankers:/build/backend/python/rerankers/run.sh,bark:/build/backend/python/bark/run.sh,diffusers:/build/backend/python/diffusers/run.sh,faster-whisper:/build/backend/python/faster-whisper/run.sh,kokoro:/build/backend/python/kokoro/run.sh,vllm:/build/backend/python/vllm/run.sh,exllama2:/build/backend/python/exllama2/run.sh"
 
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
@@ -24,6 +24,7 @@ RUN apt-get update && \
         ca-certificates \
         curl libssl-dev \
         git \
+        git-lfs \
         unzip upx-ucl && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
@@ -430,9 +431,6 @@ RUN if [[ ( "${EXTRA_BACKENDS}" =~ "kokoro" || -z "${EXTRA_BACKENDS}" ) && "$IMA
 RUN if [[ ( "${EXTRA_BACKENDS}" =~ "vllm" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
         make -C backend/python/vllm \
     ; fi && \
-    if [[ ( "${EXTRA_BACKENDS}" =~ "autogptq" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
-        make -C backend/python/autogptq \
-    ; fi && \
     if [[ ( "${EXTRA_BACKENDS}" =~ "bark" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
         make -C backend/python/bark \
     ; fi && \

LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2023-2024 Ettore Di Giacinto (mudler@localai.io)
+Copyright (c) 2023-2025 Ettore Di Giacinto (mudler@localai.io)
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

Makefile

@@ -6,9 +6,7 @@ BINARY_NAME=local-ai
 DETECT_LIBS?=true
 
 # llama.cpp versions
-GOLLAMA_REPO?=https://github.com/go-skynet/go-llama.cpp
-GOLLAMA_VERSION?=2b57a8ae43e4699d3dc5d1496a1ccd42922993be
-CPPLLAMA_VERSION?=6152129d05870cb38162c422c6ba80434e021e9f
+CPPLLAMA_VERSION?=6408210082cc0a61b992b487be7e2ff2efbb9e36
 
 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggerganov/whisper.cpp
@@ -23,8 +21,8 @@ BARKCPP_REPO?=https://github.com/PABannier/bark.cpp.git
 BARKCPP_VERSION?=v1.0.0
 
 # stablediffusion.cpp (ggml)
-STABLEDIFFUSION_GGML_REPO?=https://github.com/leejet/stable-diffusion.cpp
-STABLEDIFFUSION_GGML_VERSION?=5eb15ef4d022bef4a391de4f5f6556e81fbb5024
+STABLEDIFFUSION_GGML_REPO?=https://github.com/richiejp/stable-diffusion.cpp
+STABLEDIFFUSION_GGML_VERSION?=53e3b17eb3d0b5760ced06a1f98320b68b34aaae
 
 ONNX_VERSION?=1.20.0
 ONNX_ARCH?=x64
@@ -151,7 +149,6 @@ ifeq ($(BUILD_TYPE),hipblas)
 	LD_LIBRARY_PATH ?= /opt/rocm/lib:/opt/rocm/llvm/lib
 	export CXX=$(ROCM_HOME)/llvm/bin/clang++
 	export CC=$(ROCM_HOME)/llvm/bin/clang
-	# llama-ggml has no hipblas support, so override it here.
 	export STABLE_BUILD_TYPE=
 	export GGML_HIP=1
 	GPU_TARGETS ?= gfx900,gfx906,gfx908,gfx940,gfx941,gfx942,gfx90a,gfx1030,gfx1031,gfx1100,gfx1101
@@ -188,7 +185,6 @@ ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-avx
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-avx2
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-avx512
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-fallback
-ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-ggml
 ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp-grpc
 ALL_GRPC_BACKENDS+=backend-assets/util/llama-cpp-rpc-server
 ALL_GRPC_BACKENDS+=backend-assets/grpc/whisper
@@ -222,19 +218,6 @@ endif
 
 all: help
 
-## go-llama.cpp
-sources/go-llama.cpp:
-	mkdir -p sources/go-llama.cpp
-	cd sources/go-llama.cpp && \
-	git init && \
-	git remote add origin $(GOLLAMA_REPO) && \
-	git fetch origin && \
-	git checkout $(GOLLAMA_VERSION) && \
-	git submodule update --init --recursive --depth 1 --single-branch
-
-sources/go-llama.cpp/libbinding.a: sources/go-llama.cpp
-	$(MAKE) -C sources/go-llama.cpp BUILD_TYPE=$(STABLE_BUILD_TYPE) libbinding.a
-
 ## bark.cpp
 sources/bark.cpp:
 	git clone --recursive $(BARKCPP_REPO) sources/bark.cpp && \
@@ -277,11 +260,7 @@ backend/go/image/stablediffusion-ggml/libsd.a: sources/stablediffusion-ggml.cpp
 	$(MAKE) -C backend/go/image/stablediffusion-ggml libsd.a
 
 backend-assets/grpc/stablediffusion-ggml: backend/go/image/stablediffusion-ggml/libsd.a backend-assets/grpc
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/backend/go/image/stablediffusion-ggml/ LIBRARY_PATH=$(CURDIR)/backend/go/image/stablediffusion-ggml/ \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/stablediffusion-ggml ./backend/go/image/stablediffusion-ggml/
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/stablediffusion-ggml
-endif
+	$(MAKE) -C backend/go/image/stablediffusion-ggml CGO_LDFLAGS="$(CGO_LDFLAGS)" stablediffusion-ggml
 
 sources/onnxruntime:
 	mkdir -p sources/onnxruntime
@@ -310,19 +289,17 @@ sources/whisper.cpp:
 sources/whisper.cpp/libwhisper.a: sources/whisper.cpp
 	cd sources/whisper.cpp && $(MAKE) libwhisper.a libggml.a
 
-get-sources: sources/go-llama.cpp sources/go-piper sources/stablediffusion-ggml.cpp sources/bark.cpp sources/whisper.cpp backend/cpp/llama/llama.cpp
+get-sources: sources/go-piper sources/stablediffusion-ggml.cpp sources/bark.cpp sources/whisper.cpp backend/cpp/llama/llama.cpp
 
 replace:
 	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp=$(CURDIR)/sources/whisper.cpp
 	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp/bindings/go=$(CURDIR)/sources/whisper.cpp/bindings/go
 	$(GOCMD) mod edit -replace github.com/mudler/go-piper=$(CURDIR)/sources/go-piper
-	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(CURDIR)/sources/go-llama.cpp
 
 dropreplace:
 	$(GOCMD) mod edit -dropreplace github.com/ggerganov/whisper.cpp
 	$(GOCMD) mod edit -dropreplace github.com/ggerganov/whisper.cpp/bindings/go
 	$(GOCMD) mod edit -dropreplace github.com/mudler/go-piper
-	$(GOCMD) mod edit -dropreplace github.com/go-skynet/go-llama.cpp
 
 prepare-sources: get-sources replace
 	$(GOCMD) mod download
@@ -330,7 +307,6 @@ prepare-sources: get-sources replace
 ## GENERIC
 rebuild: ## Rebuilds the project
 	$(GOCMD) clean -cache
-	$(MAKE) -C sources/go-llama.cpp clean
 	$(MAKE) -C sources/whisper.cpp clean
 	$(MAKE) -C sources/go-piper clean
 	$(MAKE) build
@@ -434,7 +410,7 @@ run: prepare ## run local-ai
 test-models/testmodel.ggml:
 	mkdir test-models
 	mkdir test-dir
-	wget -q https://huggingface.co/TheBloke/orca_mini_3B-GGML/resolve/main/orca-mini-3b.ggmlv3.q4_0.bin -O test-models/testmodel.ggml
+	wget -q https://huggingface.co/RichardErkhov/Qwen_-_Qwen2-1.5B-Instruct-gguf/resolve/main/Qwen2-1.5B-Instruct.Q2_K.gguf -O test-models/testmodel.ggml
 	wget -q https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin -O test-models/whisper-en
 	wget -q https://huggingface.co/mudler/all-MiniLM-L6-v2/resolve/main/ggml-model-q4_0.bin -O test-models/bert
 	wget -q https://cdn.openai.com/whisper/draft-20220913a/micro-machines.wav -O test-dir/audio.wav
@@ -449,8 +425,7 @@ test: prepare test-models/testmodel.ggml grpcs
 	export GO_TAGS="tts debug"
 	$(MAKE) prepare-test
 	HUGGINGFACE_GRPC=$(abspath ./)/backend/python/transformers/run.sh 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 && !llama-gguf"  --flake-attempts $(TEST_FLAKES) --fail-fast -v -r $(TEST_PATHS)
-	$(MAKE) test-llama
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="!llama-gguf"  --flake-attempts $(TEST_FLAKES) --fail-fast -v -r $(TEST_PATHS)
 	$(MAKE) test-llama-gguf
 	$(MAKE) test-tts
 	$(MAKE) test-stablediffusion
@@ -479,10 +454,6 @@ teardown-e2e:
 	rm -rf $(TEST_DIR) || true
 	docker stop $$(docker ps -q --filter ancestor=localai-tests)
 
-test-llama: prepare-test
-	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
-
 test-llama-gguf: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
 	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama-gguf" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
@@ -534,18 +505,10 @@ protogen-go-clean:
 	$(RM) bin/*
 
 .PHONY: protogen-python
-protogen-python: autogptq-protogen bark-protogen coqui-protogen diffusers-protogen exllama2-protogen rerankers-protogen transformers-protogen kokoro-protogen vllm-protogen faster-whisper-protogen
+protogen-python: bark-protogen coqui-protogen diffusers-protogen exllama2-protogen rerankers-protogen transformers-protogen kokoro-protogen vllm-protogen faster-whisper-protogen
 
 .PHONY: protogen-python-clean
-protogen-python-clean: autogptq-protogen-clean bark-protogen-clean coqui-protogen-clean diffusers-protogen-clean  exllama2-protogen-clean rerankers-protogen-clean transformers-protogen-clean kokoro-protogen-clean vllm-protogen-clean faster-whisper-protogen-clean
-
-.PHONY: autogptq-protogen
-autogptq-protogen:
-	$(MAKE) -C backend/python/autogptq protogen
-
-.PHONY: autogptq-protogen-clean
-autogptq-protogen-clean:
-	$(MAKE) -C backend/python/autogptq protogen-clean
+protogen-python-clean: bark-protogen-clean coqui-protogen-clean diffusers-protogen-clean  exllama2-protogen-clean rerankers-protogen-clean transformers-protogen-clean kokoro-protogen-clean vllm-protogen-clean faster-whisper-protogen-clean
 
 .PHONY: bark-protogen
 bark-protogen:
@@ -622,7 +585,6 @@ vllm-protogen-clean:
 ## GRPC
 # Note: it is duplicated in the Dockerfile
 prepare-extra-conda-environments: protogen-python
-	$(MAKE) -C backend/python/autogptq
 	$(MAKE) -C backend/python/bark
 	$(MAKE) -C backend/python/coqui
 	$(MAKE) -C backend/python/diffusers
@@ -760,13 +722,6 @@ backend-assets/util/llama-cpp-rpc-server: backend-assets/grpc/llama-cpp-grpc
 	mkdir -p backend-assets/util/
 	cp -rf backend/cpp/llama-grpc/llama.cpp/build/bin/rpc-server backend-assets/util/llama-cpp-rpc-server
 
-backend-assets/grpc/llama-ggml: sources/go-llama.cpp sources/go-llama.cpp/libbinding.a backend-assets/grpc
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-llama.cpp LIBRARY_PATH=$(CURDIR)/sources/go-llama.cpp \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/llama-ggml ./backend/go/llm/llama-ggml/
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/llama-ggml
-endif
-
 backend-assets/grpc/bark-cpp: backend/go/bark/libbark.a backend-assets/grpc
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/backend/go/bark/ LIBRARY_PATH=$(CURDIR)/backend/go/bark/ \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/bark-cpp ./backend/go/bark/
@@ -841,7 +796,8 @@ docker-aio-all:
 
 docker-image-intel:
 	docker build \
-		--build-arg BASE_IMAGE=intel/oneapi-basekit:2025.0.0-0-devel-ubuntu22.04 \
+		--progress plain \
+		--build-arg BASE_IMAGE=intel/oneapi-basekit:2025.1.0-0-devel-ubuntu24.04 \
 		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
 		--build-arg GO_TAGS="none" \
 		--build-arg MAKEFLAGS="$(DOCKER_MAKEFLAGS)" \
@@ -849,7 +805,7 @@ docker-image-intel:
 
 docker-image-intel-xpu:
 	docker build \
-		--build-arg BASE_IMAGE=intel/oneapi-basekit:2025.0.0-0-devel-ubuntu22.04 \
+		--build-arg BASE_IMAGE=intel/oneapi-basekit:2025.1.0-0-devel-ubuntu22.04 \
 		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
 		--build-arg GO_TAGS="none" \
 		--build-arg MAKEFLAGS="$(DOCKER_MAKEFLAGS)" \
@@ -861,7 +817,7 @@ swagger:
 
 .PHONY: gen-assets
 gen-assets:
-	$(GOCMD) run core/dependencies_manager/manager.go embedded/webui_static.yaml core/http/static/assets
+	$(GOCMD) run core/dependencies_manager/manager.go webui_static.yaml core/http/static/assets
 
 ## Documentation
 docs/layouts/_default:

README.md

@@ -1,7 +1,6 @@
 <h1 align="center">
   <br>
-  <img height="300" src="https://github.com/go-skynet/LocalAI/assets/2420543/0966aa2a-166e-4f99-a3e5-6c915fc997dd"> <br>
-    LocalAI
+  <img height="300" src="./core/http/static/logo.png"> <br>
 <br>
 </h1>
 
@@ -48,9 +47,58 @@
 
 [![tests](https://github.com/go-skynet/LocalAI/actions/workflows/test.yml/badge.svg)](https://github.com/go-skynet/LocalAI/actions/workflows/test.yml)[![Build and Release](https://github.com/go-skynet/LocalAI/actions/workflows/release.yaml/badge.svg)](https://github.com/go-skynet/LocalAI/actions/workflows/release.yaml)[![build container images](https://github.com/go-skynet/LocalAI/actions/workflows/image.yml/badge.svg)](https://github.com/go-skynet/LocalAI/actions/workflows/image.yml)[![Bump dependencies](https://github.com/go-skynet/LocalAI/actions/workflows/bump_deps.yaml/badge.svg)](https://github.com/go-skynet/LocalAI/actions/workflows/bump_deps.yaml)[![Artifact Hub](https://img.shields.io/endpoint?url=https://artifacthub.io/badge/repository/localai)](https://artifacthub.io/packages/search?repo=localai)
 
-**LocalAI** is the free, Open Source OpenAI alternative. LocalAI act as a drop-in replacement REST API that’s compatible with OpenAI (Elevenlabs, Anthropic... ) API specifications for local AI inferencing. It allows you to run LLMs, generate images, audio (and not only) locally or on-prem with consumer grade hardware, supporting multiple model families. Does not require GPU. It is created and maintained by [Ettore Di Giacinto](https://github.com/mudler).
+**LocalAI** is the free, Open Source OpenAI alternative. LocalAI act as a drop-in replacement REST API that's compatible with OpenAI (Elevenlabs, Anthropic... ) API specifications for local AI inferencing. It allows you to run LLMs, generate images, audio (and not only) locally or on-prem with consumer grade hardware, supporting multiple model families. Does not require GPU. It is created and maintained by [Ettore Di Giacinto](https://github.com/mudler).
 
-![screen](https://github.com/mudler/LocalAI/assets/2420543/20b5ccd2-8393-44f0-aaf6-87a23806381e)
+
+## 📚🆕 Local Stack Family
+
+🆕 LocalAI is now part of a comprehensive suite of AI tools designed to work together:
+
+<table>
+  <tr>
+    <td width="50%" valign="top">
+      <a href="https://github.com/mudler/LocalAGI">
+        <img src="https://raw.githubusercontent.com/mudler/LocalAGI/refs/heads/main/webui/react-ui/public/logo_2.png" width="300" alt="LocalAGI Logo">
+      </a>
+    </td>
+    <td width="50%" valign="top">
+      <h3><a href="https://github.com/mudler/LocalAGI">LocalAGI</a></h3>
+      <p>A powerful Local AI agent management platform that serves as a drop-in replacement for OpenAI's Responses API, enhanced with advanced agentic capabilities.</p>
+    </td>
+  </tr>
+  <tr>
+    <td width="50%" valign="top">
+      <a href="https://github.com/mudler/LocalRecall">
+        <img src="https://raw.githubusercontent.com/mudler/LocalRecall/refs/heads/main/static/localrecall_horizontal.png" width="300" alt="LocalRecall Logo">
+      </a>
+    </td>
+    <td width="50%" valign="top">
+      <h3><a href="https://github.com/mudler/LocalRecall">LocalRecall</a></h3>
+      <p>A REST-ful API and knowledge base management system that provides persistent memory and storage capabilities for AI agents.</p>
+    </td>
+  </tr>
+</table>
+
+## Screenshots
+
+
+| Talk Interface | Generate Audio |
+| --- | --- |
+| ![Screenshot 2025-03-31 at 12-01-36 LocalAI - Talk](./docs/assets/images/screenshots/screenshot_tts.png) | ![Screenshot 2025-03-31 at 12-01-29 LocalAI - Generate audio with voice-en-us-ryan-low](./docs/assets/images/screenshots/screenshot_tts.png) |
+
+| Models Overview | Generate Images |
+| --- | --- |
+| ![Screenshot 2025-03-31 at 12-01-20 LocalAI - Models](./docs/assets/images/screenshots/screenshot_gallery.png) | ![Screenshot 2025-03-31 at 12-31-41 LocalAI - Generate images with flux 1-dev](./docs/assets/images/screenshots/screenshot_image.png) |
+
+| Chat Interface | Home |
+| --- | --- |
+| ![Screenshot 2025-03-31 at 11-57-44 LocalAI - Chat with localai-functioncall-qwen2 5-7b-v0 5](./docs/assets/images/screenshots/screenshot_chat.png) | ![Screenshot 2025-03-31 at 11-57-23 LocalAI API - c2a39e3 (c2a39e3639227cfd94ffffe9f5691239acc275a8)](./docs/assets/images/screenshots/screenshot_home.png) |
+
+| Login | Swarm |
+| --- | --- |
+|![Screenshot 2025-03-31 at 12-09-59 ](./docs/assets/images/screenshots/screenshot_login.png) | ![Screenshot 2025-03-31 at 12-10-39 LocalAI - P2P dashboard](./docs/assets/images/screenshots/screenshot_p2p.png) |
+
+## 💻 Quickstart
 
 Run the installer script:
 
@@ -59,17 +107,21 @@ curl https://localai.io/install.sh | sh
 ```
 
 Or run with docker:
+
+### CPU only image:
 ```bash
-# CPU only image:
 docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-cpu
-
-# Nvidia GPU:
+```
+### Nvidia GPU:
+```bash
 docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-gpu-nvidia-cuda-12
-
-# CPU and GPU image (bigger size):
+```
+### CPU and GPU image (bigger size):
+```bash
 docker run -ti --name local-ai -p 8080:8080 localai/localai:latest
-
-# AIO images (it will pre-download a set of models ready for use, see https://localai.io/basics/container/)
+```
+### AIO images (it will pre-download a set of models ready for use, see https://localai.io/basics/container/)
+```bash
 docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-aio-cpu
 ```
 
@@ -88,10 +140,13 @@ local-ai run https://gist.githubusercontent.com/.../phi-2.yaml
 local-ai run oci://localai/phi-2:latest
 ```
 
-[💻 Getting started](https://localai.io/basics/getting_started/index.html)
+For more information, see [💻 Getting started](https://localai.io/basics/getting_started/index.html)
 
 ## 📰 Latest project news
 
+- Apr 2025: [LocalAGI](https://github.com/mudler/LocalAGI) and [LocalRecall](https://github.com/mudler/LocalRecall) join the LocalAI family stack.
+- Apr 2025: WebUI overhaul, AIO images updates
+- Feb 2025: Backend cleanup, Breaking changes, new backends (kokoro, OutelTTS, faster-whisper), Nvidia L4T images
 - Jan 2025: LocalAI model release: https://huggingface.co/mudler/LocalAI-functioncall-phi-4-v0.3, SANA support in diffusers: https://github.com/mudler/LocalAI/pull/4603
 - Dec 2024: stablediffusion.cpp backend (ggml) added ( https://github.com/mudler/LocalAI/pull/4289 )
 - Nov 2024: Bark.cpp backend added ( https://github.com/mudler/LocalAI/pull/4287 )
@@ -105,19 +160,6 @@ local-ai run oci://localai/phi-2:latest
 
 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
-- WebUI improvements: https://github.com/mudler/LocalAI/issues/2156
-- Backends v2: https://github.com/mudler/LocalAI/issues/1126
-- Improving UX v2: https://github.com/mudler/LocalAI/issues/1373
-- Assistant API: https://github.com/mudler/LocalAI/issues/1273
-- Vulkan: https://github.com/mudler/LocalAI/issues/1647
-- Anthropic API: https://github.com/mudler/LocalAI/issues/1808
-
-If you want to help and contribute, issues up for grabs: https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3A%22up+for+grabs%22
-
 ## 🚀 [Features](https://localai.io/features/)
 
 - 📖 [Text generation with GPTs](https://localai.io/features/text-generation/) (`llama.cpp`, `transformers`, `vllm` ... [:book: and more](https://localai.io/model-compatibility/index.html#model-compatibility-table))
@@ -131,12 +173,10 @@ If you want to help and contribute, issues up for grabs: https://github.com/mudl
 - 🥽 [Vision API](https://localai.io/features/gpt-vision/)
 - 📈 [Reranker API](https://localai.io/features/reranker/)
 - 🆕🖧 [P2P Inferencing](https://localai.io/features/distribute/)
+- [Agentic capabilities](https://github.com/mudler/LocalAGI)
 - 🔊 Voice activity detection (Silero-VAD support)
 - 🌍 Integrated WebUI!
 
-## 💻 Usage
-
-Check out the [Getting started](https://localai.io/basics/getting_started/index.html) section in our documentation.
 
 ### 🔗 Community and integrations
 
@@ -212,7 +252,7 @@ A huge thank you to our generous sponsors who support this project covering CI e
 
 <p align="center">
   <a href="https://www.spectrocloud.com/" target="blank">
-    <img height="200" src="https://github.com/go-skynet/LocalAI/assets/2420543/68a6f3cb-8a65-4a4d-99b5-6417a8905512">
+    <img height="200" src="https://github.com/user-attachments/assets/72eab1dd-8b93-4fc0-9ade-84db49f24962">
   </a>
   <a href="https://www.premai.io/" target="blank">
     <img height="200" src="https://github.com/mudler/LocalAI/assets/2420543/42e4ca83-661e-4f79-8e46-ae43689683d6"> <br>

aio/cpu/embeddings.yaml

@@ -1,7 +1,7 @@
-name: text-embedding-ada-002
 embeddings: true
+name: text-embedding-ada-002
 parameters:
-  model: huggingface://hugging-quants/Llama-3.2-1B-Instruct-Q4_K_M-GGUF/llama-3.2-1b-instruct-q4_k_m.gguf
+  model: huggingface://bartowski/granite-embedding-107m-multilingual-GGUF/granite-embedding-107m-multilingual-f16.gguf
 
 usage: |
     You can test this model with curl like this:

aio/cpu/text-to-text.yaml

@@ -1,101 +1,57 @@
-name: gpt-4
-mmap: true
-parameters:
-  model: huggingface://NousResearch/Hermes-2-Pro-Llama-3-8B-GGUF/Hermes-2-Pro-Llama-3-8B-Q4_K_M.gguf
 context_size: 8192
-
-stopwords:
-- "<|im_end|>"
-- "<dummy32000>"
-- "</tool_call>"
-- "<|eot_id|>"
-- "<|end_of_text|>"
-
+f16: true
 function:
-  # disable injecting the "answer" tool
-  disable_no_action: true
-
   grammar:
-    # This allows the grammar to also return messages
-    mixed_mode: true
-    # Suffix to add to the grammar
-    #prefix: '<tool_call>\n'
-    # Force parallel calls in the grammar
-    # parallel_calls: true
-
-  return_name_in_function_response: true
-  # Without grammar uncomment the lines below
-  # Warning: this is relying only on the capability of the
-  # LLM model to generate the correct function call.
-  json_regex_match: 
-   - "(?s)<tool_call>(.*?)</tool_call>"
-   - "(?s)<tool_call>(.*?)"
-  replace_llm_results:
-  # Drop the scratchpad content from responses
-  - key: "(?s)<scratchpad>.*</scratchpad>"
-    value: ""
-  replace_function_results: 
-  # Replace everything that is not JSON array or object
-  # 
-  - key: '(?s)^[^{\[]*'
-    value: ""
-  - key: '(?s)[^}\]]*$'
-    value: ""
-  - key: "'([^']*?)'"
-    value: "_DQUOTE_${1}_DQUOTE_"
-  - key: '\\"'
-    value: "__TEMP_QUOTE__"
-  - key: "\'"
-    value: "'"
-  - key: "_DQUOTE_"
-    value: '"'
-  - key: "__TEMP_QUOTE__"
-    value: '"'
-  # Drop the scratchpad content from responses
-  - key: "(?s)<scratchpad>.*</scratchpad>"
-    value: ""
-
+    no_mixed_free_string: true
+    schema_type: llama3.1 # or JSON is supported too (json)
+  response_regex:
+  - <function=(?P<name>\w+)>(?P<arguments>.*)</function>
+mmap: true
+name: gpt-4
+parameters:
+  model: Hermes-3-Llama-3.2-3B-Q4_K_M.gguf
+stopwords:
+- <|im_end|>
+- <dummy32000>
+- <|eot_id|>
+- <|end_of_text|>
 template:
   chat: |
-    {{.Input -}}
-    <|im_start|>assistant
+    <|begin_of_text|><|start_header_id|>system<|end_header_id|>
+    You are a helpful assistant<|eot_id|><|start_header_id|>user<|end_header_id|>
+    {{.Input }}
+    <|start_header_id|>assistant<|end_header_id|>
   chat_message: |
-    <|im_start|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "tool"}}tool{{else if eq .RoleName "user"}}user{{end}}
-    {{- if .FunctionCall }}
-    <tool_call>
-    {{- else if eq .RoleName "tool" }}
-    <tool_response>
-    {{- end }}
-    {{- if .Content}}
-    {{.Content }}
-    {{- end }}
-    {{- if .FunctionCall}}
-    {{toJson .FunctionCall}}
-    {{- end }}
-    {{- if .FunctionCall }}
-    </tool_call>
-    {{- else if eq .RoleName "tool" }}
-    </tool_response>
-    {{- end }}<|im_end|>
+    <|start_header_id|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "tool"}}tool{{else if eq .RoleName "user"}}user{{end}}<|end_header_id|>
+    {{ if .FunctionCall -}}
+    {{ else if eq .RoleName "tool" -}}
+    The Function was executed and the response was:
+    {{ end -}}
+    {{ if .Content -}}
+    {{.Content -}}
+    {{ else if .FunctionCall -}}
+    {{ range .FunctionCall }}
+    [{{.FunctionCall.Name}}({{.FunctionCall.Arguments}})]
+    {{ end }}
+    {{ end -}}
+    <|eot_id|>
   completion: |
     {{.Input}}
-  function: |-
-    <|im_start|>system
-    You are a function calling AI model.
-    Here are the available tools:
-    <tools>
-    {{range .Functions}}
-    {'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
-    {{end}}
-    </tools>
-    You should call the tools provided to you sequentially
-    Please use <scratchpad> XML tags to record your reasoning and planning before you call the functions as follows:
-    <scratchpad>
-    {step-by-step reasoning and plan in bullet points}
-    </scratchpad>
-    For each function call return a json object with function name and arguments within <tool_call> XML tags as follows:
-    <tool_call>
-    {"arguments": <args-dict>, "name": <function-name>}
-    </tool_call><|im_end|>
-    {{.Input -}}
-    <|im_start|>assistant
+  function: |
+    <|start_header_id|>system<|end_header_id|>
+    You are an expert in composing functions. You are given a question and a set of possible functions.
+    Based on the question, you will need to make one or more function/tool calls to achieve the purpose.
+    If none of the functions can be used, point it out. If the given question lacks the parameters required by the function, also point it out. You should only return the function call in tools call sections.
+    If you decide to invoke any of the function(s), you MUST put it in the format as follows:
+    [func_name1(params_name1=params_value1,params_name2=params_value2,...),func_name2(params_name1=params_value1,params_name2=params_value2,...)]
+    You SHOULD NOT include any other text in the response.
+    Here is a list of functions in JSON format that you can invoke.
+    {{toJson .Functions}}
+    <|eot_id|><|start_header_id|>user<|end_header_id|>
+    {{.Input}}
+    <|eot_id|><|start_header_id|>assistant<|end_header_id|>
+
+download_files:
+- filename: Hermes-3-Llama-3.2-3B-Q4_K_M.gguf
+  sha256: 2e220a14ba4328fee38cf36c2c068261560f999fadb5725ce5c6d977cb5126b5
+  uri: huggingface://bartowski/Hermes-3-Llama-3.2-3B-GGUF/Hermes-3-Llama-3.2-3B-Q4_K_M.gguf

aio/cpu/vad.yaml

@@ -0,0 +1,8 @@
+backend: silero-vad
+name: silero-vad
+parameters:
+  model: silero-vad.onnx
+download_files:
+- filename: silero-vad.onnx
+  uri: https://huggingface.co/onnx-community/silero-vad/resolve/main/onnx/model.onnx
+  sha256: a4a068cd6cf1ea8355b84327595838ca748ec29a25bc91fc82e6c299ccdc5808

aio/cpu/vision.yaml

@@ -1,31 +1,49 @@
-backend: llama-cpp
 context_size: 4096
 f16: true
 mmap: true
+mmproj: minicpm-v-2_6-mmproj-f16.gguf
 name: gpt-4o
-
-roles:
-  user: "USER:"
-  assistant: "ASSISTANT:"
-  system: "SYSTEM:"
-
-mmproj: bakllava-mmproj.gguf
 parameters:
-  model: bakllava.gguf
-
+  model: minicpm-v-2_6-Q4_K_M.gguf
+stopwords:
+- <|im_end|>
+- <dummy32000>
+- </s>
+- <|endoftext|>
 template:
   chat: |
-    A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human's questions.
+    {{.Input -}}
+    <|im_start|>assistant
+  chat_message: |
+    <|im_start|>{{ .RoleName }}
+    {{ if .FunctionCall -}}
+    Function call:
+    {{ else if eq .RoleName "tool" -}}
+    Function response:
+    {{ end -}}
+    {{ if .Content -}}
+    {{.Content }}
+    {{ end -}}
+    {{ if .FunctionCall -}}
+    {{toJson .FunctionCall}}
+    {{ end -}}<|im_end|>
+  completion: |
     {{.Input}}
-    ASSISTANT:
+  function: |
+    <|im_start|>system
+    You are a function calling AI model. You are provided with functions to execute. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions. Here are the available tools:
+    {{range .Functions}}
+    {'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
+    {{end}}
+    For each function call return a json object with function name and arguments
+    <|im_end|>
+    {{.Input -}}
+    <|im_start|>assistant
 
 download_files:
-- filename: bakllava.gguf
-  uri: huggingface://mys/ggml_bakllava-1/ggml-model-q4_k.gguf
-- filename: bakllava-mmproj.gguf
-  uri: huggingface://mys/ggml_bakllava-1/mmproj-model-f16.gguf
-
-usage: |
-    curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
-        "model": "gpt-4-vision-preview",
-        "messages": [{"role": "user", "content": [{"type":"text", "text": "What is in the image?"}, {"type": "image_url", "image_url": {"url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg" }}], "temperature": 0.9}]}'
+- filename: minicpm-v-2_6-Q4_K_M.gguf
+  sha256: 3a4078d53b46f22989adbf998ce5a3fd090b6541f112d7e936eb4204a04100b1
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/ggml-model-Q4_K_M.gguf
+- filename: minicpm-v-2_6-mmproj-f16.gguf
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/mmproj-model-f16.gguf
+  sha256: 4485f68a0f1aa404c391e788ea88ea653c100d8e98fe572698f701e5809711fd

aio/entrypoint.sh

@@ -129,7 +129,7 @@ detect_gpu
 detect_gpu_size
 
 PROFILE="${PROFILE:-$GPU_SIZE}" # default to cpu
-export MODELS="${MODELS:-/aio/${PROFILE}/embeddings.yaml,/aio/${PROFILE}/rerank.yaml,/aio/${PROFILE}/text-to-speech.yaml,/aio/${PROFILE}/image-gen.yaml,/aio/${PROFILE}/text-to-text.yaml,/aio/${PROFILE}/speech-to-text.yaml,/aio/${PROFILE}/vision.yaml}"
+export MODELS="${MODELS:-/aio/${PROFILE}/embeddings.yaml,/aio/${PROFILE}/rerank.yaml,/aio/${PROFILE}/text-to-speech.yaml,/aio/${PROFILE}/image-gen.yaml,/aio/${PROFILE}/text-to-text.yaml,/aio/${PROFILE}/speech-to-text.yaml,/aio/${PROFILE}/vad.yaml,/aio/${PROFILE}/vision.yaml}"
 
 check_vars
 

aio/gpu-8g/embeddings.yaml

@@ -1,7 +1,7 @@
+embeddings: true
 name: text-embedding-ada-002
-backend: sentencetransformers
 parameters:
-  model: all-MiniLM-L6-v2
+  model: huggingface://bartowski/granite-embedding-107m-multilingual-GGUF/granite-embedding-107m-multilingual-f16.gguf
 
 usage: |
     You can test this model with curl like this:

aio/gpu-8g/text-to-text.yaml

@@ -1,101 +1,53 @@
-name: gpt-4
-mmap: true
-parameters:
-  model: huggingface://NousResearch/Hermes-2-Pro-Llama-3-8B-GGUF/Hermes-2-Pro-Llama-3-8B-Q4_K_M.gguf
-context_size: 8192
-
-stopwords:
-- "<|im_end|>"
-- "<dummy32000>"
-- "</tool_call>"
-- "<|eot_id|>"
-- "<|end_of_text|>"
-
+context_size: 4096
+f16: true
 function:
-  # disable injecting the "answer" tool
-  disable_no_action: true
-
+  capture_llm_results:
+  - (?s)<Thought>(.*?)</Thought>
   grammar:
-    # This allows the grammar to also return messages
-    mixed_mode: true
-    # Suffix to add to the grammar
-    #prefix: '<tool_call>\n'
-    # Force parallel calls in the grammar
-    # parallel_calls: true
-
-  return_name_in_function_response: true
-  # Without grammar uncomment the lines below
-  # Warning: this is relying only on the capability of the
-  # LLM model to generate the correct function call.
-  json_regex_match: 
-   - "(?s)<tool_call>(.*?)</tool_call>"
-   - "(?s)<tool_call>(.*?)"
+    properties_order: name,arguments
+  json_regex_match:
+  - (?s)<Output>(.*?)</Output>
   replace_llm_results:
-  # Drop the scratchpad content from responses
-  - key: "(?s)<scratchpad>.*</scratchpad>"
+  - key: (?s)<Thought>(.*?)</Thought>
     value: ""
-  replace_function_results: 
-  # Replace everything that is not JSON array or object
-  # 
-  - key: '(?s)^[^{\[]*'
-    value: ""
-  - key: '(?s)[^}\]]*$'
-    value: ""
-  - key: "'([^']*?)'"
-    value: "_DQUOTE_${1}_DQUOTE_"
-  - key: '\\"'
-    value: "__TEMP_QUOTE__"
-  - key: "\'"
-    value: "'"
-  - key: "_DQUOTE_"
-    value: '"'
-  - key: "__TEMP_QUOTE__"
-    value: '"'
-  # Drop the scratchpad content from responses
-  - key: "(?s)<scratchpad>.*</scratchpad>"
-    value: ""
-
+mmap: true
+name: gpt-4
+parameters:
+  model: localai-functioncall-qwen2.5-7b-v0.5-q4_k_m.gguf
+stopwords:
+- <|im_end|>
+- <dummy32000>
+- </s>
 template:
   chat: |
     {{.Input -}}
     <|im_start|>assistant
   chat_message: |
-    <|im_start|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "tool"}}tool{{else if eq .RoleName "user"}}user{{end}}
-    {{- if .FunctionCall }}
-    <tool_call>
-    {{- else if eq .RoleName "tool" }}
-    <tool_response>
-    {{- end }}
-    {{- if .Content}}
+    <|im_start|>{{ .RoleName }}
+    {{ if .FunctionCall -}}
+    Function call:
+    {{ else if eq .RoleName "tool" -}}
+    Function response:
+    {{ end -}}
+    {{ if .Content -}}
     {{.Content }}
-    {{- end }}
-    {{- if .FunctionCall}}
+    {{ end -}}
+    {{ if .FunctionCall -}}
     {{toJson .FunctionCall}}
-    {{- end }}
-    {{- if .FunctionCall }}
-    </tool_call>
-    {{- else if eq .RoleName "tool" }}
-    </tool_response>
-    {{- end }}<|im_end|>
+    {{ end -}}<|im_end|>
   completion: |
     {{.Input}}
-  function: |-
+  function: |
     <|im_start|>system
-    You are a function calling AI model.
-    Here are the available tools:
-    <tools>
+    You are an AI assistant that executes function calls, and these are the tools at your disposal:
     {{range .Functions}}
     {'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
     {{end}}
-    </tools>
-    You should call the tools provided to you sequentially
-    Please use <scratchpad> XML tags to record your reasoning and planning before you call the functions as follows:
-    <scratchpad>
-    {step-by-step reasoning and plan in bullet points}
-    </scratchpad>
-    For each function call return a json object with function name and arguments within <tool_call> XML tags as follows:
-    <tool_call>
-    {"arguments": <args-dict>, "name": <function-name>}
-    </tool_call><|im_end|>
+    <|im_end|>
     {{.Input -}}
-    <|im_start|>assistant
+    <|im_start|>assistant
+
+download_files:
+- filename: localai-functioncall-phi-4-v0.3-q4_k_m.gguf
+  sha256: 23fee048ded2a6e2e1a7b6bbefa6cbf83068f194caa9552aecbaa00fec8a16d5
+  uri: huggingface://mudler/LocalAI-functioncall-phi-4-v0.3-Q4_K_M-GGUF/localai-functioncall-phi-4-v0.3-q4_k_m.gguf

aio/gpu-8g/vad.yaml

@@ -0,0 +1,8 @@
+backend: silero-vad
+name: silero-vad
+parameters:
+  model: silero-vad.onnx
+download_files:
+- filename: silero-vad.onnx
+  uri: https://huggingface.co/onnx-community/silero-vad/resolve/main/onnx/model.onnx
+  sha256: a4a068cd6cf1ea8355b84327595838ca748ec29a25bc91fc82e6c299ccdc5808

aio/gpu-8g/vision.yaml

@@ -1,35 +1,49 @@
-backend: llama-cpp
 context_size: 4096
 f16: true
 mmap: true
+mmproj: minicpm-v-2_6-mmproj-f16.gguf
 name: gpt-4o
-
-roles:
-  user: "USER:"
-  assistant: "ASSISTANT:"
-  system: "SYSTEM:"
-
-mmproj: llava-v1.6-7b-mmproj-f16.gguf
 parameters:
-  model: llava-v1.6-mistral-7b.Q5_K_M.gguf
-  temperature: 0.2
-  top_k: 40
-  top_p: 0.95
-  seed: -1
-
+  model: minicpm-v-2_6-Q4_K_M.gguf
+stopwords:
+- <|im_end|>
+- <dummy32000>
+- </s>
+- <|endoftext|>
 template:
   chat: |
-    A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human's questions.
+    {{.Input -}}
+    <|im_start|>assistant
+  chat_message: |
+    <|im_start|>{{ .RoleName }}
+    {{ if .FunctionCall -}}
+    Function call:
+    {{ else if eq .RoleName "tool" -}}
+    Function response:
+    {{ end -}}
+    {{ if .Content -}}
+    {{.Content }}
+    {{ end -}}
+    {{ if .FunctionCall -}}
+    {{toJson .FunctionCall}}
+    {{ end -}}<|im_end|>
+  completion: |
     {{.Input}}
-    ASSISTANT:
+  function: |
+    <|im_start|>system
+    You are a function calling AI model. You are provided with functions to execute. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions. Here are the available tools:
+    {{range .Functions}}
+    {'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
+    {{end}}
+    For each function call return a json object with function name and arguments
+    <|im_end|>
+    {{.Input -}}
+    <|im_start|>assistant
 
 download_files:
-- filename: llava-v1.6-mistral-7b.Q5_K_M.gguf
-  uri: huggingface://cjpais/llava-1.6-mistral-7b-gguf/llava-v1.6-mistral-7b.Q5_K_M.gguf
-- filename: llava-v1.6-7b-mmproj-f16.gguf
-  uri: huggingface://cjpais/llava-1.6-mistral-7b-gguf/mmproj-model-f16.gguf
-
-usage: |
-    curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
-        "model": "gpt-4-vision-preview",
-        "messages": [{"role": "user", "content": [{"type":"text", "text": "What is in the image?"}, {"type": "image_url", "image_url": {"url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg" }}], "temperature": 0.9}]}'
+- filename: minicpm-v-2_6-Q4_K_M.gguf
+  sha256: 3a4078d53b46f22989adbf998ce5a3fd090b6541f112d7e936eb4204a04100b1
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/ggml-model-Q4_K_M.gguf
+- filename: minicpm-v-2_6-mmproj-f16.gguf
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/mmproj-model-f16.gguf
+  sha256: 4485f68a0f1aa404c391e788ea88ea653c100d8e98fe572698f701e5809711fd

aio/intel/embeddings.yaml

@@ -1,7 +1,7 @@
+embeddings: true
 name: text-embedding-ada-002
-backend: sentencetransformers
 parameters:
-  model: all-MiniLM-L6-v2
+  model: huggingface://bartowski/granite-embedding-107m-multilingual-GGUF/granite-embedding-107m-multilingual-f16.gguf
 
 usage: |
     You can test this model with curl like this:

aio/intel/text-to-text.yaml

@@ -1,103 +1,53 @@
-name: gpt-4
-mmap: false
-context_size: 8192
-
-f16: false
-parameters:
-  model: huggingface://NousResearch/Hermes-2-Pro-Llama-3-8B-GGUF/Hermes-2-Pro-Llama-3-8B-Q4_K_M.gguf
-
-stopwords:
-- "<|im_end|>"
-- "<dummy32000>"
-- "</tool_call>"
-- "<|eot_id|>"
-- "<|end_of_text|>"
-
+context_size: 4096
+f16: true
 function:
-  # disable injecting the "answer" tool
-  disable_no_action: true
-
+  capture_llm_results:
+  - (?s)<Thought>(.*?)</Thought>
   grammar:
-    # This allows the grammar to also return messages
-    mixed_mode: true
-    # Suffix to add to the grammar
-    #prefix: '<tool_call>\n'
-    # Force parallel calls in the grammar
-    # parallel_calls: true
-
-  return_name_in_function_response: true
-  # Without grammar uncomment the lines below
-  # Warning: this is relying only on the capability of the
-  # LLM model to generate the correct function call.
-  json_regex_match: 
-   - "(?s)<tool_call>(.*?)</tool_call>"
-   - "(?s)<tool_call>(.*?)"
+    properties_order: name,arguments
+  json_regex_match:
+  - (?s)<Output>(.*?)</Output>
   replace_llm_results:
-  # Drop the scratchpad content from responses
-  - key: "(?s)<scratchpad>.*</scratchpad>"
+  - key: (?s)<Thought>(.*?)</Thought>
     value: ""
-  replace_function_results: 
-  # Replace everything that is not JSON array or object
-  # 
-  - key: '(?s)^[^{\[]*'
-    value: ""
-  - key: '(?s)[^}\]]*$'
-    value: ""
-  - key: "'([^']*?)'"
-    value: "_DQUOTE_${1}_DQUOTE_"
-  - key: '\\"'
-    value: "__TEMP_QUOTE__"
-  - key: "\'"
-    value: "'"
-  - key: "_DQUOTE_"
-    value: '"'
-  - key: "__TEMP_QUOTE__"
-    value: '"'
-  # Drop the scratchpad content from responses
-  - key: "(?s)<scratchpad>.*</scratchpad>"
-    value: ""
-
+mmap: true
+name: gpt-4
+parameters:
+  model: localai-functioncall-qwen2.5-7b-v0.5-q4_k_m.gguf
+stopwords:
+- <|im_end|>
+- <dummy32000>
+- </s>
 template:
   chat: |
     {{.Input -}}
     <|im_start|>assistant
   chat_message: |
-    <|im_start|>{{if eq .RoleName "assistant"}}assistant{{else if eq .RoleName "system"}}system{{else if eq .RoleName "tool"}}tool{{else if eq .RoleName "user"}}user{{end}}
-    {{- if .FunctionCall }}
-    <tool_call>
-    {{- else if eq .RoleName "tool" }}
-    <tool_response>
-    {{- end }}
-    {{- if .Content}}
+    <|im_start|>{{ .RoleName }}
+    {{ if .FunctionCall -}}
+    Function call:
+    {{ else if eq .RoleName "tool" -}}
+    Function response:
+    {{ end -}}
+    {{ if .Content -}}
     {{.Content }}
-    {{- end }}
-    {{- if .FunctionCall}}
+    {{ end -}}
+    {{ if .FunctionCall -}}
     {{toJson .FunctionCall}}
-    {{- end }}
-    {{- if .FunctionCall }}
-    </tool_call>
-    {{- else if eq .RoleName "tool" }}
-    </tool_response>
-    {{- end }}<|im_end|>
+    {{ end -}}<|im_end|>
   completion: |
     {{.Input}}
-  function: |-
+  function: |
     <|im_start|>system
-    You are a function calling AI model.
-    Here are the available tools:
-    <tools>
+    You are an AI assistant that executes function calls, and these are the tools at your disposal:
     {{range .Functions}}
     {'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
     {{end}}
-    </tools>
-    You should call the tools provided to you sequentially
-    Please use <scratchpad> XML tags to record your reasoning and planning before you call the functions as follows:
-    <scratchpad>
-    {step-by-step reasoning and plan in bullet points}
-    </scratchpad>
-    For each function call return a json object with function name and arguments within <tool_call> XML tags as follows:
-    <tool_call>
-    {"arguments": <args-dict>, "name": <function-name>}
-    </tool_call><|im_end|>
+    <|im_end|>
     {{.Input -}}
     <|im_start|>assistant
+
+download_files:
+- filename: localai-functioncall-phi-4-v0.3-q4_k_m.gguf
+  sha256: 23fee048ded2a6e2e1a7b6bbefa6cbf83068f194caa9552aecbaa00fec8a16d5
+  uri: huggingface://mudler/LocalAI-functioncall-phi-4-v0.3-Q4_K_M-GGUF/localai-functioncall-phi-4-v0.3-q4_k_m.gguf

aio/intel/vad.yaml

@@ -0,0 +1,8 @@
+backend: silero-vad
+name: silero-vad
+parameters:
+  model: silero-vad.onnx
+download_files:
+- filename: silero-vad.onnx
+  uri: https://huggingface.co/onnx-community/silero-vad/resolve/main/onnx/model.onnx
+  sha256: a4a068cd6cf1ea8355b84327595838ca748ec29a25bc91fc82e6c299ccdc5808

aio/intel/vision.yaml

@@ -1,35 +1,50 @@
-backend: llama-cpp
 context_size: 4096
-mmap: false
-f16: false
+f16: true
+mmap: true
+mmproj: minicpm-v-2_6-mmproj-f16.gguf
 name: gpt-4o
-
-roles:
-  user: "USER:"
-  assistant: "ASSISTANT:"
-  system: "SYSTEM:"
-
-mmproj: llava-v1.6-7b-mmproj-f16.gguf
 parameters:
-  model: llava-v1.6-mistral-7b.Q5_K_M.gguf
-  temperature: 0.2
-  top_k: 40
-  top_p: 0.95
-  seed: -1
-
+  model: minicpm-v-2_6-Q4_K_M.gguf
+stopwords:
+- <|im_end|>
+- <dummy32000>
+- </s>
+- <|endoftext|>
 template:
   chat: |
-    A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human's questions.
+    {{.Input -}}
+    <|im_start|>assistant
+  chat_message: |
+    <|im_start|>{{ .RoleName }}
+    {{ if .FunctionCall -}}
+    Function call:
+    {{ else if eq .RoleName "tool" -}}
+    Function response:
+    {{ end -}}
+    {{ if .Content -}}
+    {{.Content }}
+    {{ end -}}
+    {{ if .FunctionCall -}}
+    {{toJson .FunctionCall}}
+    {{ end -}}<|im_end|>
+  completion: |
     {{.Input}}
-    ASSISTANT:
+  function: |
+    <|im_start|>system
+    You are a function calling AI model. You are provided with functions to execute. You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions. Here are the available tools:
+    {{range .Functions}}
+    {'type': 'function', 'function': {'name': '{{.Name}}', 'description': '{{.Description}}', 'parameters': {{toJson .Parameters}} }}
+    {{end}}
+    For each function call return a json object with function name and arguments
+    <|im_end|>
+    {{.Input -}}
+    <|im_start|>assistant
+
 
 download_files:
-- filename: llava-v1.6-mistral-7b.Q5_K_M.gguf
-  uri: huggingface://cjpais/llava-1.6-mistral-7b-gguf/llava-v1.6-mistral-7b.Q5_K_M.gguf
-- filename: llava-v1.6-7b-mmproj-f16.gguf
-  uri: huggingface://cjpais/llava-1.6-mistral-7b-gguf/mmproj-model-f16.gguf
-
-usage: |
-    curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
-        "model": "gpt-4-vision-preview",
-        "messages": [{"role": "user", "content": [{"type":"text", "text": "What is in the image?"}, {"type": "image_url", "image_url": {"url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg" }}], "temperature": 0.9}]}'
+- filename: minicpm-v-2_6-Q4_K_M.gguf
+  sha256: 3a4078d53b46f22989adbf998ce5a3fd090b6541f112d7e936eb4204a04100b1
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/ggml-model-Q4_K_M.gguf
+- filename: minicpm-v-2_6-mmproj-f16.gguf
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/mmproj-model-f16.gguf
+  sha256: 4485f68a0f1aa404c391e788ea88ea653c100d8e98fe572698f701e5809711fd

backend/backend.proto

@@ -163,6 +163,10 @@ message Reply {
   double timing_token_generation = 5;
 }
 
+message GrammarTrigger {
+  string word = 1;
+}
+
 message ModelOptions {
   string Model = 1;
   int32 ContextSize = 2;
@@ -186,11 +190,7 @@ message ModelOptions {
   int32 NGQA = 20;
   string ModelFile = 21;
 
-  // AutoGPTQ
-  string Device = 22;
-  bool UseTriton = 23;
-  string ModelBaseName = 24;
-  bool UseFastTokenizer = 25;
+
 
   // Diffusers
   string PipelineType = 26;
@@ -224,6 +224,11 @@ message ModelOptions {
   int32  MaxModelLen = 54;
   int32  TensorParallelSize = 55;
   string LoadFormat = 58;
+  bool   DisableLogStatus = 66;
+  string DType = 67;
+  int32  LimitImagePerPrompt = 68;
+  int32  LimitVideoPerPrompt = 69;
+  int32  LimitAudioPerPrompt = 70;
 
   string MMProj = 41;
 
@@ -247,6 +252,8 @@ message ModelOptions {
 
   string CacheTypeKey = 63;
   string CacheTypeValue = 64;
+
+  repeated GrammarTrigger GrammarTriggers = 65;
 }
 
 message Result {

backend/cpp/llama/CMakeLists.txt

@@ -2,7 +2,7 @@
 ## XXX: In some versions of CMake clip wasn't being built before llama.
 ## This is an hack for now, but it should be fixed in the future.
 set(TARGET myclip)
-add_library(${TARGET} clip.cpp clip.h llava.cpp llava.h)
+add_library(${TARGET} clip.cpp clip.h clip-impl.h llava.cpp llava.h)
 install(TARGETS ${TARGET} LIBRARY)
 target_include_directories(myclip PUBLIC .)
 target_include_directories(myclip PUBLIC ../..)

backend/cpp/llama/Makefile

@@ -8,7 +8,7 @@ ONEAPI_VARS?=/opt/intel/oneapi/setvars.sh
 TARGET?=--target grpc-server
 
 # Disable Shared libs as we are linking on static gRPC and we can't mix shared and static
-CMAKE_ARGS+=-DBUILD_SHARED_LIBS=OFF
+CMAKE_ARGS+=-DBUILD_SHARED_LIBS=OFF -DLLAMA_CURL=OFF
 
 # If build type is cublas, then we set -DGGML_CUDA=ON to CMAKE_ARGS automatically
 ifeq ($(BUILD_TYPE),cublas)
@@ -36,11 +36,18 @@ else ifeq ($(OS),Darwin)
 endif
 
 ifeq ($(BUILD_TYPE),sycl_f16)
-	CMAKE_ARGS+=-DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_SYCL_F16=ON
+	CMAKE_ARGS+=-DGGML_SYCL=ON \
+		-DCMAKE_C_COMPILER=icx \
+		-DCMAKE_CXX_COMPILER=icpx \
+		-DCMAKE_CXX_FLAGS="-fsycl" \
+		-DGGML_SYCL_F16=ON
 endif
 
 ifeq ($(BUILD_TYPE),sycl_f32)
-	CMAKE_ARGS+=-DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx
+	CMAKE_ARGS+=-DGGML_SYCL=ON \
+		-DCMAKE_C_COMPILER=icx \
+		-DCMAKE_CXX_COMPILER=icpx \
+		-DCMAKE_CXX_FLAGS="-fsycl"
 endif
 
 llama.cpp:
@@ -77,4 +84,4 @@ ifneq (,$(findstring sycl,$(BUILD_TYPE)))
 else
 	+cd llama.cpp && mkdir -p build && cd build && cmake .. $(CMAKE_ARGS) && cmake --build . --config Release $(TARGET)
 endif
-	cp llama.cpp/build/bin/grpc-server .
+	cp llama.cpp/build/bin/grpc-server .

backend/cpp/llama/grpc-server.cpp

@@ -217,6 +217,7 @@ struct llama_client_slot
 
     bool infill = false;
     bool embedding = false;
+    bool reranker = false;
     bool has_next_token = true;
     bool truncated = false;
     bool stopped_eos = false;
@@ -467,6 +468,10 @@ struct llama_server_context
     bool all_slots_are_idle = false;
     bool add_bos_token      = true;
     bool has_eos_token      = true;
+    bool has_gpu = false;
+
+    bool grammar_lazy = false;
+    std::vector<common_grammar_trigger> grammar_triggers;
 
     int32_t n_ctx;  // total context for all clients / slots
 
@@ -505,12 +510,15 @@ struct llama_server_context
     bool load_model(const common_params &params_)
     {
         params = params_;
-        if (!params.mmproj.empty()) {
+        if (!params.mmproj.path.empty()) {
             multimodal = true;
             LOG_INFO("Multi Modal Mode Enabled", {});
-            clp_ctx = clip_model_load(params.mmproj.c_str(), /*verbosity=*/ 1);
+            clp_ctx = clip_init(params.mmproj.path.c_str(), clip_context_params {
+                /* use_gpu */ has_gpu,
+                /*verbosity=*/ GGML_LOG_LEVEL_INFO,
+            });
             if(clp_ctx == nullptr) {
-                LOG_ERR("unable to load clip model: %s", params.mmproj.c_str());
+                LOG_ERR("unable to load clip model: %s", params.mmproj.path.c_str());
                 return false;
             }
 
@@ -524,10 +532,16 @@ struct llama_server_context
         ctx = common_init.context.release();
         if (model == nullptr)
         {
-            LOG_ERR("unable to load model: %s", params.model.c_str());
+            LOG_ERR("unable to load model: %s", params.model.path.c_str());
             return false;
         }
 
+        // Enable reranking if embeddings are enabled - moved after context initialization
+        if (params.embedding) {
+            params.reranking = true;
+            LOG_INFO("Reranking enabled (embeddings are enabled)", {});
+        }
+
         if (multimodal) {
             const int n_embd_clip = clip_n_mmproj_embd(clp_ctx);
             const int n_embd_llm  = llama_model_n_embd(model);
@@ -706,6 +720,8 @@ struct llama_server_context
         slot->sparams.grammar           = json_value(data, "grammar",           default_sparams.grammar);
         slot->sparams.n_probs           = json_value(data, "n_probs",           default_sparams.n_probs);
         slot->sparams.min_keep          = json_value(data, "min_keep",          default_sparams.min_keep);
+        slot->sparams.grammar_triggers = grammar_triggers;
+        slot->sparams.grammar_lazy = grammar_lazy;
 
         if (slot->n_predict > 0 && slot->params.n_predict > slot->n_predict) {
             // Might be better to reject the request with a 400 ?
@@ -1150,6 +1166,14 @@ struct llama_server_context
             slot.has_next_token = false;
         }
 
+        if (slot.n_past >= slot.n_ctx) {
+            slot.truncated      = true;
+            slot.stopped_limit = true;
+            slot.has_next_token = false;
+
+            LOG_VERBOSE("stopped due to running out of context capacity", {});
+        }
+
         if (result.tok == llama_vocab_eos(vocab) || llama_vocab_is_eog(vocab, result.tok))
         {
             slot.stopped_eos = true;
@@ -1337,7 +1361,7 @@ struct llama_server_context
         queue_results.send(res);
     }
 
-    void send_embedding(llama_client_slot &slot)
+    void send_embedding(llama_client_slot &slot, const llama_batch & batch)
     {
         task_result res;
         res.id = slot.task_id;
@@ -1359,16 +1383,96 @@ struct llama_server_context
         else
         {
             const float *data = llama_get_embeddings(ctx);
-            std::vector<float> embedding(data, data + n_embd);
+            std::vector<float> embd_res(n_embd, 0.0f);
+            std::vector<std::vector<float>> embedding;
+            for (int i = 0; i < batch.n_tokens; ++i) {
+                if (!batch.logits[i] || batch.seq_id[i][0] != slot.id) {
+                    continue;
+                }
+
+                const float * embd = llama_get_embeddings_seq(ctx, batch.seq_id[i][0]);
+                if (embd == NULL) {
+                    embd = llama_get_embeddings_ith(ctx, i);
+                }
+
+                if (embd == NULL) {
+                    LOG("failed to get embeddings");
+
+                    continue;
+                }
+
+                // normalize only when there is pooling
+                // TODO: configurable
+                if (llama_pooling_type(ctx) != LLAMA_POOLING_TYPE_NONE) {
+                    common_embd_normalize(embd, embd_res.data(), n_embd, 2);
+                    embedding.push_back(embd_res);
+                } else {
+                    embedding.push_back({ embd, embd + n_embd });
+                }
+            }
+
+            // OAI compat
             res.result_json = json
             {
-                {"embedding", embedding },
+                {"embedding", embedding[0] },
             };
         }
         queue_results.send(res);
     }
 
-    void request_completion(int task_id, json data, bool infill, bool embedding, int multitask_id)
+    void send_rerank(llama_client_slot &slot, const llama_batch & batch)
+    {
+        task_result res;
+        res.id = slot.task_id;
+        res.multitask_id = slot.multitask_id;
+        res.error = false;
+        res.stop = true;
+
+        float score = -1e6f; // Default score if we fail to get embeddings
+
+        if (!params.reranking)
+        {
+            LOG_WARNING("reranking disabled", {
+                {"params.reranking", params.reranking},
+            });
+        }
+        else if (ctx == nullptr)
+        {
+            LOG_ERR("context is null, cannot perform reranking");
+            res.error = true;
+        }
+        else
+        {
+            for (int i = 0; i < batch.n_tokens; ++i) {
+                if (!batch.logits[i] || batch.seq_id[i][0] != slot.id) {
+                    continue;
+                }
+
+                const float * embd = llama_get_embeddings_seq(ctx, batch.seq_id[i][0]);
+                if (embd == NULL) {
+                    embd = llama_get_embeddings_ith(ctx, i);
+                }
+
+                if (embd == NULL) {
+                    LOG("failed to get embeddings");
+                    continue;
+                }
+
+                score = embd[0];
+            }
+        }
+
+        // Format result as JSON similar to the embedding function
+        res.result_json = json
+        {
+            {"score", score},
+            {"tokens", slot.num_prompt_tokens}
+        };
+        
+        queue_results.send(res);
+    }
+
+    void request_completion(int task_id, json data, bool infill, bool embedding, bool rerank, int multitask_id)
     {
         task_server task;
         task.id = task_id;
@@ -1376,6 +1480,7 @@ struct llama_server_context
         task.data = std::move(data);
         task.infill_mode = infill;
         task.embedding_mode = embedding;
+        task.reranking_mode = rerank;
         task.type = TASK_TYPE_COMPLETION;
         task.multitask_id = multitask_id;
 
@@ -1507,7 +1612,7 @@ struct llama_server_context
             subtask_data["prompt"] = subtask_data["prompt"][i];
 
             // subtasks inherit everything else (infill mode, embedding mode, etc.)
-            request_completion(subtask_ids[i], subtask_data, multiprompt_task.infill_mode, multiprompt_task.embedding_mode, multitask_id);
+            request_completion(subtask_ids[i], subtask_data, multiprompt_task.infill_mode, multiprompt_task.embedding_mode, multiprompt_task.reranking_mode, multitask_id);
         }
     }
 
@@ -1546,6 +1651,7 @@ struct llama_server_context
 
                 slot->infill       = task.infill_mode;
                 slot->embedding    = task.embedding_mode;
+                slot->reranker    = task.reranking_mode;
                 slot->task_id      = task.id;
                 slot->multitask_id = task.multitask_id;
 
@@ -1622,17 +1728,17 @@ struct llama_server_context
             {
                 if (slot.is_processing() && system_tokens.size() + slot.cache_tokens.size() >= (size_t) slot.n_ctx)
                 {
+                    // this check is redundant (for good)
+                    // we should never get here, because generation should already stopped in process_token()
+
                     // START LOCALAI changes
                     // Temporary disable context-shifting as it can lead to infinite loops (issue: https://github.com/ggerganov/llama.cpp/issues/3969)
                     // See: https://github.com/mudler/LocalAI/issues/1333
                     // Context is exhausted, release the slot
                     slot.release();
                     send_final_response(slot);
-                    slot.cache_tokens.clear();
-                    slot.n_past = 0;
-                    slot.truncated = false;
-                    slot.has_next_token = true;
-                    LOG("Context exhausted. Slot %d released (%d tokens in cache)\n", slot.id, (int) slot.cache_tokens.size());
+                    slot.has_next_token = false;
+                    LOG_ERROR("context is exhausted, release the slot", {});
 
                     continue;
                     // END LOCALAI changes
@@ -1983,7 +2089,15 @@ struct llama_server_context
                 // prompt evaluated for embedding
                 if (slot.embedding)
                 {
-                    send_embedding(slot);
+                    send_embedding(slot, batch_view);
+                    slot.release();
+                    slot.i_batch = -1;
+                    continue;
+                }
+
+                if (slot.reranker)
+                {
+                    send_rerank(slot, batch_view);
                     slot.release();
                     slot.i_batch = -1;
                     continue;
@@ -2077,7 +2191,11 @@ static void append_to_generated_text_from_generated_token_probs(llama_server_con
 }
 
 std::function<void(int)> shutdown_handler;
-inline void signal_handler(int signal) { shutdown_handler(signal); }
+
+inline void signal_handler(int signal) {
+    exit(1);
+}
+
 
 /////////////////////////////////
 ////////////////////////////////
@@ -2273,15 +2391,15 @@ static std::string get_all_kv_cache_types() {
 }
 
 static void params_parse(const backend::ModelOptions* request,
-                                common_params & params) {
+                                common_params & params, llama_server_context &llama) {
    
     // this is comparable to: https://github.com/ggerganov/llama.cpp/blob/d9b33fe95bd257b36c84ee5769cc048230067d6f/examples/server/server.cpp#L1809
 
-    params.model = request->modelfile();
+    params.model.path = request->modelfile();
     if (!request->mmproj().empty()) {
     // get the directory of modelfile
-      std::string model_dir = params.model.substr(0, params.model.find_last_of("/\\"));
-      params.mmproj = model_dir + "/"+ request->mmproj();
+      std::string model_dir = params.model.path.substr(0, params.model.path.find_last_of("/\\"));
+      params.mmproj.path = model_dir + "/"+ request->mmproj();
     }
     //  params.model_alias ??
     params.model_alias =  request->modelfile();
@@ -2311,6 +2429,20 @@ static void params_parse(const backend::ModelOptions* request,
         add_rpc_devices(std::string(llama_grpc_servers));
     }
     
+     // decode options. Options are in form optname:optvale, or if booleans only optname.
+    for (int i = 0; i < request->options_size(); i++) {
+        std::string opt = request->options(i);
+        char *optname = strtok(&opt[0], ":");
+        char *optval = strtok(NULL, ":");
+        if (optval == NULL) {
+            optval = "true";
+        }
+
+        if (!strcmp(optname, "gpu")) {
+            llama.has_gpu = true;
+        }
+    }
+
     // TODO: Add yarn
 
     if (!request->tensorsplit().empty()) {
@@ -2342,7 +2474,7 @@ static void params_parse(const backend::ModelOptions* request,
         scale_factor = request->lorascale();
      }
      // get the directory of modelfile
-     std::string model_dir = params.model.substr(0, params.model.find_last_of("/\\"));
+     std::string model_dir = params.model.path.substr(0, params.model.path.find_last_of("/\\"));
      params.lora_adapters.push_back({ model_dir + "/"+request->loraadapter(), scale_factor });
     }
     params.use_mlock = request->mlock();
@@ -2374,6 +2506,21 @@ static void params_parse(const backend::ModelOptions* request,
     if ( request->ropefreqscale() != 0.0f ) {
         params.rope_freq_scale = request->ropefreqscale();
     }
+
+    if (request->grammartriggers_size() > 0) {
+        LOG_INFO("configuring grammar triggers", {});
+        llama.grammar_lazy = true;
+        for (int i = 0; i < request->grammartriggers_size(); i++) {
+            common_grammar_trigger trigger;
+	    trigger.type = COMMON_GRAMMAR_TRIGGER_TYPE_WORD;
+            trigger.value = request->grammartriggers(i).word();
+	    // trigger.at_start = request->grammartriggers(i).at_start();
+            llama.grammar_triggers.push_back(trigger);
+            LOG_INFO("grammar trigger", {
+                { "word", trigger.value },
+            });
+        }
+    }
 }
 
 
@@ -2389,7 +2536,7 @@ public:
   grpc::Status LoadModel(ServerContext* context, const backend::ModelOptions* request, backend::Result* result) {
     // Implement LoadModel RPC
     common_params params;
-    params_parse(request, params);
+    params_parse(request, params, llama);
 
     llama_backend_init();
     llama_numa_init(params.numa);
@@ -2411,7 +2558,7 @@ public:
         json data = parse_options(true, request, llama);
         const int task_id = llama.queue_tasks.get_new_id();
         llama.queue_results.add_waiting_task_id(task_id);
-        llama.request_completion(task_id, data, false, false, -1);
+        llama.request_completion(task_id, data, false, false, false, -1);
         while (true)
         {
             task_result result = llama.queue_results.recv(task_id);
@@ -2465,7 +2612,7 @@ public:
         json data = parse_options(false, request, llama);
         const int task_id = llama.queue_tasks.get_new_id();
         llama.queue_results.add_waiting_task_id(task_id);
-        llama.request_completion(task_id, data, false, false, -1);
+        llama.request_completion(task_id, data, false, false, false, -1);
         std::string completion_text;
         task_result result = llama.queue_results.recv(task_id);
         if (!result.error && result.stop) {
@@ -2502,7 +2649,7 @@ public:
         json data = parse_options(false, request, llama);
         const int task_id = llama.queue_tasks.get_new_id();
         llama.queue_results.add_waiting_task_id(task_id);
-        llama.request_completion(task_id, { {"prompt", data["embeddings"]}, { "n_predict", 0}, {"image_data", ""} }, false, true, -1);
+        llama.request_completion(task_id, { {"prompt", data["embeddings"]}, { "n_predict", 0}, {"image_data", ""} }, false, true, false, -1);
         // get the result
         task_result result = llama.queue_results.recv(task_id);
         //std::cout << "Embedding result JSON" << result.result_json.dump() << std::endl;
@@ -2522,6 +2669,58 @@ public:
         return grpc::Status::OK;
     }
 
+    grpc::Status TokenizeString(ServerContext* context, const backend::PredictOptions* request, backend::TokenizationResponse* response){
+         json data = parse_options(false, request, llama);
+
+         std::vector<llama_token> tokens = llama.tokenize(data["prompt"],false);
+
+         for (int i=0 ; i< tokens.size(); i++){
+            response->add_tokens(tokens[i]);
+         }
+
+        return grpc::Status::OK;
+    }
+
+    grpc::Status Rerank(ServerContext* context, const backend::RerankRequest* request, backend::RerankResult* rerankResult) {
+        // Create a JSON object with the query and documents
+        json data = {
+            {"prompt", request->query()},
+            {"documents", request->documents()},
+            {"top_n", request->top_n()}
+        };
+
+        // Generate a new task ID
+        const int task_id = llama.queue_tasks.get_new_id();
+        llama.queue_results.add_waiting_task_id(task_id);
+
+        // Queue the task with reranking mode enabled
+        llama.request_completion(task_id, data, false, false, true, -1);
+
+        // Get the result
+        task_result result = llama.queue_results.recv(task_id);
+        llama.queue_results.remove_waiting_task_id(task_id);
+
+        if (!result.error && result.stop) {
+            // Set usage information
+            backend::Usage* usage = rerankResult->mutable_usage();
+            usage->set_total_tokens(result.result_json.value("tokens", 0));
+            usage->set_prompt_tokens(result.result_json.value("tokens", 0));
+
+            // Get the score from the result
+            float score = result.result_json.value("score", 0.0f);
+
+            // Create document results for each input document
+            for (int i = 0; i < request->documents_size(); i++) {
+                backend::DocumentResult* doc_result = rerankResult->add_results();
+                doc_result->set_index(i);
+                doc_result->set_text(request->documents(i));
+                doc_result->set_relevance_score(score);
+            }
+        }
+
+        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();
 
@@ -2554,7 +2753,9 @@ void RunServer(const std::string& server_address) {
   ServerBuilder builder;
   builder.AddListeningPort(server_address, grpc::InsecureServerCredentials());
   builder.RegisterService(&service);
-
+  builder.SetMaxMessageSize(50 * 1024 * 1024); // 50MB
+  builder.SetMaxSendMessageSize(50 * 1024 * 1024); // 50MB
+  builder.SetMaxReceiveMessageSize(50 * 1024 * 1024); // 50MB
   std::unique_ptr<Server> server(builder.BuildAndStart());
   std::cout << "Server listening on " << server_address << std::endl;
   server->Wait();
@@ -2563,6 +2764,20 @@ void RunServer(const std::string& server_address) {
 int main(int argc, char** argv) {
   std::string server_address("localhost:50051");
 
+#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
+    struct sigaction sigint_action;
+    sigint_action.sa_handler = signal_handler;
+    sigemptyset (&sigint_action.sa_mask);
+    sigint_action.sa_flags = 0;
+    sigaction(SIGINT, &sigint_action, NULL);
+    sigaction(SIGTERM, &sigint_action, NULL);
+#elif defined (_WIN32)
+    auto console_ctrl_handler = +[](DWORD ctrl_type) -> BOOL {
+        return (ctrl_type == CTRL_C_EVENT) ? (signal_handler(SIGINT), true) : false;
+    };
+    SetConsoleCtrlHandler(reinterpret_cast<PHANDLER_ROUTINE>(console_ctrl_handler), true);
+#endif
+
   // Define long and short options
   struct option long_options[] = {
       {"addr", required_argument, nullptr, 'a'},

backend/cpp/llama/prepare.sh

@@ -21,6 +21,7 @@ fi
 ## XXX: In some versions of CMake clip wasn't being built before llama.
 ## This is an hack for now, but it should be fixed in the future.
 cp -rfv llama.cpp/examples/llava/clip.h llama.cpp/examples/grpc-server/clip.h
+cp -rfv llama.cpp/examples/llava/clip-impl.h llama.cpp/examples/grpc-server/clip-impl.h
 cp -rfv llama.cpp/examples/llava/llava.cpp llama.cpp/examples/grpc-server/llava.cpp
 echo '#include "llama.h"' > llama.cpp/examples/grpc-server/llava.h
 cat llama.cpp/examples/llava/llava.h >> llama.cpp/examples/grpc-server/llava.h

backend/cpp/llama/utils.hpp

@@ -61,6 +61,7 @@ struct task_server {
     json data;
     bool infill_mode = false;
     bool embedding_mode = false;
+    bool reranking_mode = false;
     int multitask_id = -1;
 };
 

backend/go/image/stablediffusion-ggml/Makefile

@@ -8,6 +8,13 @@ ONEAPI_VARS?=/opt/intel/oneapi/setvars.sh
 # keep standard at C11 and C++11
 CXXFLAGS = -I. -I$(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp/thirdparty -I$(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp/ggml/include -I$(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp -O3 -DNDEBUG -std=c++17 -fPIC
 
+GOCMD?=go
+CGO_LDFLAGS?=
+# Avoid parent make file overwriting CGO_LDFLAGS which is needed for hipblas
+CGO_LDFLAGS_SYCL=
+GO_TAGS?=
+LD_FLAGS?=
+
 # Disable Shared libs as we are linking on static gRPC and we can't mix shared and static
 CMAKE_ARGS+=-DBUILD_SHARED_LIBS=OFF
 
@@ -21,7 +28,7 @@ else ifeq ($(BUILD_TYPE),openblas)
 # If build type is clblas (openCL) we set -DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
 else ifeq ($(BUILD_TYPE),clblas)
 	CMAKE_ARGS+=-DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
-# If it's hipblas we do have also to set CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++ 
+# If it's hipblas we do have also to set CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++
 else ifeq ($(BUILD_TYPE),hipblas)
 	CMAKE_ARGS+=-DGGML_HIP=ON
 # If it's OSX, DO NOT embed the metal library - -DGGML_METAL_EMBED_LIBRARY=ON requires further investigation
@@ -36,16 +43,35 @@ else ifeq ($(OS),Darwin)
 	endif
 endif
 
-# ifeq ($(BUILD_TYPE),sycl_f16)
-# 	CMAKE_ARGS+=-DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_SYCL_F16=ON -DSD_SYCL=ON -DGGML_SYCL_F16=ON
-# endif
+ifeq ($(BUILD_TYPE),sycl_f16)
+	CMAKE_ARGS+=-DGGML_SYCL=ON \
+		-DCMAKE_C_COMPILER=icx \
+		-DCMAKE_CXX_COMPILER=icpx \
+		-DSD_SYCL=ON \
+		-DGGML_SYCL_F16=ON
+	CC=icx
+	CXX=icpx
+	CGO_LDFLAGS_SYCL += -fsycl -L${DNNLROOT}/lib -ldnnl ${MKLROOT}/lib/intel64/libmkl_sycl.a -fiopenmp -fopenmp-targets=spir64 -lOpenCL
+	CGO_LDFLAGS_SYCL += $(shell pkg-config --libs mkl-static-lp64-gomp)
+	CGO_CXXFLAGS += -fiopenmp -fopenmp-targets=spir64
+	CGO_CXXFLAGS += $(shell pkg-config --cflags mkl-static-lp64-gomp )
+endif
 
-# ifeq ($(BUILD_TYPE),sycl_f32)
-# 	CMAKE_ARGS+=-DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DSD_SYCL=ON
-# endif
+ifeq ($(BUILD_TYPE),sycl_f32)
+	CMAKE_ARGS+=-DGGML_SYCL=ON \
+		-DCMAKE_C_COMPILER=icx \
+		-DCMAKE_CXX_COMPILER=icpx \
+		-DSD_SYCL=ON
+	CC=icx
+	CXX=icpx
+	CGO_LDFLAGS_SYCL += -fsycl -L${DNNLROOT}/lib -ldnnl ${MKLROOT}/lib/intel64/libmkl_sycl.a -fiopenmp -fopenmp-targets=spir64 -lOpenCL
+	CGO_LDFLAGS_SYCL += $(shell pkg-config --libs mkl-static-lp64-gomp)
+	CGO_CXXFLAGS += -fiopenmp -fopenmp-targets=spir64
+	CGO_CXXFLAGS += $(shell pkg-config --cflags mkl-static-lp64-gomp )
+endif
 
 # warnings
-CXXFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function
+# CXXFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function
 
 # Find all .a archives in ARCHIVE_DIR
 # (ggml can have different backends cpu, cuda, etc., each backend generates a .a archive)
@@ -86,11 +112,24 @@ endif
 	$(MAKE) $(COMBINED_LIB)
 
 gosd.o:
+ifneq (,$(findstring sycl,$(BUILD_TYPE)))
+	+bash -c "source $(ONEAPI_VARS); \
+	$(CXX) $(CXXFLAGS) gosd.cpp -o gosd.o -c"
+else
 	$(CXX) $(CXXFLAGS) gosd.cpp -o gosd.o -c
+endif
 
 libsd.a: gosd.o
 	cp $(INCLUDE_PATH)/build/libstable-diffusion.a ./libsd.a
 	$(AR) rcs libsd.a gosd.o
 
+stablediffusion-ggml:
+	CGO_LDFLAGS="$(CGO_LDFLAGS) $(CGO_LDFLAGS_SYCL)" C_INCLUDE_PATH="$(INCLUDE_PATH)" LIBRARY_PATH="$(LIBRARY_PATH)" \
+	CC="$(CC)" CXX="$(CXX)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o ../../../../backend-assets/grpc/stablediffusion-ggml ./
+ifneq ($(UPX),)
+	$(UPX) ../../../../backend-assets/grpc/stablediffusion-ggml
+endif
+
 clean:
-	rm -rf gosd.o libsd.a build $(COMBINED_LIB)
+	rm -rf gosd.o libsd.a build $(COMBINED_LIB)

backend/go/image/stablediffusion-ggml/gosd.cpp

@@ -35,6 +35,8 @@ const char* sample_method_str[] = {
     "ipndm",
     "ipndm_v",
     "lcm",
+    "ddim_trailing",
+    "tcd",
 };
 
 // Names of the sigma schedule overrides, same order as sample_schedule in stable-diffusion.h
@@ -173,6 +175,7 @@ int gen_image(char *text, char *negativeText, int width, int height, int steps,
                             -1, //clip_skip
                             cfg_scale, // sfg_scale
                             3.5f,
+			    0, // eta
                             width,
                             height,
                             sample_method, 

backend/go/llm/llama-ggml/llama.go

@@ -1,204 +0,0 @@
-package main
-
-// This is a wrapper to statisfy the GRPC service interface
-// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
-import (
-	"fmt"
-
-	"github.com/go-skynet/go-llama.cpp"
-	"github.com/mudler/LocalAI/pkg/grpc/base"
-	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
-)
-
-type LLM struct {
-	base.SingleThread
-
-	llama *llama.LLama
-}
-
-func (llm *LLM) Load(opts *pb.ModelOptions) error {
-	ropeFreqBase := float32(10000)
-	ropeFreqScale := float32(1)
-
-	if opts.RopeFreqBase != 0 {
-		ropeFreqBase = opts.RopeFreqBase
-	}
-	if opts.RopeFreqScale != 0 {
-		ropeFreqScale = opts.RopeFreqScale
-	}
-
-	llamaOpts := []llama.ModelOption{
-		llama.WithRopeFreqBase(ropeFreqBase),
-		llama.WithRopeFreqScale(ropeFreqScale),
-	}
-
-	if opts.NGQA != 0 {
-		llamaOpts = append(llamaOpts, llama.WithGQA(int(opts.NGQA)))
-	}
-
-	if opts.RMSNormEps != 0 {
-		llamaOpts = append(llamaOpts, llama.WithRMSNormEPS(opts.RMSNormEps))
-	}
-
-	if opts.ContextSize != 0 {
-		llamaOpts = append(llamaOpts, llama.SetContext(int(opts.ContextSize)))
-	}
-	if opts.F16Memory {
-		llamaOpts = append(llamaOpts, llama.EnableF16Memory)
-	}
-	if opts.Embeddings {
-		llamaOpts = append(llamaOpts, llama.EnableEmbeddings)
-	}
-	if opts.NGPULayers != 0 {
-		llamaOpts = append(llamaOpts, llama.SetGPULayers(int(opts.NGPULayers)))
-	}
-
-	llamaOpts = append(llamaOpts, llama.SetMMap(opts.MMap))
-	llamaOpts = append(llamaOpts, llama.SetMainGPU(opts.MainGPU))
-	llamaOpts = append(llamaOpts, llama.SetTensorSplit(opts.TensorSplit))
-	if opts.NBatch != 0 {
-		llamaOpts = append(llamaOpts, llama.SetNBatch(int(opts.NBatch)))
-	} else {
-		llamaOpts = append(llamaOpts, llama.SetNBatch(512))
-	}
-
-	if opts.NUMA {
-		llamaOpts = append(llamaOpts, llama.EnableNUMA)
-	}
-
-	if opts.LowVRAM {
-		llamaOpts = append(llamaOpts, llama.EnabelLowVRAM)
-	}
-
-	model, err := llama.New(opts.ModelFile, llamaOpts...)
-	llm.llama = model
-
-	return err
-}
-
-func buildPredictOptions(opts *pb.PredictOptions) []llama.PredictOption {
-	ropeFreqBase := float32(10000)
-	ropeFreqScale := float32(1)
-
-	if opts.RopeFreqBase != 0 {
-		ropeFreqBase = opts.RopeFreqBase
-	}
-	if opts.RopeFreqScale != 0 {
-		ropeFreqScale = opts.RopeFreqScale
-	}
-	predictOptions := []llama.PredictOption{
-		llama.SetTemperature(opts.Temperature),
-		llama.SetTopP(opts.TopP),
-		llama.SetTopK(int(opts.TopK)),
-		llama.SetTokens(int(opts.Tokens)),
-		llama.SetThreads(int(opts.Threads)),
-		llama.WithGrammar(opts.Grammar),
-		llama.SetRopeFreqBase(ropeFreqBase),
-		llama.SetRopeFreqScale(ropeFreqScale),
-		llama.SetNegativePromptScale(opts.NegativePromptScale),
-		llama.SetNegativePrompt(opts.NegativePrompt),
-	}
-
-	if opts.PromptCacheAll {
-		predictOptions = append(predictOptions, llama.EnablePromptCacheAll)
-	}
-
-	if opts.PromptCacheRO {
-		predictOptions = append(predictOptions, llama.EnablePromptCacheRO)
-	}
-
-	// Expected absolute path
-	if opts.PromptCachePath != "" {
-		predictOptions = append(predictOptions, llama.SetPathPromptCache(opts.PromptCachePath))
-	}
-
-	if opts.Mirostat != 0 {
-		predictOptions = append(predictOptions, llama.SetMirostat(int(opts.Mirostat)))
-	}
-
-	if opts.MirostatETA != 0 {
-		predictOptions = append(predictOptions, llama.SetMirostatETA(opts.MirostatETA))
-	}
-
-	if opts.MirostatTAU != 0 {
-		predictOptions = append(predictOptions, llama.SetMirostatTAU(opts.MirostatTAU))
-	}
-
-	if opts.Debug {
-		predictOptions = append(predictOptions, llama.Debug)
-	}
-
-	predictOptions = append(predictOptions, llama.SetStopWords(opts.StopPrompts...))
-
-	if opts.PresencePenalty != 0 {
-		predictOptions = append(predictOptions, llama.SetPenalty(opts.PresencePenalty))
-	}
-
-	if opts.NKeep != 0 {
-		predictOptions = append(predictOptions, llama.SetNKeep(int(opts.NKeep)))
-	}
-
-	if opts.Batch != 0 {
-		predictOptions = append(predictOptions, llama.SetBatch(int(opts.Batch)))
-	}
-
-	if opts.F16KV {
-		predictOptions = append(predictOptions, llama.EnableF16KV)
-	}
-
-	if opts.IgnoreEOS {
-		predictOptions = append(predictOptions, llama.IgnoreEOS)
-	}
-
-	if opts.Seed != 0 {
-		predictOptions = append(predictOptions, llama.SetSeed(int(opts.Seed)))
-	}
-
-	//predictOptions = append(predictOptions, llama.SetLogitBias(c.Seed))
-
-	predictOptions = append(predictOptions, llama.SetFrequencyPenalty(opts.FrequencyPenalty))
-	predictOptions = append(predictOptions, llama.SetMlock(opts.MLock))
-	predictOptions = append(predictOptions, llama.SetMemoryMap(opts.MMap))
-	predictOptions = append(predictOptions, llama.SetPredictionMainGPU(opts.MainGPU))
-	predictOptions = append(predictOptions, llama.SetPredictionTensorSplit(opts.TensorSplit))
-	predictOptions = append(predictOptions, llama.SetTailFreeSamplingZ(opts.TailFreeSamplingZ))
-	predictOptions = append(predictOptions, llama.SetTypicalP(opts.TypicalP))
-	return predictOptions
-}
-
-func (llm *LLM) Predict(opts *pb.PredictOptions) (string, error) {
-	return llm.llama.Predict(opts.Prompt, buildPredictOptions(opts)...)
-}
-
-func (llm *LLM) PredictStream(opts *pb.PredictOptions, results chan string) error {
-	predictOptions := buildPredictOptions(opts)
-
-	predictOptions = append(predictOptions, llama.SetTokenCallback(func(token string) bool {
-		results <- token
-		return true
-	}))
-
-	go func() {
-		_, err := llm.llama.Predict(opts.Prompt, predictOptions...)
-		if err != nil {
-			fmt.Println("err: ", err)
-		}
-		close(results)
-	}()
-
-	return nil
-}
-
-func (llm *LLM) Embeddings(opts *pb.PredictOptions) ([]float32, error) {
-	predictOptions := buildPredictOptions(opts)
-
-	if len(opts.EmbeddingTokens) > 0 {
-		tokens := []int{}
-		for _, t := range opts.EmbeddingTokens {
-			tokens = append(tokens, int(t))
-		}
-		return llm.llama.TokenEmbeddings(tokens, predictOptions...)
-	}
-
-	return llm.llama.Embeddings(opts.Embeddings, predictOptions...)
-}

backend/go/llm/llama-ggml/main.go

@@ -1,19 +0,0 @@
-package main
-
-import (
-	"flag"
-
-	grpc "github.com/mudler/LocalAI/pkg/grpc"
-)
-
-var (
-	addr = flag.String("addr", "localhost:50051", "the address to connect to")
-)
-
-func main() {
-	flag.Parse()
-
-	if err := grpc.StartServer(*addr, &LLM{}); err != nil {
-		panic(err)
-	}
-}

backend/python/autogptq/Makefile

@@ -1,17 +0,0 @@
-.PHONY: autogptq
-autogptq: protogen
-	bash install.sh
-
-.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 -rf venv __pycache__

backend/python/autogptq/README.md

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

backend/python/autogptq/backend.py

@@ -1,153 +0,0 @@
-#!/usr/bin/env python3
-from concurrent import futures
-import argparse
-import signal
-import sys
-import os
-import time
-import base64
-
-import grpc
-import backend_pb2
-import backend_pb2_grpc
-
-from auto_gptq import AutoGPTQForCausalLM
-from transformers import AutoTokenizer, AutoModelForCausalLM
-from transformers import TextGenerationPipeline
-
-_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 Health(self, request, context):
-        return backend_pb2.Reply(message=bytes("OK", 'utf-8'))
-    def LoadModel(self, request, context):
-        try:
-            device = "cuda:0"
-            if request.Device != "":
-                device = request.Device
-
-            # support loading local model files
-            model_path = os.path.join(os.environ.get('MODELS_PATH', './'), request.Model)
-            tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=True, trust_remote_code=request.TrustRemoteCode)
-
-            # support model `Qwen/Qwen-VL-Chat-Int4`
-            if "qwen-vl" in request.Model.lower():
-                self.model_name = "Qwen-VL-Chat"
-                model = AutoModelForCausalLM.from_pretrained(model_path, 
-                    trust_remote_code=request.TrustRemoteCode,
-                    device_map="auto").eval()
-            else:
-                model = AutoGPTQForCausalLM.from_quantized(model_path,
-                    model_basename=request.ModelBaseName,
-                    use_safetensors=True,
-                    trust_remote_code=request.TrustRemoteCode,
-                    device=device,
-                    use_triton=request.UseTriton,
-                    quantize_config=None)
-            
-            self.model = model
-            self.tokenizer = tokenizer
-        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 = 1.0
-        if request.Penalty != 0.0:
-            penalty = request.Penalty
-        tokens = 512
-        if request.Tokens != 0:
-            tokens = request.Tokens
-        top_p = 0.95
-        if request.TopP != 0.0:
-            top_p = request.TopP
-
-        
-        prompt_images = self.recompile_vl_prompt(request)
-        compiled_prompt = prompt_images[0]
-        print(f"Prompt: {compiled_prompt}", file=sys.stderr)
-
-        # Implement Predict RPC
-        pipeline = TextGenerationPipeline(
-            model=self.model, 
-            tokenizer=self.tokenizer,
-            max_new_tokens=tokens,
-            temperature=request.Temperature,
-            top_p=top_p,
-            repetition_penalty=penalty,
-            )
-        t = pipeline(compiled_prompt)[0]["generated_text"]
-        print(f"generated_text: {t}", file=sys.stderr)
-        
-        if compiled_prompt in t:
-            t = t.replace(compiled_prompt, "")
-        # house keeping. Remove the image files from /tmp folder
-        for img_path in prompt_images[1]:
-            try:
-                os.remove(img_path)
-            except Exception as e:
-                print(f"Error removing image file: {img_path}, {e}", file=sys.stderr)
-
-        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 recompile_vl_prompt(self, request):
-        prompt = request.Prompt
-        image_paths = []
-
-        if "qwen-vl" in self.model_name.lower():
-            # request.Images is an array which contains base64 encoded images. Iterate the request.Images array, decode and save each image to /tmp folder with a random filename.
-            # Then, save the image file paths to an array "image_paths".
-            # read "request.Prompt", replace "[img-%d]" with the image file paths in the order they appear in "image_paths". Save the new prompt to "prompt".
-            for i, img in enumerate(request.Images):
-                timestamp = str(int(time.time() * 1000))  # Generate timestamp
-                img_path = f"/tmp/vl-{timestamp}.jpg"  # Use timestamp in filename
-                with open(img_path, "wb") as f:
-                    f.write(base64.b64decode(img))
-                image_paths.append(img_path)
-                prompt = prompt.replace(f"[img-{i}]", "<img>" + img_path + "</img>,")
-        else:
-            prompt = request.Prompt
-        return (prompt, image_paths)
-
-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/autogptq/install.sh

@@ -1,14 +0,0 @@
-#!/bin/bash
-set -e
-
-source $(dirname $0)/../common/libbackend.sh
-
-# This is here because the Intel pip index is broken and returns 200 status codes for every package name, it just doesn't return any package links.
-# This makes uv think that the package exists in the Intel pip index, and by default it stops looking at other pip indexes once it finds a match.
-# We need uv to continue falling through to the pypi default index to find optimum[openvino] in the pypi index
-# the --upgrade actually allows us to *downgrade* torch to the version provided in the Intel pip index
-if [ "x${BUILD_PROFILE}" == "xintel" ]; then
-    EXTRA_PIP_INSTALL_FLAGS+=" --upgrade --index-strategy=unsafe-first-match"
-fi
-
-installRequirements

backend/python/autogptq/requirements-cublas11.txt

@@ -1,2 +0,0 @@
---extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.4.1+cu118

backend/python/autogptq/requirements-cublas12.txt

@@ -1 +0,0 @@
-torch==2.4.1

backend/python/autogptq/requirements-hipblas.txt

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

backend/python/autogptq/requirements-intel.txt

@@ -1,6 +0,0 @@
---extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
-torch==2.3.1+cxx11.abi
-oneccl_bind_pt==2.3.100+xpu
-optimum[openvino]
-setuptools

backend/python/autogptq/requirements.txt

@@ -1,6 +0,0 @@
-accelerate
-auto-gptq==0.7.1
-grpcio==1.69.0
-protobuf
-certifi
-transformers

backend/python/autogptq/run.sh

@@ -1,4 +0,0 @@
-#!/bin/bash
-source $(dirname $0)/../common/libbackend.sh
-
-startBackend $@

backend/python/autogptq/test.sh

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

backend/python/bark/backend.py

@@ -61,7 +61,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         return backend_pb2.Result(success=True)
 
 def serve(address):
-    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS))
+    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS),
+        options=[
+            ('grpc.max_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_send_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_receive_message_length', 50 * 1024 * 1024),  # 50MB
+        ])
     backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
     server.add_insecure_port(address)
     server.start()

backend/python/bark/requirements.txt

@@ -1,4 +1,4 @@
 bark==0.1.5
-grpcio==1.69.0
+grpcio==1.71.0
 protobuf
 certifi

backend/python/common/template/requirements.txt

@@ -1,3 +1,3 @@
-grpcio==1.69.0
+grpcio==1.71.0
 protobuf
 grpcio-tools

backend/python/coqui/backend.py

@@ -86,7 +86,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         return backend_pb2.Result(success=True)
 
 def serve(address):
-    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS))
+    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS),
+        options=[
+            ('grpc.max_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_send_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_receive_message_length', 50 * 1024 * 1024),  # 50MB
+        ])
     backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
     server.add_insecure_port(address)
     server.start()

backend/python/coqui/requirements-cpu.txt

@@ -1,4 +1,4 @@
-transformers
+transformers==4.48.3
 accelerate
 torch==2.4.1
 coqui-tts

backend/python/coqui/requirements-cublas11.txt

@@ -1,6 +1,6 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
 torch==2.4.1+cu118
 torchaudio==2.4.1+cu118
-transformers
+transformers==4.48.3
 accelerate
 coqui-tts

backend/python/coqui/requirements-cublas12.txt

@@ -1,5 +1,5 @@
 torch==2.4.1
 torchaudio==2.4.1
-transformers
+transformers==4.48.3
 accelerate
 coqui-tts

backend/python/coqui/requirements-hipblas.txt

@@ -1,6 +1,6 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
 torch==2.4.1+rocm6.0
 torchaudio==2.4.1+rocm6.0
-transformers
+transformers==4.48.3
 accelerate
 coqui-tts

backend/python/coqui/requirements-intel.txt

@@ -5,6 +5,6 @@ torchaudio==2.3.1+cxx11.abi
 oneccl_bind_pt==2.3.100+xpu
 optimum[openvino]
 setuptools
-transformers
+transformers==4.48.3
 accelerate
 coqui-tts

backend/python/coqui/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.69.0
+grpcio==1.71.0
 protobuf
 certifi
 packaging==24.1

backend/python/diffusers/backend.py

@@ -19,7 +19,7 @@ import grpc
 
 from diffusers import SanaPipeline, StableDiffusion3Pipeline, StableDiffusionXLPipeline, StableDiffusionDepth2ImgPipeline, DPMSolverMultistepScheduler, StableDiffusionPipeline, DiffusionPipeline, \
     EulerAncestralDiscreteScheduler, FluxPipeline, FluxTransformer2DModel
-from diffusers import StableDiffusionImg2ImgPipeline, AutoPipelineForText2Image, ControlNetModel, StableVideoDiffusionPipeline
+from diffusers import StableDiffusionImg2ImgPipeline, AutoPipelineForText2Image, ControlNetModel, StableVideoDiffusionPipeline, Lumina2Text2ImgPipeline
 from diffusers.pipelines.stable_diffusion import safety_checker
 from diffusers.utils import load_image, export_to_video
 from compel import Compel, ReturnedEmbeddingsType
@@ -159,6 +159,18 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                 torchType = torch.float16
                 variant = "fp16"
 
+            options = request.Options
+
+            # empty dict
+            self.options = {}
+
+            # The options are a list of strings in this form optname:optvalue
+            # We are storing all the options in a dict so we can use it later when
+            # generating the images
+            for opt in options:
+                key, value = opt.split(":")
+                self.options[key] = value
+
             local = False
             modelFile = request.Model
 
@@ -275,6 +287,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
 
                     if request.LowVRAM:
                         self.pipe.enable_model_cpu_offload()
+            elif request.PipelineType == "Lumina2Text2ImgPipeline":
+                self.pipe = Lumina2Text2ImgPipeline.from_pretrained(
+                    request.Model,
+                    torch_dtype=torch.bfloat16)
+                if request.LowVRAM:
+                    self.pipe.enable_model_cpu_offload()
             elif request.PipelineType == "SanaPipeline":
                 self.pipe = SanaPipeline.from_pretrained(
                     request.Model,
@@ -441,6 +459,9 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         # create a dictionary of parameters by using the keys from EnableParameters and the values from defaults
         kwargs = {key: options.get(key) for key in keys if key in options}
 
+        # populate kwargs from self.options.
+        kwargs.update(self.options)
+
         # Set seed
         if request.seed > 0:
             kwargs["generator"] = torch.Generator(device=self.device).manual_seed(
@@ -501,7 +522,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
 
 
 def serve(address):
-    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS))
+    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS),
+        options=[
+            ('grpc.max_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_send_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_receive_message_length', 50 * 1024 * 1024),  # 50MB
+        ])
     backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
     server.add_insecure_port(address)
     server.start()

backend/python/diffusers/requirements.txt

@@ -1,5 +1,5 @@
 setuptools
-grpcio==1.69.0
+grpcio==1.71.0
 pillow
 protobuf
 certifi

backend/python/exllama2/backend.py

@@ -105,7 +105,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
 
 
 def serve(address):
-    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS))
+    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS),
+        options=[
+            ('grpc.max_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_send_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_receive_message_length', 50 * 1024 * 1024),  # 50MB
+        ])
     backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
     server.add_insecure_port(address)
     server.start()

backend/python/exllama2/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.69.0
+grpcio==1.71.0
 protobuf
 certifi
 wheel

backend/python/faster-whisper/backend.py

@@ -62,7 +62,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         return backend_pb2.TranscriptResult(segments=resultSegments, text=text)
 
 def serve(address):
-    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS))
+    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS),
+        options=[
+            ('grpc.max_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_send_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_receive_message_length', 50 * 1024 * 1024),  # 50MB
+        ])
     backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
     server.add_insecure_port(address)
     server.start()

backend/python/faster-whisper/requirements.txt

@@ -1,3 +1,3 @@
-grpcio==1.69.0
+grpcio==1.71.0
 protobuf
 grpcio-tools

backend/python/kokoro/backend.py

@@ -99,7 +99,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         return backend_pb2.Result(success=True)
 
 def serve(address):
-    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS))
+    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS),
+        options=[
+            ('grpc.max_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_send_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_receive_message_length', 50 * 1024 * 1024),  # 50MB
+        ])
     backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
     server.add_insecure_port(address)
     server.start()

backend/python/kokoro/requirements.txt

@@ -1,4 +1,4 @@
-grpcio==1.69.0
+grpcio==1.71.0
 protobuf
 phonemizer
 scipy

backend/python/rerankers/backend.py

@@ -91,7 +91,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         return backend_pb2.RerankResult(usage=usage, results=results)
 
 def serve(address):
-    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS))
+    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS),
+        options=[
+            ('grpc.max_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_send_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_receive_message_length', 50 * 1024 * 1024),  # 50MB
+        ])
     backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
     server.add_insecure_port(address)
     server.start()

backend/python/rerankers/requirements.txt

@@ -1,3 +1,3 @@
-grpcio==1.69.0
+grpcio==1.71.0
 protobuf
 certifi

backend/python/transformers/backend.py

@@ -559,7 +559,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
 
 async def serve(address):
     # Start asyncio gRPC server
-    server = grpc.aio.server(migration_thread_pool=futures.ThreadPoolExecutor(max_workers=MAX_WORKERS))
+    server = grpc.aio.server(migration_thread_pool=futures.ThreadPoolExecutor(max_workers=MAX_WORKERS),
+        options=[
+            ('grpc.max_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_send_message_length', 50 * 1024 * 1024),  # 50MB
+            ('grpc.max_receive_message_length', 50 * 1024 * 1024),  # 50MB
+        ])
     # Add the servicer to the server
     backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
     # Bind the server to the address

backend/python/transformers/requirements-cpu.txt

@@ -5,4 +5,4 @@ accelerate
 transformers
 bitsandbytes
 outetts
-sentence-transformers==3.3.1
+sentence-transformers==3.4.1

backend/python/transformers/requirements-cublas11.txt

@@ -6,4 +6,4 @@ accelerate
 transformers
 bitsandbytes
 outetts
-sentence-transformers==3.3.1
+sentence-transformers==3.4.1

backend/python/transformers/requirements-cublas12.txt

@@ -5,4 +5,4 @@ numba==0.60.0
 transformers
 bitsandbytes
 outetts
-sentence-transformers==3.3.1
+sentence-transformers==3.4.1

backend/python/transformers/requirements-hipblas.txt

@@ -7,4 +7,4 @@ numba==0.60.0
 bitsandbytes
 outetts
 bitsandbytes
-sentence-transformers==3.3.1
+sentence-transformers==3.4.1