debu2

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/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/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
-}

.devcontainer-scripts/postcreate.sh

@@ -1,17 +0,0 @@
-#!/bin/bash
-
-cd /workspace
-
-# Get the files into the volume without a bind mount
-if [ ! -d ".git" ]; then
-    git clone https://github.com/mudler/LocalAI.git .
-else
-    git fetch
-fi
-
-echo "Standard Post-Create script completed."
-
-if [ -f "/devcontainer-customization/postcreate.sh" ]; then
-    echo "Launching customization postcreate.sh"
-    bash "/devcontainer-customization/postcreate.sh"
-fi

.devcontainer-scripts/poststart.sh

@@ -1,16 +0,0 @@
-#!/bin/bash
-
-cd /workspace
-
-# Grab the pre-stashed backend assets to avoid build issues
-cp -r /build/backend-assets /workspace/backend-assets
-
-# Ensures generated source files are present upon load
-make prepare
-
-echo "Standard Post-Start script completed."
-
-if [ -f "/devcontainer-customization/poststart.sh" ]; then
-    echo "Launching customization poststart.sh"
-    bash "/devcontainer-customization/poststart.sh"
-fi

.devcontainer-scripts/utils.sh

@@ -1,55 +0,0 @@
-#!/bin/bash
-
-# This file contains some really simple functions that are useful when building up customization scripts.
-
-
-# Checks if the git config has a user registered - and sets it up if not.
-#
-# Param 1: name
-# Param 2: email
-#
-config_user() {
-    echo "Configuring git for $1 <$2>"
-    local gcn=$(git config --global user.name)
-    if [ -z "${gcn}" ]; then
-        echo "Setting up git user / remote"
-        git config --global user.name "$1"
-        git config --global user.email "$2"
-        
-    fi
-}
-
-# Checks if the git remote is configured - and sets it up if not. Fetches either way.
-#
-# Param 1: remote name
-# Param 2: remote url
-#
-config_remote() {
-    echo "Adding git remote and fetching $2 as $1"
-    local gr=$(git remote -v | grep $1)
-    if [ -z "${gr}" ]; then
-        git remote add $1 $2
-    fi
-    git fetch $1
-}
-
-# Setup special .ssh files
-# Prints out lines of text to make things pretty
-# Param 1: bash array, filenames relative to the customization directory that should be copied to ~/.ssh
-setup_ssh() {
-    echo "starting ~/.ssh directory setup..."
-    mkdir -p "${HOME}.ssh"
-    chmod 0700 "${HOME}/.ssh"
-    echo "-----"
-    local files=("$@")
-    for file in "${files[@]}" ; do
-        local cfile="/devcontainer-customization/${file}"
-        local hfile="${HOME}/.ssh/${file}"
-        if [ ! -f "${hfile}" ]; then
-            echo "copying \"${file}\""
-            cp "${cfile}" "${hfile}"
-            chmod 600 "${hfile}"
-        fi
-    done
-    echo "~/.ssh directory setup complete!"
-}

.devcontainer/customization/README.md

@@ -1,25 +0,0 @@
-Place any additional resources your environment requires in this directory
-
-Script hooks are currently called for:
-`postcreate.sh` and `poststart.sh`
-
-If files with those names exist here, they will be called at the end of the normal script.
-
-This is a good place to set things like `git config --global user.name` are set - and to handle any other files that are mounted via this directory.
-
-To assist in doing so, `source /.devcontainer-scripts/utils.sh` will provide utility functions that may be useful - for example:
-
-```
-#!/bin/bash
-
-source "/.devcontainer-scripts/utils.sh"
-
-sshfiles=("config", "key.pub")
-
-setup_ssh "${sshfiles[@]}"
-
-config_user "YOUR NAME" "YOUR EMAIL"
-
-config_remote "REMOTE NAME" "REMOTE URL"
-
-```

.devcontainer/devcontainer.json

@@ -1,24 +0,0 @@
-{
-    "$schema": "https://raw.githubusercontent.com/devcontainers/spec/main/schemas/devContainer.schema.json",
-    "name": "LocalAI",
-    "workspaceFolder": "/workspace",
-    "dockerComposeFile": [ "./docker-compose-devcontainer.yml" ],
-    "service": "api",
-    "shutdownAction": "stopCompose",
-    "customizations": {
-        "vscode": {
-            "extensions": [
-                "golang.go",
-                "ms-vscode.makefile-tools",
-                "ms-azuretools.vscode-docker",
-                "ms-python.python",
-                "ms-python.debugpy",
-                "wayou.vscode-todo-highlight",
-                "waderyan.gitblame"
-            ]
-        }
-    },
-    "forwardPorts": [8080, 3000],
-    "postCreateCommand": "bash /.devcontainer-scripts/postcreate.sh",
-    "postStartCommand": "bash /.devcontainer-scripts/poststart.sh"
-}

.devcontainer/docker-compose-devcontainer.yml

@@ -1,48 +0,0 @@
-services:
-  api:
-    build:
-      context: ..
-      dockerfile: Dockerfile
-      target: devcontainer
-      args:
-      - FFMPEG=true
-      - IMAGE_TYPE=extras
-      - GO_TAGS=stablediffusion p2p tts
-    env_file:
-      - ../.env
-    ports:
-      - 8080:8080
-    volumes:
-      - localai_workspace:/workspace
-      - ../models:/host-models
-      - ./customization:/devcontainer-customization
-    command: /bin/sh -c "while sleep 1000; do :; done"
-    cap_add:
-      - SYS_PTRACE
-    security_opt:
-      - seccomp:unconfined
-  prometheus:
-    image: prom/prometheus
-    container_name: prometheus
-    command:
-      - '--config.file=/etc/prometheus/prometheus.yml'
-    ports:
-      - 9090:9090
-    restart: unless-stopped
-    volumes:
-      - ./prometheus:/etc/prometheus
-      - prom_data:/prometheus
-  grafana:
-    image: grafana/grafana
-    container_name: grafana
-    ports:
-      - 3000:3000
-    restart: unless-stopped
-    environment:
-      - GF_SECURITY_ADMIN_USER=admin
-      - GF_SECURITY_ADMIN_PASSWORD=grafana
-    volumes:
-      - ./grafana:/etc/grafana/provisioning/datasources
-volumes:
-  prom_data:
-  localai_workspace:

.devcontainer/grafana/datasource.yml

@@ -1,10 +0,0 @@
-
-apiVersion: 1
-
-datasources:
-- name: Prometheus
-  type: prometheus
-  url: http://prometheus:9090 
-  isDefault: true
-  access: proxy
-  editable: true

.devcontainer/prometheus/prometheus.yml

@@ -1,21 +0,0 @@
-global:
-  scrape_interval: 15s
-  scrape_timeout: 10s
-  evaluation_interval: 15s
-alerting:
-  alertmanagers:
-  - static_configs:
-    - targets: []
-    scheme: http
-    timeout: 10s
-    api_version: v1
-scrape_configs:
-- job_name: prometheus
-  honor_timestamps: true
-  scrape_interval: 15s
-  scrape_timeout: 10s
-  metrics_path: /metrics
-  scheme: http
-  static_configs:
-  - targets:
-    - localhost:9090

.dockerignore

@@ -1,7 +1,6 @@
 .idea
 .github
 .vscode
-.devcontainer
 models
 examples/chatbot-ui/models
 examples/rwkv/models

.env

@@ -79,9 +79,6 @@
 ### Enable to run parallel requests
 # LOCALAI_PARALLEL_REQUESTS=true
 
-# Enable to allow p2p mode
-# LOCALAI_P2P=true
-
 ### Watchdog settings
 ###
 # Enables watchdog to kill backends that are inactive for too much time

.gitattributes

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

.github/bump_deps.sh

@@ -6,17 +6,4 @@ VAR=$3
 
 LAST_COMMIT=$(curl -s -H "Accept: application/vnd.github.VERSION.sha" "https://api.github.com/repos/$REPO/commits/$BRANCH")
 
-# Read $VAR from Makefile (only first match)
-set +e
-CURRENT_COMMIT="$(grep -m1 "^$VAR?=" Makefile | cut -d'=' -f2)"
-set -e
-
 sed -i Makefile -e "s/$VAR?=.*/$VAR?=$LAST_COMMIT/"
-
-if [ -z "$CURRENT_COMMIT" ]; then
-    echo "Could not find $VAR in Makefile."
-    exit 0
-fi
-
-echo "Changes: https://github.com/$REPO/compare/${CURRENT_COMMIT}..${LAST_COMMIT}" >> "${VAR}_message.txt"
-echo "${LAST_COMMIT}" >> "${VAR}_commit.txt"

.github/check_and_update.py

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

.github/ci/modelslist.go

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

.github/dependabot.yml

@@ -9,8 +9,6 @@ updates:
     directory: "/"
     schedule:
       interval: "weekly"
-    ignore:
-    - dependency-name: "github.com/mudler/LocalAI/pkg/grpc/proto"
   - package-ecosystem: "github-actions"
     # Workflow files stored in the default location of `.github/workflows`. (You don't need to specify `/.github/workflows` for `directory`. You can use `directory: "/"`.)
     directory: "/"
@@ -69,6 +67,10 @@ updates:
     directory: "/backend/python/parler-tts"
     schedule:
       interval: "weekly"
+  - package-ecosystem: "pip"
+    directory: "/backend/python/petals"
+    schedule:
+      interval: "weekly"
   - package-ecosystem: "pip"
     directory: "/backend/python/rerankers"
     schedule:

.github/labeler.yml

@@ -1,11 +1,6 @@
 enhancements:
  - head-branch: ['^feature', 'feature']
 
-dependencies:
-- any:
-  - changed-files:
-    - any-glob-to-any-file: 'Makefile'
-
 kind/documentation:
 - any:
   - changed-files:

.github/workflows/bump_deps.yaml

@@ -12,14 +12,23 @@ jobs:
           - repository: "ggerganov/llama.cpp"
             variable: "CPPLLAMA_VERSION"
             branch: "master"
+          - repository: "go-skynet/go-ggml-transformers.cpp"
+            variable: "GOGGMLTRANSFORMERS_VERSION"
+            branch: "master"
+          - repository: "donomii/go-rwkv.cpp"
+            variable: "RWKV_VERSION"
+            branch: "main"
           - repository: "ggerganov/whisper.cpp"
             variable: "WHISPER_CPP_VERSION"
             branch: "master"
-          - repository: "PABannier/bark.cpp"
-            variable: "BARKCPP_VERSION"
+          - repository: "go-skynet/go-bert.cpp"
+            variable: "BERT_VERSION"
+            branch: "master"
+          - repository: "go-skynet/bloomz.cpp"
+            variable: "BLOOMZ_VERSION"
             branch: "main"
-          - repository: "leejet/stable-diffusion.cpp"
-            variable: "STABLEDIFFUSION_GGML_VERSION"
+          - repository: "mudler/go-ggllm.cpp"
+            variable: "GOGGLLM_VERSION"
             branch: "master"
           - repository: "mudler/go-stable-diffusion"
             variable: "STABLEDIFFUSION_VERSION"
@@ -31,30 +40,17 @@ jobs:
     steps:
       - uses: actions/checkout@v4
       - name: Bump dependencies 🔧
-        id: bump
         run: |
           bash .github/bump_deps.sh ${{ matrix.repository }} ${{ matrix.branch }} ${{ matrix.variable }}
-          {
-            echo 'message<<EOF'
-            cat "${{ matrix.variable }}_message.txt"
-            echo EOF
-          } >> "$GITHUB_OUTPUT"
-          {
-            echo 'commit<<EOF'
-            cat "${{ matrix.variable }}_commit.txt"
-            echo EOF
-          } >> "$GITHUB_OUTPUT"
-          rm -rfv ${{ matrix.variable }}_message.txt
-          rm -rfv ${{ matrix.variable }}_commit.txt
       - name: Create Pull Request
-        uses: peter-evans/create-pull-request@v7
+        uses: peter-evans/create-pull-request@v6
         with:
           token: ${{ secrets.UPDATE_BOT_TOKEN }}
           push-to-fork: ci-forks/LocalAI
           commit-message: ':arrow_up: Update ${{ matrix.repository }}'
-          title: 'chore: :arrow_up: Update ${{ matrix.repository }} to `${{ steps.bump.outputs.commit }}`'
+          title: 'chore: :arrow_up: Update ${{ matrix.repository }}'
           branch: "update/${{ matrix.variable }}"
-          body: ${{ steps.bump.outputs.message }}
+          body: Bump of ${{ matrix.repository }} version
           signoff: true
 
 

.github/workflows/bump_docs.yaml

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

.github/workflows/checksum_checker.yaml

@@ -23,7 +23,7 @@ jobs:
           sudo pip install --upgrade pip
           pip install huggingface_hub
       - name: 'Setup yq'
-        uses: dcarbone/install-yq-action@v1.3.1
+        uses: dcarbone/install-yq-action@v1.1.1
         with:
           version: 'v4.44.2'
           download-compressed: true
@@ -36,12 +36,12 @@ jobs:
           sudo chmod 777 /hf_cache
           bash .github/checksum_checker.sh gallery/index.yaml
       - name: Create Pull Request
-        uses: peter-evans/create-pull-request@v7
+        uses: peter-evans/create-pull-request@v6
         with:
           token: ${{ secrets.UPDATE_BOT_TOKEN }}
           push-to-fork: ci-forks/LocalAI
           commit-message: ':arrow_up: Checksum updates in gallery/index.yaml'
-          title: 'chore(model-gallery): :arrow_up: update checksum'
+          title: 'models(gallery): :arrow_up: update checksum'
           branch: "update/checksum"
           body: Updating checksums in gallery/index.yaml
           signoff: true

.github/workflows/deploy-explorer.yaml

@@ -1,64 +0,0 @@
-name: Explorer deployment
-
-on:
-  push:
-    branches:
-      - master
-    tags:
-      - 'v*'
-
-concurrency:
-  group: ci-deploy-${{ github.head_ref || github.ref }}-${{ github.repository }}
-
-jobs:
-  build-linux:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Clone
-        uses: actions/checkout@v4
-        with:
-          submodules: true
-      - uses: actions/setup-go@v5
-        with:
-          go-version: '1.21.x'
-          cache: false
-      - name: Dependencies
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y wget curl build-essential ffmpeg protobuf-compiler ccache upx-ucl gawk cmake libgmock-dev
-          go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
-          go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
-          make protogen-go
-      - name: Build api
-        run: |
-          CGO_ENABLED=0 make build-api
-      - name: rm
-        uses: appleboy/ssh-action@v1.2.0
-        with:
-            host: ${{ secrets.EXPLORER_SSH_HOST }}
-            username: ${{ secrets.EXPLORER_SSH_USERNAME }}
-            key: ${{ secrets.EXPLORER_SSH_KEY }}
-            port: ${{ secrets.EXPLORER_SSH_PORT }}
-            script: |
-                sudo rm -rf local-ai/ || true
-      - name: copy file via ssh
-        uses: appleboy/scp-action@v0.1.7
-        with:
-            host: ${{ secrets.EXPLORER_SSH_HOST }}
-            username: ${{ secrets.EXPLORER_SSH_USERNAME }}
-            key: ${{ secrets.EXPLORER_SSH_KEY }}
-            port: ${{ secrets.EXPLORER_SSH_PORT }}
-            source: "local-ai"
-            overwrite: true
-            rm: true
-            target: ./local-ai
-      - name: restarting
-        uses: appleboy/ssh-action@v1.2.0
-        with:
-            host: ${{ secrets.EXPLORER_SSH_HOST }}
-            username: ${{ secrets.EXPLORER_SSH_USERNAME }}
-            key: ${{ secrets.EXPLORER_SSH_KEY }}
-            port: ${{ secrets.EXPLORER_SSH_PORT }}
-            script: |
-                sudo cp -rfv local-ai/local-ai /usr/bin/local-ai
-                sudo systemctl restart local-ai

.github/workflows/generate_intel_image.yaml

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

.github/workflows/image-pr.yml

@@ -47,7 +47,7 @@ jobs:
           #   makeflags: "--jobs=3 --output-sync=target"
           - build-type: 'cublas'
             cuda-major-version: "12"
-            cuda-minor-version: "0"
+            cuda-minor-version: "4"
             platforms: 'linux/amd64'
             tag-latest: 'false'
             tag-suffix: '-cublas-cuda12-ffmpeg'
@@ -120,7 +120,7 @@ jobs:
           #   makeflags: "--jobs=3 --output-sync=target"
           # - build-type: 'cublas'
           #   cuda-major-version: "12"
-          #   cuda-minor-version: "0"
+          #   cuda-minor-version: "4"
           #   platforms: 'linux/amd64'
           #   tag-latest: 'false'
           #   tag-suffix: '-cublas-cuda12-ffmpeg-core'

.github/workflows/image.yml

@@ -13,78 +13,6 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
-  hipblas-jobs:
-    uses: ./.github/workflows/image_build.yml
-    with:
-      tag-latest: ${{ matrix.tag-latest }}
-      tag-suffix: ${{ matrix.tag-suffix }}
-      ffmpeg: ${{ matrix.ffmpeg }}
-      image-type: ${{ matrix.image-type }}
-      build-type: ${{ matrix.build-type }}
-      cuda-major-version: ${{ matrix.cuda-major-version }}
-      cuda-minor-version: ${{ matrix.cuda-minor-version }}
-      platforms: ${{ matrix.platforms }}
-      runs-on: ${{ matrix.runs-on }}
-      base-image: ${{ matrix.base-image }}
-      grpc-base-image: ${{ matrix.grpc-base-image }}
-      aio: ${{ matrix.aio }}
-      makeflags: ${{ matrix.makeflags }}
-      latest-image: ${{ matrix.latest-image }}
-      latest-image-aio: ${{ matrix.latest-image-aio }}
-    secrets:
-      dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
-      dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
-      quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
-      quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
-    strategy:
-      # Pushing with all jobs in parallel
-      # eats the bandwidth of all the nodes
-      max-parallel: 2
-      matrix:
-        include:
-          - build-type: 'hipblas'
-            platforms: 'linux/amd64'
-            tag-latest: 'auto'
-            tag-suffix: '-hipblas-ffmpeg'
-            ffmpeg: 'true'
-            image-type: 'extras'
-            aio: "-aio-gpu-hipblas"
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
-            grpc-base-image: "ubuntu:22.04"
-            latest-image: 'latest-gpu-hipblas'
-            latest-image-aio: 'latest-aio-gpu-hipblas'
-            runs-on: 'arc-runner-set'
-            makeflags: "--jobs=3 --output-sync=target"
-          - build-type: 'hipblas'
-            platforms: 'linux/amd64'
-            tag-latest: 'false'
-            tag-suffix: '-hipblas'
-            ffmpeg: 'false'
-            image-type: 'extras'
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
-            grpc-base-image: "ubuntu:22.04"
-            runs-on: 'arc-runner-set'
-            makeflags: "--jobs=3 --output-sync=target"
-          - build-type: 'hipblas'
-            platforms: 'linux/amd64'
-            tag-latest: 'false'
-            tag-suffix: '-hipblas-ffmpeg-core'
-            ffmpeg: 'true'
-            image-type: 'core'
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
-            grpc-base-image: "ubuntu:22.04"
-            runs-on: 'arc-runner-set'
-            makeflags: "--jobs=3 --output-sync=target"
-          - build-type: 'hipblas'
-            platforms: 'linux/amd64'
-            tag-latest: 'false'
-            tag-suffix: '-hipblas-core'
-            ffmpeg: 'false'
-            image-type: 'core'
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
-            grpc-base-image: "ubuntu:22.04"
-            runs-on: 'arc-runner-set'
-            makeflags: "--jobs=3 --output-sync=target"
   self-hosted-jobs:
     uses: ./.github/workflows/image_build.yml
     with:
@@ -111,7 +39,7 @@ jobs:
     strategy:
       # Pushing with all jobs in parallel
       # eats the bandwidth of all the nodes
-      max-parallel: ${{ github.event_name != 'pull_request' && 5 || 8 }}
+      max-parallel: ${{ github.event_name != 'pull_request' && 6 || 10 }}
       matrix:
         include:
           # Extra images
@@ -147,7 +75,7 @@ jobs:
             makeflags: "--jobs=3 --output-sync=target"
           - build-type: 'cublas'
             cuda-major-version: "12"
-            cuda-minor-version: "0"
+            cuda-minor-version: "4"
             platforms: 'linux/amd64'
             tag-latest: 'false'
             tag-suffix: '-cublas-cuda12'
@@ -172,7 +100,7 @@ jobs:
             makeflags: "--jobs=3 --output-sync=target"
           - build-type: 'cublas'
             cuda-major-version: "12"
-            cuda-minor-version: "0"
+            cuda-minor-version: "4"
             platforms: 'linux/amd64'
             tag-latest: 'auto'
             tag-suffix: '-cublas-cuda12-ffmpeg'
@@ -194,6 +122,29 @@ jobs:
             base-image: "ubuntu:22.04"
             runs-on: 'arc-runner-set'
             makeflags: "--jobs=3 --output-sync=target"
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'auto'
+            tag-suffix: '-hipblas-ffmpeg'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            aio: "-aio-gpu-hipblas"
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            grpc-base-image: "ubuntu:22.04"
+            latest-image: 'latest-gpu-hipblas'
+            latest-image-aio: 'latest-aio-gpu-hipblas'
+            runs-on: 'arc-runner-set'
+            makeflags: "--jobs=3 --output-sync=target"
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas'
+            ffmpeg: 'false'
+            image-type: 'extras'
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            grpc-base-image: "ubuntu:22.04"
+            runs-on: 'arc-runner-set'
+            makeflags: "--jobs=3 --output-sync=target"
           - build-type: 'sycl_f16'
             platforms: 'linux/amd64'
             tag-latest: 'auto'
@@ -261,6 +212,26 @@ jobs:
             image-type: 'core'
             runs-on: 'arc-runner-set'
             makeflags: "--jobs=3 --output-sync=target"
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas-ffmpeg-core'
+            ffmpeg: 'true'
+            image-type: 'core'
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            grpc-base-image: "ubuntu:22.04"
+            runs-on: 'arc-runner-set'
+            makeflags: "--jobs=3 --output-sync=target"
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas-core'
+            ffmpeg: 'false'
+            image-type: 'core'
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            grpc-base-image: "ubuntu:22.04"
+            runs-on: 'arc-runner-set'
+            makeflags: "--jobs=3 --output-sync=target"
 
   core-image-build:
     uses: ./.github/workflows/image_build.yml
@@ -314,7 +285,7 @@ jobs:
             makeflags: "--jobs=4 --output-sync=target"
           - build-type: 'cublas'
             cuda-major-version: "12"
-            cuda-minor-version: "0"
+            cuda-minor-version: "4"
             platforms: 'linux/amd64'
             tag-latest: 'false'
             tag-suffix: '-cublas-cuda12-core'
@@ -336,7 +307,7 @@ jobs:
             makeflags: "--jobs=4 --output-sync=target"
           - build-type: 'cublas'
             cuda-major-version: "12"
-            cuda-minor-version: "0"
+            cuda-minor-version: "4"
             platforms: 'linux/amd64'
             tag-latest: 'false'
             tag-suffix: '-cublas-cuda12-ffmpeg-core'

.github/workflows/notify-models.yaml

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

.github/workflows/release.yaml

@@ -31,10 +31,11 @@ jobs:
         with:
           go-version: '1.21.x'
           cache: false
+
       - name: Dependencies
         run: |
           sudo apt-get update
-          sudo apt-get install build-essential ffmpeg protobuf-compiler ccache upx-ucl gawk
+          sudo apt-get install build-essential ffmpeg protobuf-compiler ccache gawk
           sudo apt-get install -qy binutils-aarch64-linux-gnu gcc-aarch64-linux-gnu g++-aarch64-linux-gnu libgmock-dev
       - name: Install CUDA Dependencies
         run: |
@@ -123,7 +124,7 @@ jobs:
             release/*
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.19
+        uses: mxschmitt/action-tmate@v3.18
         with:
           detached: true
           connect-timeout-seconds: 180
@@ -150,7 +151,7 @@ jobs:
       - name: Dependencies
         run: |
           sudo apt-get update
-          sudo apt-get install -y wget curl build-essential ffmpeg protobuf-compiler ccache upx-ucl gawk cmake libgmock-dev
+          sudo apt-get install -y wget curl build-essential ffmpeg protobuf-compiler ccache gawk cmake libgmock-dev
       - name: Intel Dependencies
         run: |
           wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB | gpg --dearmor | sudo tee /usr/share/keyrings/oneapi-archive-keyring.gpg > /dev/null
@@ -232,7 +233,7 @@ jobs:
             release/*
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.19
+        uses: mxschmitt/action-tmate@v3.18
         with:
           detached: true
           connect-timeout-seconds: 180
@@ -251,7 +252,7 @@ jobs:
       - name: Dependencies
         run: |
           sudo apt-get update
-          sudo apt-get install -y --no-install-recommends libopencv-dev protobuf-compiler ccache upx-ucl
+          sudo apt-get install -y --no-install-recommends libopencv-dev protobuf-compiler ccache
           go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
           go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
       - name: Build stablediffusion
@@ -294,7 +295,7 @@ jobs:
           export C_INCLUDE_PATH=/usr/local/include
           export CPLUS_INCLUDE_PATH=/usr/local/include
           export PATH=$PATH:$GOPATH/bin
-          export SKIP_GRPC_BACKEND=backend-assets/grpc/whisper
+
           make dist
       - uses: actions/upload-artifact@v4
         with:
@@ -308,7 +309,7 @@ jobs:
             release/*
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.19
+        uses: mxschmitt/action-tmate@v3.18
         with:
           detached: true
           connect-timeout-seconds: 180
@@ -327,7 +328,7 @@ jobs:
           cache: false
       - name: Dependencies
         run: |
-          brew install protobuf grpc libomp llvm
+          brew install protobuf grpc
           go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
           go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
       - name: Build
@@ -336,7 +337,7 @@ jobs:
           export C_INCLUDE_PATH=/usr/local/include
           export CPLUS_INCLUDE_PATH=/usr/local/include
           export PATH=$PATH:$GOPATH/bin
-          export CC=/opt/homebrew/opt/llvm/bin/clang
+
           make dist
       - uses: actions/upload-artifact@v4
         with:
@@ -350,7 +351,7 @@ jobs:
             release/*
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.19
+        uses: mxschmitt/action-tmate@v3.18
         with:
           detached: true
           connect-timeout-seconds: 180

.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@v2.21.4
+        uses: securego/gosec@master
         with:
           # we let the report trigger content trigger a failure using the GitHub Security features.
           args: '-no-fail -fmt sarif -out results.sarif ./...'

.github/workflows/test-extra.yml

@@ -123,13 +123,6 @@ jobs:
         run: |
            make --jobs=5 --output-sync=target -C backend/python/parler-tts
            make --jobs=5 --output-sync=target -C backend/python/parler-tts test
-      - name: Setup tmate session if tests fail
-        if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.19
-        with:
-          detached: true
-          connect-timeout-seconds: 180
-          limit-access-to-actor: true
 
   tests-openvoice:
     runs-on: ubuntu-latest
@@ -175,6 +168,32 @@ jobs:
            make --jobs=5 --output-sync=target -C backend/python/transformers-musicgen
            make --jobs=5 --output-sync=target -C backend/python/transformers-musicgen test
 
+
+
+  # tests-petals:
+  #   runs-on: ubuntu-latest
+  #   steps:
+  #     - name: Clone
+  #       uses: actions/checkout@v4
+  #       with:
+  #         submodules: true
+  #     - name: Dependencies
+  #       run: |
+  #         sudo apt-get update
+  #         sudo apt-get install build-essential ffmpeg
+  #         # Install UV
+  #         curl -LsSf https://astral.sh/uv/install.sh | sh
+  #         sudo apt-get install -y ca-certificates cmake curl patch python3-pip
+  #         sudo apt-get install -y libopencv-dev
+  #         pip install --user --no-cache-dir grpcio-tools==1.64.1
+
+  #     - name: Test petals
+  #       run: |
+  #          make --jobs=5 --output-sync=target -C backend/python/petals
+  #          make --jobs=5 --output-sync=target -C backend/python/petals test
+
+
+
   # tests-bark:
   #   runs-on: ubuntu-latest
   #   steps:

.github/workflows/test.yml

@@ -70,7 +70,7 @@ jobs:
       - name: Dependencies
         run: |
           sudo apt-get update
-          sudo apt-get install build-essential ccache upx-ucl curl ffmpeg
+          sudo apt-get install build-essential curl ffmpeg
           sudo apt-get install -y libgmock-dev
           curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmor > conda.gpg && \
              sudo install -o root -g root -m 644 conda.gpg /usr/share/keyrings/conda-archive-keyring.gpg && \
@@ -133,7 +133,7 @@ jobs:
           PATH="$PATH:/root/go/bin" GO_TAGS="stablediffusion tts" make --jobs 5 --output-sync=target test
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.19
+        uses: mxschmitt/action-tmate@v3.18
         with:
           detached: true
           connect-timeout-seconds: 180
@@ -178,26 +178,17 @@ jobs:
         uses: actions/checkout@v4
         with:
           submodules: true
-      - name: Dependencies
-        run: |
-          # Install protoc
-          curl -L -s https://github.com/protocolbuffers/protobuf/releases/download/v26.1/protoc-26.1-linux-x86_64.zip -o protoc.zip && \
-          unzip -j -d /usr/local/bin protoc.zip bin/protoc && \
-          rm protoc.zip
-          go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
-          go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
-          PATH="$PATH:$HOME/go/bin" make protogen-go
       - name: Build images
         run: |
           docker build --build-arg FFMPEG=true --build-arg IMAGE_TYPE=extras --build-arg EXTRA_BACKENDS=rerankers --build-arg MAKEFLAGS="--jobs=5 --output-sync=target" -t local-ai:tests -f Dockerfile .
           BASE_IMAGE=local-ai:tests DOCKER_AIO_IMAGE=local-ai-aio:test make docker-aio
       - name: Test
         run: |
-            PATH="$PATH:$HOME/go/bin" LOCALAI_MODELS_DIR=$PWD/models LOCALAI_IMAGE_TAG=test LOCALAI_IMAGE=local-ai-aio \
+          LOCALAI_MODELS_DIR=$PWD/models LOCALAI_IMAGE_TAG=test LOCALAI_IMAGE=local-ai-aio \
             make run-e2e-aio
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.19
+        uses: mxschmitt/action-tmate@v3.18
         with:
           detached: true
           connect-timeout-seconds: 180
@@ -223,19 +214,18 @@ jobs:
         run: go version
       - name: Dependencies
         run: |
-          brew install protobuf grpc make protoc-gen-go protoc-gen-go-grpc libomp llvm
-          pip install --user --no-cache-dir grpcio-tools
+          brew install protobuf grpc make protoc-gen-go protoc-gen-go-grpc
+          pip install --user --no-cache-dir grpcio-tools==1.64.1
       - name: Test
         run: |
           export C_INCLUDE_PATH=/usr/local/include
           export CPLUS_INCLUDE_PATH=/usr/local/include
-          export CC=/opt/homebrew/opt/llvm/bin/clang
           # Used to run the newer GNUMake version from brew that supports --output-sync
           export PATH="/opt/homebrew/opt/make/libexec/gnubin:$PATH"
           BUILD_TYPE="GITHUB_CI_HAS_BROKEN_METAL" CMAKE_ARGS="-DGGML_F16C=OFF -DGGML_AVX512=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF" make --jobs 4 --output-sync=target test
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
-        uses: mxschmitt/action-tmate@v3.19
+        uses: mxschmitt/action-tmate@v3.18
         with:
           detached: true
           connect-timeout-seconds: 180

.github/workflows/update_swagger.yaml

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

.gitignore

@@ -2,7 +2,6 @@
 /sources/
 __pycache__/
 *.a
-*.o
 get-sources
 prepare-sources
 /backend/cpp/llama/grpc-server
@@ -13,6 +12,7 @@ prepare-sources
 
 go-ggml-transformers
 go-gpt2
+go-rwkv
 whisper.cpp
 /bloomz
 go-bert
@@ -54,6 +54,3 @@ docs/static/gallery.html
 
 # backend virtual environments
 **/venv
-
-# per-developer customization files for the development container
-.devcontainer/customization/*

.vscode/launch.json

@@ -3,12 +3,12 @@
     "configurations": [
         {
             "name": "Python: Current File",
-            "type": "debugpy",
+            "type": "python",
             "request": "launch",
             "program": "${file}",
             "console": "integratedTerminal",
             "justMyCode": false,
-            "cwd": "${fileDirname}",
+            "cwd": "${workspaceFolder}/examples/langchain-chroma",
             "env": {
                 "OPENAI_API_BASE": "http://localhost:8080/v1",
                 "OPENAI_API_KEY": "abc"
@@ -19,16 +19,15 @@
             "type": "go",
             "request": "launch",
             "mode": "debug",
-            "program": "${workspaceRoot}",
-            "args": [],
+            "program": "${workspaceFolder}/main.go",
+            "args": [
+                "api"
+            ],
             "env": {
-                "LOCALAI_LOG_LEVEL": "debug",
-                "LOCALAI_P2P": "true",
-                "LOCALAI_FEDERATED": "true"
-            },
-            "buildFlags": ["-tags", "stablediffusion p2p tts", "-v"],
-            "envFile": "${workspaceFolder}/.env",
-            "cwd": "${workspaceRoot}"
+                "C_INCLUDE_PATH": "${workspaceFolder}/go-llama:${workspaceFolder}/go-stable-diffusion/:${workspaceFolder}/gpt4all/gpt4all-bindings/golang/:${workspaceFolder}/go-gpt2:${workspaceFolder}/go-rwkv:${workspaceFolder}/whisper.cpp:${workspaceFolder}/go-bert:${workspaceFolder}/bloomz",
+                "LIBRARY_PATH": "${workspaceFolder}/go-llama:${workspaceFolder}/go-stable-diffusion/:${workspaceFolder}/gpt4all/gpt4all-bindings/golang/:${workspaceFolder}/go-gpt2:${workspaceFolder}/go-rwkv:${workspaceFolder}/whisper.cpp:${workspaceFolder}/go-bert:${workspaceFolder}/bloomz",
+                "DEBUG": "true"
+            }
         }
     ]
 }

CONTRIBUTING.md

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

Dockerfile

@@ -8,14 +8,12 @@ FROM ${BASE_IMAGE} AS requirements-core
 
 USER root
 
-ARG GO_VERSION=1.22.6
-ARG CMAKE_VERSION=3.26.4
-ARG CMAKE_FROM_SOURCE=false
+ARG GO_VERSION=1.22.5
 ARG TARGETARCH
 ARG TARGETVARIANT
 
 ENV DEBIAN_FRONTEND=noninteractive
-ENV EXTERNAL_GRPC_BACKENDS="coqui:/build/backend/python/coqui/run.sh,huggingface-embeddings:/build/backend/python/sentencetransformers/run.sh,transformers:/build/backend/python/transformers/run.sh,sentencetransformers:/build/backend/python/sentencetransformers/run.sh,rerankers:/build/backend/python/rerankers/run.sh,autogptq:/build/backend/python/autogptq/run.sh,bark:/build/backend/python/bark/run.sh,diffusers:/build/backend/python/diffusers/run.sh,openvoice:/build/backend/python/openvoice/run.sh,vall-e-x:/build/backend/python/vall-e-x/run.sh,vllm:/build/backend/python/vllm/run.sh,mamba:/build/backend/python/mamba/run.sh,exllama2:/build/backend/python/exllama2/run.sh,transformers-musicgen:/build/backend/python/transformers-musicgen/run.sh,parler-tts:/build/backend/python/parler-tts/run.sh"
+ENV EXTERNAL_GRPC_BACKENDS="coqui:/build/backend/python/coqui/run.sh,huggingface-embeddings:/build/backend/python/sentencetransformers/run.sh,petals:/build/backend/python/petals/run.sh,transformers:/build/backend/python/transformers/run.sh,sentencetransformers:/build/backend/python/sentencetransformers/run.sh,rerankers:/build/backend/python/rerankers/run.sh,autogptq:/build/backend/python/autogptq/run.sh,bark:/build/backend/python/bark/run.sh,diffusers:/build/backend/python/diffusers/run.sh,exllama:/build/backend/python/exllama/run.sh,openvoice:/build/backend/python/openvoice/run.sh,vall-e-x:/build/backend/python/vall-e-x/run.sh,vllm:/build/backend/python/vllm/run.sh,mamba:/build/backend/python/mamba/run.sh,exllama2:/build/backend/python/exllama2/run.sh,transformers-musicgen:/build/backend/python/transformers-musicgen/run.sh,parler-tts:/build/backend/python/parler-tts/run.sh"
 
 
 RUN apt-get update && \
@@ -23,28 +21,16 @@ RUN apt-get update && \
         build-essential \
         ccache \
         ca-certificates \
-        curl libssl-dev \
+        cmake \
+        curl \
         git \
-        unzip upx-ucl && \
+        unzip && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
-# Install CMake (the version in 22.04 is too old)
-RUN <<EOT bash
-    if [ "${CMAKE_FROM_SOURCE}}" = "true" ]; then
-        curl -L -s https://github.com/Kitware/CMake/releases/download/v${CMAKE_VERSION}/cmake-${CMAKE_VERSION}.tar.gz -o cmake.tar.gz && tar xvf cmake.tar.gz && cd cmake-${CMAKE_VERSION} && ./configure && make && make install
-    else
-        apt-get update && \
-        apt-get install -y \
-            cmake && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/*
-    fi
-EOT
-
 # Install Go
 RUN curl -L -s https://go.dev/dl/go${GO_VERSION}.linux-${TARGETARCH}.tar.gz | tar -C /usr/local -xz
-ENV PATH=$PATH:/root/go/bin:/usr/local/go/bin
+ENV PATH $PATH:/root/go/bin:/usr/local/go/bin
 
 # Install grpc compilers
 RUN go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2 && \
@@ -53,18 +39,15 @@ RUN go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2 && \
 COPY --chmod=644 custom-ca-certs/* /usr/local/share/ca-certificates/
 RUN update-ca-certificates
 
-RUN test -n "$TARGETARCH" \
-    || (echo 'warn: missing $TARGETARCH, either set this `ARG` manually, or run using `docker buildkit`')
-
 # Use the variables in subsequent instructions
 RUN echo "Target Architecture: $TARGETARCH"
 RUN echo "Target Variant: $TARGETVARIANT"
 
 # Cuda
-ENV PATH=/usr/local/cuda/bin:${PATH}
+ENV PATH /usr/local/cuda/bin:${PATH}
 
 # HipBLAS requirements
-ENV PATH=/opt/rocm/bin:${PATH}
+ENV PATH /opt/rocm/bin:${PATH}
 
 # OpenBLAS requirements and stable diffusion
 RUN apt-get update && \
@@ -79,14 +62,16 @@ RUN ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2
 
 WORKDIR /build
 
+RUN test -n "$TARGETARCH" \
+    || (echo 'warn: missing $TARGETARCH, either set this `ARG` manually, or run using `docker buildkit`')
+
 ###################################
 ###################################
 
 # The requirements-extras target is for any builds with IMAGE_TYPE=extras. It should not be placed in this target unless every IMAGE_TYPE=extras build will use it
 FROM requirements-core AS requirements-extras
 
-# Install uv as a system package
-RUN curl -LsSf https://astral.sh/uv/install.sh | UV_INSTALL_DIR=/usr/bin sh
+RUN curl -LsSf https://astral.sh/uv/install.sh | sh
 ENV PATH="/root/.cargo/bin:${PATH}"
 
 RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
@@ -96,7 +81,7 @@ RUN apt-get update && \
         espeak \
         python3-pip \
         python-is-python3 \
-        python3-dev llvm \
+        python3-dev \
         python3-venv && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/* && \
@@ -114,7 +99,7 @@ FROM requirements-${IMAGE_TYPE} AS requirements-drivers
 
 ARG BUILD_TYPE
 ARG CUDA_MAJOR_VERSION=12
-ARG CUDA_MINOR_VERSION=0
+ARG CUDA_MINOR_VERSION=4
 
 ENV BUILD_TYPE=${BUILD_TYPE}
 
@@ -203,8 +188,6 @@ FROM ${GRPC_BASE_IMAGE} AS grpc
 # This is a bit of a hack, but it's required in order to be able to effectively cache this layer in CI
 ARG GRPC_MAKEFLAGS="-j4 -Otarget"
 ARG GRPC_VERSION=v1.65.0
-ARG CMAKE_FROM_SOURCE=false
-ARG CMAKE_VERSION=3.26.4
 
 ENV MAKEFLAGS=${GRPC_MAKEFLAGS}
 
@@ -213,24 +196,12 @@ WORKDIR /build
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
         ca-certificates \
-        build-essential curl libssl-dev \
+        build-essential \
+        cmake \
         git && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
-# Install CMake (the version in 22.04 is too old)
-RUN <<EOT bash
-    if [ "${CMAKE_FROM_SOURCE}}" = "true" ]; then
-        curl -L -s https://github.com/Kitware/CMake/releases/download/v${CMAKE_VERSION}/cmake-${CMAKE_VERSION}.tar.gz -o cmake.tar.gz && tar xvf cmake.tar.gz && cd cmake-${CMAKE_VERSION} && ./configure && make && make install
-    else
-        apt-get update && \
-        apt-get install -y \
-            cmake && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/*
-    fi
-EOT
-
 # We install GRPC to a different prefix here so that we can copy in only the build artifacts later
 # saves several hundred MB on the final docker image size vs copying in the entire GRPC source tree
 # and running make install in the target container
@@ -246,14 +217,13 @@ RUN git clone --recurse-submodules --jobs 4 -b ${GRPC_VERSION} --depth 1 --shall
 ###################################
 ###################################
 
-# The builder-base target has the arguments, variables, and copies shared between full builder images and the uncompiled devcontainer
-
-FROM requirements-drivers AS builder-base
+# The builder target compiles LocalAI. This target is not the target that will be uploaded to the registry.
+# Adjustments to the build process should likely be made here.
+FROM requirements-drivers AS builder
 
 ARG GO_TAGS="stablediffusion tts p2p"
 ARG GRPC_BACKENDS
 ARG MAKEFLAGS
-ARG LD_FLAGS="-s -w"
 
 ENV GRPC_BACKENDS=${GRPC_BACKENDS}
 ENV GO_TAGS=${GO_TAGS}
@@ -261,12 +231,14 @@ ENV MAKEFLAGS=${MAKEFLAGS}
 ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
 ENV NVIDIA_REQUIRE_CUDA="cuda>=${CUDA_MAJOR_VERSION}.0"
 ENV NVIDIA_VISIBLE_DEVICES=all
-ENV LD_FLAGS=${LD_FLAGS}
-
-RUN echo "GO_TAGS: $GO_TAGS" && echo "TARGETARCH: $TARGETARCH"
 
 WORKDIR /build
 
+COPY . .
+COPY .git .
+RUN echo "GO_TAGS: $GO_TAGS"
+
+RUN make prepare
 
 # We need protoc installed, and the version in 22.04 is too old.  We will create one as part installing the GRPC build below
 # but that will also being in a newer version of absl which stablediffusion cannot compile with.  This version of protoc is only
@@ -284,35 +256,8 @@ RUN <<EOT bash
     fi
 EOT
 
-
-###################################
-###################################
-
-# This first portion of builder holds the layers specifically used to build backend-assets/grpc/stablediffusion
-# In most cases, builder is the image you should be using - however, this can save build time if one just needs to copy backend-assets/grpc/stablediffusion and nothing else.
-FROM builder-base AS builder-sd
-
-# stablediffusion does not tolerate a newer version of abseil, copy only over enough elements to build it
-COPY Makefile .
-COPY go.mod .
-COPY go.sum .
-COPY backend/backend.proto ./backend/backend.proto
-COPY backend/go/image/stablediffusion ./backend/go/image/stablediffusion
-COPY pkg/grpc ./pkg/grpc
-COPY pkg/stablediffusion ./pkg/stablediffusion
-RUN git init
-RUN make sources/go-stable-diffusion
-RUN touch prepare-sources
-
-# Actually build the backend
-RUN GRPC_BACKENDS=backend-assets/grpc/stablediffusion make backend-assets/grpc/stablediffusion
-
-###################################
-###################################
-
-# The builder target compiles LocalAI. This target is not the target that will be uploaded to the registry.
-# Adjustments to the build process should likely be made here.
-FROM builder-sd AS builder
+# stablediffusion does not tolerate a newer version of abseil, build it first
+RUN GRPC_BACKENDS=backend-assets/grpc/stablediffusion make build
 
 # Install the pre-built GRPC
 COPY --from=grpc /opt/grpc /usr/local
@@ -320,20 +265,8 @@ COPY --from=grpc /opt/grpc /usr/local
 # Rebuild with defaults backends
 WORKDIR /build
 
-COPY . .
-COPY .git .
-
-RUN make prepare
-
 ## Build the binary
-## If it's CUDA or hipblas, we want to skip some of the llama-compat backends to save space
-## We only leave the most CPU-optimized variant and the fallback for the cublas/hipblas build
-## (both will use CUDA or hipblas for the actual computation)
-RUN if [ "${BUILD_TYPE}" = "cublas" ] || [ "${BUILD_TYPE}" = "hipblas" ]; then \
-        SKIP_GRPC_BACKEND="backend-assets/grpc/llama-cpp-avx backend-assets/grpc/llama-cpp-avx2" make build; \
-    else \
-        make build; \
-    fi
+RUN make build
 
 RUN if [ ! -d "/build/sources/go-piper/piper-phonemize/pi/lib/" ]; then \
         mkdir -p /build/sources/go-piper/piper-phonemize/pi/lib/ \
@@ -343,40 +276,6 @@ RUN if [ ! -d "/build/sources/go-piper/piper-phonemize/pi/lib/" ]; then \
 ###################################
 ###################################
 
-# The devcontainer target is not used on CI. It is a target for developers to use locally -
-# rather than copying files it mounts them locally and leaves building to the developer
-
-FROM builder-base AS devcontainer
-
-ARG FFMPEG
-
-COPY --from=grpc /opt/grpc /usr/local
-
-COPY --from=builder-sd /build/backend-assets/grpc/stablediffusion /build/backend-assets/grpc/stablediffusion
-
-COPY .devcontainer-scripts /.devcontainer-scripts
-
-# Add FFmpeg
-RUN if [ "${FFMPEG}" = "true" ]; then \
-        apt-get update && \
-        apt-get install -y --no-install-recommends \
-            ffmpeg && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/* \
-    ; fi
-
-RUN apt-get update && \
-    apt-get install -y --no-install-recommends \
-        ssh less wget
-# For the devcontainer, leave apt functional in case additional devtools are needed at runtime.
-
-RUN go install github.com/go-delve/delve/cmd/dlv@latest
-
-RUN go install github.com/mikefarah/yq/v4@latest
-
-###################################
-###################################
-
 # This is the final target. The result of this target will be the image uploaded to the registry.
 # If you cannot find a more suitable place for an addition, this layer is a suitable place for it.
 FROM requirements-drivers
@@ -427,7 +326,7 @@ COPY --from=builder /build/local-ai ./
 COPY --from=builder /build/sources/go-piper/piper-phonemize/pi/lib/* /usr/lib/
 
 # do not let stablediffusion rebuild (requires an older version of absl)
-COPY --from=builder-sd /build/backend-assets/grpc/stablediffusion ./backend-assets/grpc/stablediffusion
+COPY --from=builder /build/backend-assets/grpc/stablediffusion ./backend-assets/grpc/stablediffusion
 
 # Change the shell to bash so we can use [[ tests below
 SHELL ["/bin/bash", "-c"]
@@ -446,6 +345,9 @@ RUN if [[ ( "${EXTRA_BACKENDS}" =~ "coqui" || -z "${EXTRA_BACKENDS}" ) && "$IMAG
     ; fi && \
     if [[ ( "${EXTRA_BACKENDS}" =~ "transformers-musicgen" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
         make -C backend/python/transformers-musicgen \
+    ; fi && \
+    if [[ ( "${EXTRA_BACKENDS}" =~ "exllama1" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
+        make -C backend/python/exllama \
     ; fi
 
 RUN if [[ ( "${EXTRA_BACKENDS}" =~ "vall-e-x" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
@@ -454,6 +356,9 @@ RUN if [[ ( "${EXTRA_BACKENDS}" =~ "vall-e-x" || -z "${EXTRA_BACKENDS}" ) && "$I
     if [[ ( "${EXTRA_BACKENDS}" =~ "openvoice" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
         make -C backend/python/openvoice \
     ; fi && \
+    if [[ ( "${EXTRA_BACKENDS}" =~ "petals" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
+        make -C backend/python/petals \
+    ; fi && \
     if [[ ( "${EXTRA_BACKENDS}" =~ "sentencetransformers" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
         make -C backend/python/sentencetransformers \
     ; fi && \

Makefile

@@ -8,15 +8,27 @@ DETECT_LIBS?=true
 # llama.cpp versions
 GOLLAMA_REPO?=https://github.com/go-skynet/go-llama.cpp
 GOLLAMA_VERSION?=2b57a8ae43e4699d3dc5d1496a1ccd42922993be
-CPPLLAMA_VERSION?=cc98896db858df7aa40d0e16a505883ef196a482
+CPPLLAMA_VERSION?=07283b1a90e1320aae4762c7e03c879043910252
+
+# gpt4all version
+GPT4ALL_REPO?=https://github.com/nomic-ai/gpt4all
+GPT4ALL_VERSION?=27a8b020c36b0df8f8b82a252d261cda47cf44b8
+
+# go-rwkv version
+RWKV_REPO?=https://github.com/donomii/go-rwkv.cpp
+RWKV_VERSION?=661e7ae26d442f5cfebd2a0881b44e8c55949ec6
 
 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggerganov/whisper.cpp
-WHISPER_CPP_VERSION?=6266a9f9e56a5b925e9892acf650f3eb1245814d
+WHISPER_CPP_VERSION?=f68298ce06ca3edd6e6f3f21c3d0bb5f073942c3
+
+# bert.cpp version
+BERT_REPO?=https://github.com/go-skynet/go-bert.cpp
+BERT_VERSION?=710044b124545415f555e4260d16b146c725a6e4
 
 # go-piper version
 PIPER_REPO?=https://github.com/mudler/go-piper
-PIPER_VERSION?=e10ca041a885d4a8f3871d52924b47792d5e5aa0
+PIPER_VERSION?=9d0100873a7dbb0824dfea40e8cec70a1b110759
 
 # stablediffusion version
 STABLEDIFFUSION_REPO?=https://github.com/mudler/go-stable-diffusion
@@ -26,18 +38,6 @@ STABLEDIFFUSION_VERSION?=4a3cd6aeae6f66ee57eae9a0075f8c58c3a6a38f
 TINYDREAM_REPO?=https://github.com/M0Rf30/go-tiny-dream
 TINYDREAM_VERSION?=c04fa463ace9d9a6464313aa5f9cd0f953b6c057
 
-# bark.cpp
-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?=4570715727f35e5a07a76796d823824c8f42206c
-
-ONNX_VERSION?=1.20.0
-ONNX_ARCH?=x64
-ONNX_OS?=linux
-
 export BUILD_TYPE?=
 export STABLE_BUILD_TYPE?=$(BUILD_TYPE)
 export CMAKE_ARGS?=
@@ -49,7 +49,6 @@ CGO_LDFLAGS_WHISPER+=-lggml
 CUDA_LIBPATH?=/usr/local/cuda/lib64/
 GO_TAGS?=
 BUILD_ID?=
-NATIVE?=false
 
 TEST_DIR=/tmp/test
 
@@ -59,7 +58,7 @@ RANDOM := $(shell bash -c 'echo $$RANDOM')
 
 VERSION?=$(shell git describe --always --tags || echo "dev" )
 # go tool nm ./local-ai | grep Commit
-LD_FLAGS?=-s -w
+LD_FLAGS?=
 override LD_FLAGS += -X "github.com/mudler/LocalAI/internal.Version=$(VERSION)"
 override LD_FLAGS += -X "github.com/mudler/LocalAI/internal.Commit=$(shell git rev-parse HEAD)"
 
@@ -73,14 +72,6 @@ WHITE  := $(shell tput -Txterm setaf 7)
 CYAN   := $(shell tput -Txterm setaf 6)
 RESET  := $(shell tput -Txterm sgr0)
 
-UPX?=
-# check if upx exists
-ifeq (, $(shell which upx))
-	UPX=
-else
-	UPX=$(shell which upx)
-endif
-
 # Default Docker bridge IP
 E2E_BRIDGE_IP?=172.17.0.1
 
@@ -88,25 +79,7 @@ ifndef UNAME_S
 UNAME_S := $(shell uname -s)
 endif
 
-# IF native is false, we add -DGGML_NATIVE=OFF to CMAKE_ARGS
-ifeq ($(NATIVE),false)
-	CMAKE_ARGS+=-DGGML_NATIVE=OFF
-endif
-
-# Detect if we are running on arm64
-ifneq (,$(findstring aarch64,$(shell uname -m)))
-	ONNX_ARCH=aarch64
-endif
-
 ifeq ($(OS),Darwin)
-	ONNX_OS=osx
-	ifneq (,$(findstring aarch64,$(shell uname -m)))
-		ONNX_ARCH=arm64
-	else ifneq (,$(findstring arm64,$(shell uname -m)))
-		ONNX_ARCH=arm64
-	else
-		ONNX_ARCH=x86_64
-	endif
 
 	ifeq ($(OSX_SIGNING_IDENTITY),)
 		OSX_SIGNING_IDENTITY := $(shell security find-identity -v -p codesigning | grep '"' | head -n 1 | sed -E 's/.*"(.*)"/\1/')
@@ -161,10 +134,10 @@ ifeq ($(BUILD_TYPE),hipblas)
 	export CC=$(ROCM_HOME)/llvm/bin/clang
 	# llama-ggml has no hipblas support, so override it here.
 	export STABLE_BUILD_TYPE=
-	export GGML_HIP=1
+	export GGML_HIPBLAS=1
 	GPU_TARGETS ?= gfx900,gfx906,gfx908,gfx940,gfx941,gfx942,gfx90a,gfx1030,gfx1031,gfx1100,gfx1101
 	AMDGPU_TARGETS ?= "$(GPU_TARGETS)"
-	CMAKE_ARGS+=-DGGML_HIP=ON -DAMDGPU_TARGETS="$(AMDGPU_TARGETS)" -DGPU_TARGETS="$(GPU_TARGETS)"
+	CMAKE_ARGS+=-DGGML_HIPBLAS=ON -DAMDGPU_TARGETS="$(AMDGPU_TARGETS)" -DGPU_TARGETS="$(GPU_TARGETS)"
 	CGO_LDFLAGS += -O3 --rtlib=compiler-rt -unwindlib=libgcc -lhipblas -lrocblas --hip-link -L${ROCM_HOME}/lib/llvm/lib
 endif
 
@@ -202,23 +175,17 @@ ifeq ($(findstring tts,$(GO_TAGS)),tts)
 endif
 
 ALL_GRPC_BACKENDS=backend-assets/grpc/huggingface
+ALL_GRPC_BACKENDS+=backend-assets/grpc/bert-embeddings
 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-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/gpt4all
+ALL_GRPC_BACKENDS+=backend-assets/grpc/rwkv
 ALL_GRPC_BACKENDS+=backend-assets/grpc/whisper
-
-ifeq ($(ONNX_OS),linux)
-ifeq ($(ONNX_ARCH),x64)
-	ALL_GRPC_BACKENDS+=backend-assets/grpc/bark-cpp
-	ALL_GRPC_BACKENDS+=backend-assets/grpc/stablediffusion-ggml
-endif
-endif
-
 ALL_GRPC_BACKENDS+=backend-assets/grpc/local-store
-ALL_GRPC_BACKENDS+=backend-assets/grpc/silero-vad
 ALL_GRPC_BACKENDS+=$(OPTIONAL_GRPC)
 # Use filter-out to remove the specified backends
 ALL_GRPC_BACKENDS := $(filter-out $(SKIP_GRPC_BACKEND),$(ALL_GRPC_BACKENDS))
@@ -239,6 +206,19 @@ endif
 
 all: help
 
+## BERT embeddings
+sources/go-bert.cpp:
+	mkdir -p sources/go-bert.cpp
+	cd sources/go-bert.cpp && \
+	git init && \
+	git remote add origin $(BERT_REPO) && \
+	git fetch origin && \
+	git checkout $(BERT_VERSION) && \
+	git submodule update --init --recursive --depth 1 --single-branch
+
+sources/go-bert.cpp/libgobert.a: sources/go-bert.cpp
+	$(MAKE) -C sources/go-bert.cpp libgobert.a
+
 ## go-llama.cpp
 sources/go-llama.cpp:
 	mkdir -p sources/go-llama.cpp
@@ -252,23 +232,6 @@ sources/go-llama.cpp:
 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 && \
-	cd sources/bark.cpp && \
-	git checkout $(BARKCPP_VERSION) && \
-	git submodule update --init --recursive --depth 1 --single-branch
-
-sources/bark.cpp/build/libbark.a: sources/bark.cpp
-	cd sources/bark.cpp && \
-	mkdir -p build && \
-	cd build && \
-	cmake $(CMAKE_ARGS) .. && \
-	cmake --build . --config Release
-
-backend/go/bark/libbark.a: sources/bark.cpp/build/libbark.a
-	$(MAKE) -C backend/go/bark libbark.a
-
 ## go-piper
 sources/go-piper:
 	mkdir -p sources/go-piper
@@ -282,7 +245,33 @@ sources/go-piper:
 sources/go-piper/libpiper_binding.a: sources/go-piper
 	$(MAKE) -C sources/go-piper libpiper_binding.a example/main piper.o
 
-## stable diffusion (onnx)
+## GPT4ALL
+sources/gpt4all:
+	mkdir -p sources/gpt4all
+	cd sources/gpt4all && \
+	git init && \
+	git remote add origin $(GPT4ALL_REPO) && \
+	git fetch origin && \
+	git checkout $(GPT4ALL_VERSION) && \
+	git submodule update --init --recursive --depth 1 --single-branch
+
+sources/gpt4all/gpt4all-bindings/golang/libgpt4all.a: sources/gpt4all
+	$(MAKE) -C sources/gpt4all/gpt4all-bindings/golang/ libgpt4all.a
+
+## RWKV
+sources/go-rwkv.cpp:
+	mkdir -p sources/go-rwkv.cpp
+	cd sources/go-rwkv.cpp && \
+	git init && \
+	git remote add origin $(RWKV_REPO) && \
+	git fetch origin && \
+	git checkout $(RWKV_VERSION) && \
+	git submodule update --init --recursive --depth 1 --single-branch
+
+sources/go-rwkv.cpp/librwkv.a: sources/go-rwkv.cpp
+	cd sources/go-rwkv.cpp && cd rwkv.cpp &&	cmake . -DRWKV_BUILD_SHARED_LIBRARY=OFF &&	cmake --build . && 	cp librwkv.a ..
+
+## stable diffusion
 sources/go-stable-diffusion:
 	mkdir -p sources/go-stable-diffusion
 	cd sources/go-stable-diffusion && \
@@ -295,44 +284,6 @@ sources/go-stable-diffusion:
 sources/go-stable-diffusion/libstablediffusion.a: sources/go-stable-diffusion
 	CPATH="$(CPATH):/usr/include/opencv4" $(MAKE) -C sources/go-stable-diffusion libstablediffusion.a
 
-## stablediffusion (ggml)
-sources/stablediffusion-ggml.cpp:
-	git clone --recursive $(STABLEDIFFUSION_GGML_REPO) sources/stablediffusion-ggml.cpp && \
-	cd sources/stablediffusion-ggml.cpp && \
-	git checkout $(STABLEDIFFUSION_GGML_VERSION) && \
-	git submodule update --init --recursive --depth 1 --single-branch
-
-sources/stablediffusion-ggml.cpp/build/libstable-diffusion.a: sources/stablediffusion-ggml.cpp
-	cd sources/stablediffusion-ggml.cpp && \
-	mkdir -p build && \
-	cd build && \
-	cmake $(CMAKE_ARGS) .. && \
-	cmake --build . --config Release
-
-backend/go/image/stablediffusion-ggml/libsd.a: sources/stablediffusion-ggml.cpp/build/libstable-diffusion.a
-	$(MAKE) -C backend/go/image/stablediffusion-ggml libsd.a
-
-backend-assets/grpc/stablediffusion-ggml: backend/go/image/stablediffusion-ggml/libsd.a backend-assets/grpc
-	CGO_LDFLAGS="$(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),)
-	$(UPX) backend-assets/grpc/stablediffusion-ggml
-endif
-
-sources/onnxruntime:
-	mkdir -p sources/onnxruntime
-	curl -L https://github.com/microsoft/onnxruntime/releases/download/v$(ONNX_VERSION)/onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz -o sources/onnxruntime/onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz
-	cd sources/onnxruntime && tar -xvf onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz && rm onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz
-	cd sources/onnxruntime && mv onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION)/* ./
-
-backend-assets/lib/libonnxruntime.so.1: backend-assets/lib sources/onnxruntime
-	cp -rfv sources/onnxruntime/lib/* backend-assets/lib/
-ifeq ($(OS),Darwin)
-	mv backend-assets/lib/libonnxruntime.$(ONNX_VERSION).dylib backend-assets/lib/libonnxruntime.dylib
-else
-	mv backend-assets/lib/libonnxruntime.so.$(ONNX_VERSION) backend-assets/lib/libonnxruntime.so.1
-endif
-
 ## tiny-dream
 sources/go-tiny-dream:
 	mkdir -p sources/go-tiny-dream
@@ -359,22 +310,28 @@ 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 sources/go-stable-diffusion sources/go-tiny-dream backend/cpp/llama/llama.cpp
+get-sources: sources/go-llama.cpp sources/gpt4all sources/go-piper sources/go-rwkv.cpp sources/whisper.cpp sources/go-bert.cpp sources/go-stable-diffusion sources/go-tiny-dream backend/cpp/llama/llama.cpp
 
 replace:
+	$(GOCMD) mod edit -replace github.com/donomii/go-rwkv.cpp=$(CURDIR)/sources/go-rwkv.cpp
 	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp=$(CURDIR)/sources/whisper.cpp
 	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp/bindings/go=$(CURDIR)/sources/whisper.cpp/bindings/go
+	$(GOCMD) mod edit -replace github.com/go-skynet/go-bert.cpp=$(CURDIR)/sources/go-bert.cpp
 	$(GOCMD) mod edit -replace github.com/M0Rf30/go-tiny-dream=$(CURDIR)/sources/go-tiny-dream
 	$(GOCMD) mod edit -replace github.com/mudler/go-piper=$(CURDIR)/sources/go-piper
 	$(GOCMD) mod edit -replace github.com/mudler/go-stable-diffusion=$(CURDIR)/sources/go-stable-diffusion
+	$(GOCMD) mod edit -replace github.com/nomic-ai/gpt4all/gpt4all-bindings/golang=$(CURDIR)/sources/gpt4all/gpt4all-bindings/golang
 	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(CURDIR)/sources/go-llama.cpp
 
 dropreplace:
+	$(GOCMD) mod edit -dropreplace github.com/donomii/go-rwkv.cpp
 	$(GOCMD) mod edit -dropreplace github.com/ggerganov/whisper.cpp
 	$(GOCMD) mod edit -dropreplace github.com/ggerganov/whisper.cpp/bindings/go
+	$(GOCMD) mod edit -dropreplace github.com/go-skynet/go-bert.cpp
 	$(GOCMD) mod edit -dropreplace github.com/M0Rf30/go-tiny-dream
 	$(GOCMD) mod edit -dropreplace github.com/mudler/go-piper
 	$(GOCMD) mod edit -dropreplace github.com/mudler/go-stable-diffusion
+	$(GOCMD) mod edit -dropreplace github.com/nomic-ai/gpt4all/gpt4all-bindings/golang
 	$(GOCMD) mod edit -dropreplace github.com/go-skynet/go-llama.cpp
 
 prepare-sources: get-sources replace
@@ -384,8 +341,11 @@ prepare-sources: get-sources replace
 rebuild: ## Rebuilds the project
 	$(GOCMD) clean -cache
 	$(MAKE) -C sources/go-llama.cpp clean
+	$(MAKE) -C sources/gpt4all/gpt4all-bindings/golang/ clean
+	$(MAKE) -C sources/go-rwkv.cpp clean
 	$(MAKE) -C sources/whisper.cpp clean
 	$(MAKE) -C sources/go-stable-diffusion clean
+	$(MAKE) -C sources/go-bert.cpp clean
 	$(MAKE) -C sources/go-piper clean
 	$(MAKE) -C sources/go-tiny-dream clean
 	$(MAKE) build
@@ -400,9 +360,7 @@ clean: ## Remove build related file
 	rm -rf release/
 	rm -rf backend-assets/*
 	$(MAKE) -C backend/cpp/grpc clean
-	$(MAKE) -C backend/go/bark clean
 	$(MAKE) -C backend/cpp/llama clean
-	$(MAKE) -C backend/go/image/stablediffusion-ggml clean
 	rm -rf backend/cpp/llama-* || true
 	$(MAKE) dropreplace
 	$(MAKE) protogen-clean
@@ -413,16 +371,13 @@ clean-tests:
 	rm -rf test-dir
 	rm -rf core/http/backend-assets
 
-clean-dc: clean
-	cp -r /build/backend-assets /workspace/backend-assets
-
 ## Build:
 build: prepare backend-assets grpcs ## Build the project
 	$(info ${GREEN}I local-ai build info:${RESET})
 	$(info ${GREEN}I BUILD_TYPE: ${YELLOW}$(BUILD_TYPE)${RESET})
 	$(info ${GREEN}I GO_TAGS: ${YELLOW}$(GO_TAGS)${RESET})
 	$(info ${GREEN}I LD_FLAGS: ${YELLOW}$(LD_FLAGS)${RESET})
-	$(info ${GREEN}I UPX: ${YELLOW}$(UPX)${RESET})
+	ls -liah backend-assets/grpc
 ifneq ($(BACKEND_LIBS),)
 	$(MAKE) backend-assets/lib
 	cp -f $(BACKEND_LIBS) backend-assets/lib/
@@ -433,7 +388,7 @@ build-minimal:
 	BUILD_GRPC_FOR_BACKEND_LLAMA=true GRPC_BACKENDS="backend-assets/grpc/llama-cpp-avx2" GO_TAGS=p2p $(MAKE) build
 
 build-api:
-	BUILD_GRPC_FOR_BACKEND_LLAMA=true BUILD_API_ONLY=true GO_TAGS=p2p $(MAKE) build
+	BUILD_GRPC_FOR_BACKEND_LLAMA=true BUILD_API_ONLY=true GO_TAGS=none $(MAKE) build
 
 backend-assets/lib:
 	mkdir -p backend-assets/lib
@@ -444,7 +399,7 @@ ifeq ($(DETECT_LIBS),true)
 	scripts/prepare-libs.sh backend-assets/grpc/llama-cpp-avx2
 endif
 ifeq ($(OS),Darwin)
-	BUILD_TYPE=none $(MAKE) backend-assets/grpc/llama-cpp-fallback
+	$(info ${GREEN}I Skip CUDA/hipblas build on MacOS${RESET})
 else
 	$(MAKE) backend-assets/grpc/llama-cpp-cuda
 	$(MAKE) backend-assets/grpc/llama-cpp-hipblas
@@ -493,6 +448,8 @@ test-models/testmodel.ggml:
 	wget -q https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin -O test-models/whisper-en
 	wget -q https://huggingface.co/mudler/all-MiniLM-L6-v2/resolve/main/ggml-model-q4_0.bin -O test-models/bert
 	wget -q https://cdn.openai.com/whisper/draft-20220913a/micro-machines.wav -O test-dir/audio.wav
+	wget -q https://huggingface.co/mudler/rwkv-4-raven-1.5B-ggml/resolve/main/RWKV-4-Raven-1B5-v11-Eng99%2525-Other1%2525-20230425-ctx4096_Q4_0.bin -O test-models/rwkv
+	wget -q https://raw.githubusercontent.com/saharNooby/rwkv.cpp/5eb8f09c146ea8124633ab041d9ea0b1f1db4459/rwkv/20B_tokenizer.json -O test-models/rwkv.tokenizer.json
 	cp tests/models_fixtures/* test-models
 
 prepare-test: grpcs
@@ -504,7 +461,8 @@ test: prepare test-models/testmodel.ggml grpcs
 	export GO_TAGS="tts stablediffusion debug"
 	$(MAKE) prepare-test
 	HUGGINGFACE_GRPC=$(abspath ./)/backend/python/sentencetransformers/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)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="!gpt4all && !llama && !llama-gguf"  --flake-attempts $(TEST_FLAKES) --fail-fast -v -r $(TEST_PATHS)
+	$(MAKE) test-gpt4all
 	$(MAKE) test-llama
 	$(MAKE) test-llama-gguf
 	$(MAKE) test-tts
@@ -514,46 +472,50 @@ prepare-e2e:
 	mkdir -p $(TEST_DIR)
 	cp -rfv $(abspath ./tests/e2e-fixtures)/gpu.yaml $(TEST_DIR)/gpu.yaml
 	test -e $(TEST_DIR)/ggllm-test-model.bin || wget -q https://huggingface.co/TheBloke/CodeLlama-7B-Instruct-GGUF/resolve/main/codellama-7b-instruct.Q2_K.gguf -O $(TEST_DIR)/ggllm-test-model.bin
-	docker build --build-arg GRPC_BACKENDS="$(GRPC_BACKENDS)" --build-arg IMAGE_TYPE=core --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg CUDA_MAJOR_VERSION=12 --build-arg CUDA_MINOR_VERSION=0 --build-arg FFMPEG=true -t localai-tests .
+	docker build --build-arg GRPC_BACKENDS="$(GRPC_BACKENDS)" --build-arg IMAGE_TYPE=core --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg CUDA_MAJOR_VERSION=12 --build-arg CUDA_MINOR_VERSION=4 --build-arg FFMPEG=true -t localai-tests .
 
 run-e2e-image:
 	ls -liah $(abspath ./tests/e2e-fixtures)
 	docker run -p 5390:8080 -e MODELS_PATH=/models -e THREADS=1 -e DEBUG=true -d --rm -v $(TEST_DIR):/models --gpus all --name e2e-tests-$(RANDOM) localai-tests
 
-run-e2e-aio: protogen-go
+run-e2e-aio:
 	@echo 'Running e2e AIO tests'
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts $(TEST_FLAKES) -v -r ./tests/e2e-aio
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts 5 -v -r ./tests/e2e-aio
 
 test-e2e:
 	@echo 'Running e2e tests'
 	BUILD_TYPE=$(BUILD_TYPE) \
 	LOCALAI_API=http://$(E2E_BRIDGE_IP):5390/v1 \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts $(TEST_FLAKES) -v -r ./tests/e2e
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts 5 -v -r ./tests/e2e
 
 teardown-e2e:
 	rm -rf $(TEST_DIR) || true
 	docker stop $$(docker ps -q --filter ancestor=localai-tests)
 
+test-gpt4all: 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="gpt4all" --flake-attempts 5 -v -r $(TEST_PATHS)
+
 test-llama: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama" --flake-attempts 5 -v -r $(TEST_PATHS)
 
 test-llama-gguf: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama-gguf" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama-gguf" --flake-attempts 5 -v -r $(TEST_PATHS)
 
 test-tts: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="tts" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="tts" --flake-attempts 1 -v -r $(TEST_PATHS)
 
 test-stablediffusion: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stablediffusion" --flake-attempts $(TEST_FLAKES) -v -r $(TEST_PATHS)
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stablediffusion" --flake-attempts 1 -v -r $(TEST_PATHS)
 
 test-stores: backend-assets/grpc/local-store
 	mkdir -p tests/integration/backend-assets/grpc
 	cp -f backend-assets/grpc/local-store tests/integration/backend-assets/grpc/
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stores" --flake-attempts $(TEST_FLAKES) -v -r tests/integration
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stores" --flake-attempts 1 -v -r tests/integration
 
 test-container:
 	docker build --target requirements -t local-ai-test-container .
@@ -589,10 +551,10 @@ protogen-go-clean:
 	$(RM) bin/*
 
 .PHONY: protogen-python
-protogen-python: autogptq-protogen bark-protogen coqui-protogen diffusers-protogen exllama2-protogen mamba-protogen rerankers-protogen sentencetransformers-protogen transformers-protogen parler-tts-protogen transformers-musicgen-protogen vall-e-x-protogen vllm-protogen openvoice-protogen
+protogen-python: autogptq-protogen bark-protogen coqui-protogen diffusers-protogen exllama-protogen exllama2-protogen mamba-protogen petals-protogen rerankers-protogen sentencetransformers-protogen transformers-protogen parler-tts-protogen transformers-musicgen-protogen vall-e-x-protogen vllm-protogen openvoice-protogen
 
 .PHONY: protogen-python-clean
-protogen-python-clean: autogptq-protogen-clean bark-protogen-clean coqui-protogen-clean diffusers-protogen-clean  exllama2-protogen-clean mamba-protogen-clean sentencetransformers-protogen-clean rerankers-protogen-clean transformers-protogen-clean transformers-musicgen-protogen-clean parler-tts-protogen-clean vall-e-x-protogen-clean vllm-protogen-clean openvoice-protogen-clean
+protogen-python-clean: autogptq-protogen-clean bark-protogen-clean coqui-protogen-clean diffusers-protogen-clean exllama-protogen-clean exllama2-protogen-clean mamba-protogen-clean petals-protogen-clean sentencetransformers-protogen-clean rerankers-protogen-clean transformers-protogen-clean transformers-musicgen-protogen-clean parler-tts-protogen-clean vall-e-x-protogen-clean vllm-protogen-clean openvoice-protogen-clean
 
 .PHONY: autogptq-protogen
 autogptq-protogen:
@@ -626,6 +588,14 @@ diffusers-protogen:
 diffusers-protogen-clean:
 	$(MAKE) -C backend/python/diffusers protogen-clean
 
+.PHONY: exllama-protogen
+exllama-protogen:
+	$(MAKE) -C backend/python/exllama protogen
+
+.PHONY: exllama-protogen-clean
+exllama-protogen-clean:
+	$(MAKE) -C backend/python/exllama protogen-clean
+
 .PHONY: exllama2-protogen
 exllama2-protogen:
 	$(MAKE) -C backend/python/exllama2 protogen
@@ -642,6 +612,14 @@ mamba-protogen:
 mamba-protogen-clean:
 	$(MAKE) -C backend/python/mamba protogen-clean
 
+.PHONY: petals-protogen
+petals-protogen:
+	$(MAKE) -C backend/python/petals protogen
+
+.PHONY: petals-protogen-clean
+petals-protogen-clean:
+	$(MAKE) -C backend/python/petals protogen-clean
+
 .PHONY: rerankers-protogen
 rerankers-protogen:
 	$(MAKE) -C backend/python/rerankers protogen
@@ -722,6 +700,8 @@ prepare-extra-conda-environments: protogen-python
 	$(MAKE) -C backend/python/parler-tts
 	$(MAKE) -C backend/python/vall-e-x
 	$(MAKE) -C backend/python/openvoice
+	$(MAKE) -C backend/python/exllama
+	$(MAKE) -C backend/python/petals
 	$(MAKE) -C backend/python/exllama2
 
 prepare-test-extra: protogen-python
@@ -742,14 +722,25 @@ backend-assets/espeak-ng-data: sources/go-piper sources/go-piper/libpiper_bindin
 	mkdir -p backend-assets/espeak-ng-data
 	@cp -rf sources/go-piper/piper-phonemize/pi/share/espeak-ng-data/. backend-assets/espeak-ng-data
 
+backend-assets/gpt4all: sources/gpt4all sources/gpt4all/gpt4all-bindings/golang/libgpt4all.a
+	mkdir -p backend-assets/gpt4all
+	@cp sources/gpt4all/gpt4all-bindings/golang/buildllm/*.so backend-assets/gpt4all/ || true
+	@cp sources/gpt4all/gpt4all-bindings/golang/buildllm/*.dylib backend-assets/gpt4all/ || true
+	@cp sources/gpt4all/gpt4all-bindings/golang/buildllm/*.dll backend-assets/gpt4all/ || true
+
 backend-assets/grpc: protogen-go replace
 	mkdir -p backend-assets/grpc
 
+backend-assets/grpc/bert-embeddings: sources/go-bert.cpp sources/go-bert.cpp/libgobert.a backend-assets/grpc
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-bert.cpp LIBRARY_PATH=$(CURDIR)/sources/go-bert.cpp \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/bert-embeddings ./backend/go/llm/bert/
+
+backend-assets/grpc/gpt4all: sources/gpt4all sources/gpt4all/gpt4all-bindings/golang/libgpt4all.a backend-assets/gpt4all backend-assets/grpc
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/gpt4all/gpt4all-bindings/golang/ LIBRARY_PATH=$(CURDIR)/sources/gpt4all/gpt4all-bindings/golang/ \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gpt4all ./backend/go/llm/gpt4all/
+
 backend-assets/grpc/huggingface: backend-assets/grpc
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/huggingface ./backend/go/llm/langchain/
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/huggingface
-endif
 
 backend/cpp/llama/llama.cpp:
 	LLAMA_VERSION=$(CPPLLAMA_VERSION) $(MAKE) -C backend/cpp/llama llama.cpp
@@ -804,6 +795,10 @@ backend-assets/grpc/llama-cpp-fallback: backend-assets/grpc backend/cpp/llama/ll
 	$(info ${GREEN}I llama-cpp build info:fallback${RESET})
 	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off" $(MAKE) VARIANT="llama-fallback" build-llama-cpp-grpc-server
 	cp -rfv backend/cpp/llama-fallback/grpc-server backend-assets/grpc/llama-cpp-fallback
+# TODO: every binary should have its own folder instead, so can have different metal implementations
+ifeq ($(BUILD_TYPE),metal)
+	cp backend/cpp/llama-fallback/llama.cpp/build/bin/default.metallib backend-assets/grpc/
+endif
 
 backend-assets/grpc/llama-cpp-cuda: backend-assets/grpc backend/cpp/llama/llama.cpp
 	cp -rf backend/cpp/llama backend/cpp/llama-cuda
@@ -816,7 +811,7 @@ backend-assets/grpc/llama-cpp-hipblas: backend-assets/grpc backend/cpp/llama/lla
 	cp -rf backend/cpp/llama backend/cpp/llama-hipblas
 	$(MAKE) -C backend/cpp/llama-hipblas purge
 	$(info ${GREEN}I llama-cpp build info:hipblas${RESET})
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off" BUILD_TYPE="hipblas" $(MAKE) VARIANT="llama-hipblas" build-llama-cpp-grpc-server
+	BUILD_TYPE="hipblas" $(MAKE) VARIANT="llama-hipblas" build-llama-cpp-grpc-server
 	cp -rfv backend/cpp/llama-hipblas/grpc-server backend-assets/grpc/llama-cpp-hipblas
 
 backend-assets/grpc/llama-cpp-sycl_f16: backend-assets/grpc backend/cpp/llama/llama.cpp
@@ -847,57 +842,29 @@ backend-assets/util/llama-cpp-rpc-server: backend-assets/grpc/llama-cpp-grpc
 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 "$(GO_TAGS)" -o backend-assets/grpc/bark-cpp ./backend/go/bark/
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/bark-cpp
-endif
 
 backend-assets/grpc/piper: sources/go-piper sources/go-piper/libpiper_binding.a backend-assets/grpc backend-assets/espeak-ng-data
 	CGO_CXXFLAGS="$(PIPER_CGO_CXXFLAGS)" CGO_LDFLAGS="$(PIPER_CGO_LDFLAGS)" LIBRARY_PATH=$(CURDIR)/sources/go-piper \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/piper ./backend/go/tts/
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/piper
-endif
+
+backend-assets/grpc/rwkv: sources/go-rwkv.cpp sources/go-rwkv.cpp/librwkv.a backend-assets/grpc
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-rwkv.cpp LIBRARY_PATH=$(CURDIR)/sources/go-rwkv.cpp \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/rwkv ./backend/go/llm/rwkv
 
 backend-assets/grpc/stablediffusion: sources/go-stable-diffusion sources/go-stable-diffusion/libstablediffusion.a backend-assets/grpc
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" CPATH="$(CPATH):$(CURDIR)/sources/go-stable-diffusion/:/usr/include/opencv4" LIBRARY_PATH=$(CURDIR)/sources/go-stable-diffusion/ \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/stablediffusion ./backend/go/image/stablediffusion
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/stablediffusion
-endif
-
-backend-assets/grpc/silero-vad: backend-assets/grpc backend-assets/lib/libonnxruntime.so.1
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" CPATH="$(CPATH):$(CURDIR)/sources/onnxruntime/include/" LIBRARY_PATH=$(CURDIR)/backend-assets/lib \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/silero-vad ./backend/go/vad/silero
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/silero-vad
-endif
 
 backend-assets/grpc/tinydream: sources/go-tiny-dream sources/go-tiny-dream/libtinydream.a backend-assets/grpc
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" LIBRARY_PATH=$(CURDIR)/go-tiny-dream \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/tinydream ./backend/go/image/tinydream
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/tinydream
-endif
 
 backend-assets/grpc/whisper: sources/whisper.cpp sources/whisper.cpp/libwhisper.a backend-assets/grpc
 	CGO_LDFLAGS="$(CGO_LDFLAGS) $(CGO_LDFLAGS_WHISPER)" C_INCLUDE_PATH="$(CURDIR)/sources/whisper.cpp/include:$(CURDIR)/sources/whisper.cpp/ggml/include" LIBRARY_PATH=$(CURDIR)/sources/whisper.cpp \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/whisper ./backend/go/transcribe/whisper
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/whisper
-endif
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/whisper ./backend/go/transcribe/
 
 backend-assets/grpc/local-store: backend-assets/grpc
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/local-store ./backend/go/stores/
-ifneq ($(UPX),)
-	$(UPX) backend-assets/grpc/local-store
-endif
 
 grpcs: prepare $(GRPC_BACKENDS)
 
@@ -939,7 +906,7 @@ docker-aio-all:
 
 docker-image-intel:
 	docker build \
-		--build-arg BASE_IMAGE=intel/oneapi-basekit:2025.0.0-0-devel-ubuntu22.04 \
+		--build-arg BASE_IMAGE=intel/oneapi-basekit:2024.2.0-devel-ubuntu22.04 \
 		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
 		--build-arg GO_TAGS="none" \
 		--build-arg MAKEFLAGS="$(DOCKER_MAKEFLAGS)" \
@@ -947,7 +914,7 @@ docker-image-intel:
 
 docker-image-intel-xpu:
 	docker build \
-		--build-arg BASE_IMAGE=intel/oneapi-basekit:2025.0.0-0-devel-ubuntu22.04 \
+		--build-arg BASE_IMAGE=intel/oneapi-basekit:2024.2.0-devel-ubuntu22.04 \
 		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
 		--build-arg GO_TAGS="none" \
 		--build-arg MAKEFLAGS="$(DOCKER_MAKEFLAGS)" \

README.md

@@ -38,13 +38,9 @@
 </a>
 </p>
 
-<p align="center">
-<a href="https://trendshift.io/repositories/1484" target="_blank"><img src="https://trendshift.io/api/badge/repositories/1484" alt="go-skynet%2FLocalAI | Trendshift" style="width: 250px; height: 55px;" width="250" height="55"/></a>
-</p>
-
 > :bulb: Get help - [❓FAQ](https://localai.io/faq/) [💭Discussions](https://github.com/go-skynet/LocalAI/discussions) [:speech_balloon: Discord](https://discord.gg/uJAeKSAGDy) [:book: Documentation website](https://localai.io/)
 >
-> [💻 Quickstart](https://localai.io/basics/getting_started/) [🖼️ Models](https://models.localai.io/) [🚀 Roadmap](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap) [🥽 Demo](https://demo.localai.io) [🌍 Explorer](https://explorer.localai.io) [🛫 Examples](https://github.com/mudler/LocalAI-examples) 
+> [💻 Quickstart](https://localai.io/basics/getting_started/) [📣 News](https://localai.io/basics/news/) [ 🛫 Examples ](https://github.com/go-skynet/LocalAI/tree/master/examples/) [ 🖼️ Models ](https://localai.io/models/) [ 🚀 Roadmap ](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap)
 
 [![tests](https://github.com/go-skynet/LocalAI/actions/workflows/test.yml/badge.svg)](https://github.com/go-skynet/LocalAI/actions/workflows/test.yml)[![Build and Release](https://github.com/go-skynet/LocalAI/actions/workflows/release.yaml/badge.svg)](https://github.com/go-skynet/LocalAI/actions/workflows/release.yaml)[![build container images](https://github.com/go-skynet/LocalAI/actions/workflows/image.yml/badge.svg)](https://github.com/go-skynet/LocalAI/actions/workflows/image.yml)[![Bump dependencies](https://github.com/go-skynet/LocalAI/actions/workflows/bump_deps.yaml/badge.svg)](https://github.com/go-skynet/LocalAI/actions/workflows/bump_deps.yaml)[![Artifact Hub](https://img.shields.io/endpoint?url=https://artifacthub.io/badge/repository/localai)](https://artifacthub.io/packages/search?repo=localai)
 
@@ -60,43 +56,22 @@ curl https://localai.io/install.sh | sh
 
 Or run with docker:
 ```bash
-# CPU only image:
-docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-cpu
-
-# Nvidia GPU:
-docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-gpu-nvidia-cuda-12
-
-# CPU and GPU image (bigger size):
-docker run -ti --name local-ai -p 8080:8080 localai/localai:latest
-
-# AIO images (it will pre-download a set of models ready for use, see https://localai.io/basics/container/)
 docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-aio-cpu
-```
-
-To load models:
-
-```bash
-# From the model gallery (see available models with `local-ai models list`, in the WebUI from the model tab, or visiting https://models.localai.io)
-local-ai run llama-3.2-1b-instruct:q4_k_m
-# Start LocalAI with the phi-2 model directly from huggingface
-local-ai run huggingface://TheBloke/phi-2-GGUF/phi-2.Q8_0.gguf
-# Install and run a model from the Ollama OCI registry
-local-ai run ollama://gemma:2b
-# Run a model from a configuration file
-local-ai run https://gist.githubusercontent.com/.../phi-2.yaml
-# Install and run a model from a standard OCI registry (e.g., Docker Hub)
-local-ai run oci://localai/phi-2:latest
+# Alternative images:
+# - if you have an Nvidia GPU:
+# docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-aio-gpu-nvidia-cuda-12
+# - without preconfigured models
+# docker run -ti --name local-ai -p 8080:8080 localai/localai:latest
+# - without preconfigured models for Nvidia GPUs
+# docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-gpu-nvidia-cuda-12 
 ```
 
 [💻 Getting started](https://localai.io/basics/getting_started/index.html)
 
-## 📰 Latest project news
+## 🔥🔥 Hot topics / Roadmap
+
+[Roadmap](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap)
 
-- 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 )
-- Nov 2024: Voice activity detection models (**VAD**) added to the API: https://github.com/mudler/LocalAI/pull/4204
-- Oct 2024: examples moved to [LocalAI-examples](https://github.com/mudler/LocalAI-examples)
-- Aug 2024:  🆕 FLUX-1, [P2P Explorer](https://explorer.localai.io)
 - July 2024: 🔥🔥 🆕 P2P Dashboard, LocalAI Federated mode and AI Swarms: https://github.com/mudler/LocalAI/pull/2723
 - June 2024: 🆕 You can browse now the model gallery without LocalAI! Check out https://models.localai.io
 - June 2024: Support for models from OCI registries: https://github.com/mudler/LocalAI/pull/2628
@@ -107,13 +82,8 @@ local-ai run oci://localai/phi-2:latest
 - May 2024: Chat, TTS, and Image generation in the WebUI: https://github.com/mudler/LocalAI/pull/2222
 - April 2024: Reranker API: https://github.com/mudler/LocalAI/pull/2121
 
-Roadmap items: [List of issues](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap)
+Hot topics (looking for contributors):
 
-## 🔥🔥 Hot topics (looking for help):
-
-- Multimodal with vLLM and Video understanding: https://github.com/mudler/LocalAI/pull/3729
-- Realtime API https://github.com/mudler/LocalAI/issues/3714
-- 🔥🔥 Distributed, P2P Global community pools: https://github.com/mudler/LocalAI/issues/3113
 - WebUI improvements: https://github.com/mudler/LocalAI/issues/2156
 - Backends v2: https://github.com/mudler/LocalAI/issues/1126
 - Improving UX v2: https://github.com/mudler/LocalAI/issues/1373
@@ -166,9 +136,6 @@ Other:
 - Slack bot https://github.com/mudler/LocalAGI/tree/main/examples/slack
 - Shell-Pilot(Interact with LLM using LocalAI models via pure shell scripts on your Linux or MacOS system) https://github.com/reid41/shell-pilot
 - Telegram bot https://github.com/mudler/LocalAI/tree/master/examples/telegram-bot
-- Another Telegram Bot https://github.com/JackBekket/Hellper
-- Auto-documentation https://github.com/JackBekket/Reflexia
-- Github bot which answer on issues, with code and documentation as context https://github.com/JackBekket/GitHelper
 - Github Actions: https://github.com/marketplace/actions/start-localai
 - Examples: https://github.com/mudler/LocalAI/tree/master/examples/
   
@@ -183,7 +150,6 @@ Other:
 
 ## :book: 🎥 [Media, Blogs, Social](https://localai.io/basics/news/#media-blogs-social)
 
-- [Run Visual studio code with LocalAI (SUSE)](https://www.suse.com/c/running-ai-locally/)
 - 🆕 [Run LocalAI on Jetson Nano Devkit](https://mudler.pm/posts/local-ai-jetson-nano-devkit/)
 - [Run LocalAI on AWS EKS with Pulumi](https://www.pulumi.com/blog/low-code-llm-apps-with-local-ai-flowise-and-pulumi/)
 - [Run LocalAI on AWS](https://staleks.hashnode.dev/installing-localai-on-aws-ec2-instance)
@@ -243,6 +209,7 @@ LocalAI couldn't have been built without the help of great software already avai
 - https://github.com/antimatter15/alpaca.cpp
 - https://github.com/EdVince/Stable-Diffusion-NCNN
 - https://github.com/ggerganov/whisper.cpp
+- https://github.com/saharNooby/rwkv.cpp
 - https://github.com/rhasspy/piper
 
 ## 🤗 Contributors

aio/cpu/embeddings.yaml

@@ -1,7 +1,7 @@
 name: text-embedding-ada-002
-embeddings: true
+backend: bert-embeddings
 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://mudler/all-MiniLM-L6-v2/ggml-model-q4_0.bin
 
 usage: |
     You can test this model with curl like this:

aio/cpu/vision.yaml

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

aio/gpu-8g/vision.yaml

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

aio/intel/image-gen.yaml

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

aio/intel/vision.yaml

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

backend/backend.proto

@@ -16,7 +16,6 @@ service Backend {
   rpc GenerateImage(GenerateImageRequest) returns (Result) {}
   rpc AudioTranscription(TranscriptRequest) returns (TranscriptResult) {}
   rpc TTS(TTSRequest) returns (Result) {}
-  rpc SoundGeneration(SoundGenerationRequest) returns (Result) {}
   rpc TokenizeString(PredictOptions) returns (TokenizationResponse) {}
   rpc Status(HealthMessage) returns (StatusResponse) {}
 
@@ -26,21 +25,6 @@ service Backend {
   rpc StoresFind(StoresFindOptions) returns (StoresFindResult) {}
 
   rpc Rerank(RerankRequest) returns (RerankResult) {}
-
-  rpc GetMetrics(MetricsRequest) returns (MetricsResponse);
-
-  rpc VAD(VADRequest) returns (VADResponse) {}
-}
-
-// Define the empty request
-message MetricsRequest {}
-
-message MetricsResponse {
-  int32 slot_id = 1;
-  string prompt_json_for_slot = 2;  // Stores the prompt as a JSON string.
-  float tokens_per_second = 3;
-  int32 tokens_generated = 4;
-  int32 prompt_tokens_processed = 5;
 }
 
 message RerankRequest {
@@ -149,9 +133,6 @@ message PredictOptions {
   repeated string Images = 42;
   bool UseTokenizerTemplate = 43;
   repeated Message Messages = 44;
-  repeated string Videos = 45;
-  repeated string Audios = 46;
-  string CorrelationId = 47;
 }
 
 // The response message containing the result
@@ -221,7 +202,6 @@ message ModelOptions {
   int32  SwapSpace = 53;
   int32  MaxModelLen = 54;
   int32  TensorParallelSize = 55;
-  string LoadFormat = 58;
 
   string MMProj = 41;
 
@@ -235,13 +215,6 @@ message ModelOptions {
 
   bool FlashAttention = 56;
   bool NoKVOffload = 57;
-
-  string ModelPath = 59;
-
-  repeated string LoraAdapters = 60;
-  repeated float LoraScales = 61;
-
-  repeated string Options = 62;
 }
 
 message Result {
@@ -297,30 +270,6 @@ message TTSRequest {
   optional string language = 5;
 }
 
-message VADRequest {
-  repeated float audio = 1;
-}
-
-message VADSegment {
-  float start = 1;
-  float end = 2;
-}
-
-message VADResponse {
-  repeated VADSegment segments = 1;
-}
-
-message SoundGenerationRequest {
-  string text = 1;
-  string model = 2;
-  string dst = 3;
-  optional float duration = 4;
-  optional float temperature = 5;
-  optional bool sample = 6;
-  optional string src = 7;
-  optional int32 src_divisor = 8;
-}
-
 message TokenizationResponse {
   int32 length = 1;
   repeated int32 tokens = 2;

backend/cpp/llama/Makefile

@@ -22,7 +22,7 @@ else ifeq ($(BUILD_TYPE),clblas)
 	CMAKE_ARGS+=-DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
 # If it's hipblas we do have also to set CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++ 
 else ifeq ($(BUILD_TYPE),hipblas)
-	CMAKE_ARGS+=-DGGML_HIP=ON
+	CMAKE_ARGS+=-DGGML_HIPBLAS=ON
 # If it's OSX, DO NOT embed the metal library - -DGGML_METAL_EMBED_LIBRARY=ON requires further investigation
 # But if it's OSX without metal, disable it here
 else ifeq ($(OS),Darwin)
@@ -30,7 +30,9 @@ else ifeq ($(OS),Darwin)
 		CMAKE_ARGS+=-DGGML_METAL=OFF
 	else
 		CMAKE_ARGS+=-DGGML_METAL=ON
-		CMAKE_ARGS+=-DGGML_METAL_EMBED_LIBRARY=ON
+# Until this is tested properly, we disable embedded metal file
+# as we already embed it as part of the LocalAI assets
+		CMAKE_ARGS+=-DGGML_METAL_EMBED_LIBRARY=OFF
 		TARGET+=--target ggml-metal
 	endif
 endif

backend/cpp/llama/grpc-server.cpp

@@ -13,15 +13,15 @@
 #include <getopt.h>
 #include "clip.h"
 #include "llava.h"
-#include "log.h"
 #include "stb_image.h"
 #include "common.h"
 #include "json.hpp"
 #include "llama.h"
+#include "grammar-parser.h"
 #include "backend.pb.h"
 #include "backend.grpc.pb.h"
 #include "utils.hpp"
-#include "sampling.h"
+
 // include std::regex
 #include <cstddef>
 #include <thread>
@@ -113,7 +113,7 @@ static std::string tokens_to_str(llama_context *ctx, Iter begin, Iter end)
     std::string ret;
     for (; begin != end; ++begin)
     {
-        ret += common_token_to_piece(ctx, *begin);
+        ret += llama_token_to_piece(ctx, *begin);
     }
     return ret;
 }
@@ -121,7 +121,7 @@ static std::string tokens_to_str(llama_context *ctx, Iter begin, Iter end)
 // format incomplete utf-8 multibyte character for output
 static std::string tokens_to_output_formatted_string(const llama_context *ctx, const llama_token token)
 {
-    std::string out = token == -1 ? "" : common_token_to_piece(ctx, token);
+    std::string out = token == -1 ? "" : llama_token_to_piece(ctx, token);
     // if the size is 1 and first bit is 1, meaning it's a partial character
     //   (size > 1 meaning it's already a known token)
     if (out.size() == 1 && (out[0] & 0x80) == 0x80)
@@ -203,8 +203,8 @@ struct llama_client_slot
     std::string stopping_word;
 
     // sampling
-    struct common_params_sampling sparams;
-    common_sampler *ctx_sampling = nullptr;
+    struct llama_sampling_params sparams;
+    llama_sampling_context *ctx_sampling = nullptr;
 
     int32_t ga_i = 0;   // group-attention state
     int32_t ga_n = 1;   // group-attention factor
@@ -257,7 +257,7 @@ struct llama_client_slot
         images.clear();
     }
 
-    bool has_budget(common_params &global_params) {
+    bool has_budget(gpt_params &global_params) {
         if (params.n_predict == -1 && global_params.n_predict == -1)
         {
             return true; // limitless
@@ -391,39 +391,6 @@ struct llama_metrics {
     }
 };
 
-struct llava_embd_batch {
-    std::vector<llama_pos>      pos;
-    std::vector<int32_t>        n_seq_id;
-    std::vector<llama_seq_id>   seq_id_0;
-    std::vector<llama_seq_id *> seq_ids;
-    std::vector<int8_t>         logits;
-    llama_batch batch;
-    llava_embd_batch(float * embd, int32_t n_tokens, llama_pos pos_0, llama_seq_id seq_id) {
-        pos     .resize(n_tokens);
-        n_seq_id.resize(n_tokens);
-        seq_ids .resize(n_tokens + 1);
-        logits  .resize(n_tokens);
-        seq_id_0.resize(1);
-        seq_id_0[0] = seq_id;
-        seq_ids [n_tokens] = nullptr;
-        batch = {
-            /*n_tokens       =*/ n_tokens,
-            /*tokens         =*/ nullptr,
-            /*embd           =*/ embd,
-            /*pos            =*/ pos.data(),
-            /*n_seq_id       =*/ n_seq_id.data(),
-            /*seq_id         =*/ seq_ids.data(),
-            /*logits         =*/ logits.data(),
-        };
-        for (int i = 0; i < n_tokens; i++) {
-            batch.pos     [i] = pos_0 + i;
-            batch.n_seq_id[i] = 1;
-            batch.seq_id  [i] = seq_id_0.data();
-            batch.logits  [i] = false;
-        }
-    }
-};
-
 struct llama_server_context
 {
     llama_model *model = nullptr;
@@ -431,7 +398,7 @@ struct llama_server_context
 
     clip_ctx *clp_ctx = nullptr;
 
-    common_params params;
+    gpt_params params;
 
     llama_batch batch;
 
@@ -474,7 +441,7 @@ struct llama_server_context
         }
     }
 
-    bool load_model(const common_params &params_)
+    bool load_model(const gpt_params &params_)
     {
         params = params_;
         if (!params.mmproj.empty()) {
@@ -482,7 +449,7 @@ struct llama_server_context
             LOG_INFO("Multi Modal Mode Enabled", {});
             clp_ctx = clip_model_load(params.mmproj.c_str(), /*verbosity=*/ 1);
             if(clp_ctx == nullptr) {
-                LOG_ERR("unable to load clip model: %s", params.mmproj.c_str());
+                LOG_ERROR("unable to load clip model", {{"model", params.mmproj}});
                 return false;
             }
 
@@ -491,12 +458,10 @@ struct llama_server_context
             }
         }
 
-        common_init_result common_init = common_init_from_params(params);
-        model = common_init.model;
-        ctx = common_init.context;
+        std::tie(model, ctx) = llama_init_from_gpt_params(params);
         if (model == nullptr)
         {
-            LOG_ERR("unable to load model: %s", params.model.c_str());
+            LOG_ERROR("unable to load model", {{"model", params.model}});
             return false;
         }
 
@@ -504,7 +469,7 @@ struct llama_server_context
             const int n_embd_clip = clip_n_mmproj_embd(clp_ctx);
             const int n_embd_llm  = llama_n_embd(model);
             if (n_embd_clip != n_embd_llm) {
-                LOG("%s: embedding dim of the multimodal projector (%d) is not equal to that of LLaMA (%d). Make sure that you use the correct mmproj file.\n", __func__, n_embd_clip, n_embd_llm);
+                LOG_TEE("%s: embedding dim of the multimodal projector (%d) is not equal to that of LLaMA (%d). Make sure that you use the correct mmproj file.\n", __func__, n_embd_clip, n_embd_llm);
                 llama_free(ctx);
                 llama_free_model(model);
                 return false;
@@ -513,7 +478,7 @@ struct llama_server_context
 
         n_ctx = llama_n_ctx(ctx);
 
-        add_bos_token = llama_add_bos_token(model);
+        add_bos_token = llama_should_add_bos_token(model);
 
         return true;
     }
@@ -523,21 +488,11 @@ struct llama_server_context
         std::vector<char> buf(1);
         int res = llama_chat_apply_template(model, nullptr, chat, 1, true, buf.data(), buf.size());
         if (res < 0) {
-            LOG_ERR("The chat template comes with this model is not yet supported, falling back to chatml. This may cause the model to output suboptimal responses", __func__);
+            LOG_ERROR("The chat template comes with this model is not yet supported, falling back to chatml. This may cause the model to output suboptimal responses", {});
             sparams.chat_template = "<|im_start|>"; // llama_chat_apply_template only checks if <|im_start|> exist in the template
         }
     }
 
-    llama_client_slot* get_active_slot() {
-        for (llama_client_slot& slot : slots) {
-            // Check if the slot is currently processing
-            if (slot.is_processing()) {
-                return &slot;  // Return the active slot
-            }
-        }
-        return nullptr;  // No active slot found
-    }
-
     void initialize() {
         // create slots
         all_slots_are_idle = true;
@@ -611,12 +566,12 @@ struct llama_server_context
                     std::vector<llama_token> p;
                     if (first)
                     {
-                        p = common_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
+                        p = ::llama_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
                         first = false;
                     }
                     else
                     {
-                        p = common_tokenize(ctx, s, false, TMP_FORCE_SPECIAL);
+                        p = ::llama_tokenize(ctx, s, false, TMP_FORCE_SPECIAL);
                     }
                     prompt_tokens.insert(prompt_tokens.end(), p.begin(), p.end());
                 }
@@ -633,7 +588,7 @@ struct llama_server_context
         else
         {
             auto s = json_prompt.template get<std::string>();
-            prompt_tokens = common_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
+            prompt_tokens = ::llama_tokenize(ctx, s, add_bos, TMP_FORCE_SPECIAL);
         }
 
         return prompt_tokens;
@@ -662,7 +617,7 @@ struct llama_server_context
 
     bool launch_slot_with_data(llama_client_slot* &slot, json data) {
         slot_params default_params;
-        common_params_sampling default_sparams;
+        llama_sampling_params default_sparams;
  
         slot->params.stream             = json_value(data, "stream",            false);
         slot->params.cache_prompt       = json_value(data, "cache_prompt",      false);
@@ -670,7 +625,8 @@ struct llama_server_context
         slot->sparams.top_k             = json_value(data, "top_k",             default_sparams.top_k);
         slot->sparams.top_p             = json_value(data, "top_p",             default_sparams.top_p);
         slot->sparams.min_p             = json_value(data, "min_p",             default_sparams.min_p);
-        slot->sparams.typ_p             = json_value(data, "typical_p",         default_sparams.typ_p);
+        slot->sparams.tfs_z             = json_value(data, "tfs_z",             default_sparams.tfs_z);
+        slot->sparams.typical_p         = json_value(data, "typical_p",         default_sparams.typical_p);
         slot->sparams.temp              = json_value(data, "temperature",       default_sparams.temp);
         slot->sparams.dynatemp_range    = json_value(data, "dynatemp_range",    default_sparams.dynatemp_range);
         slot->sparams.dynatemp_exponent = json_value(data, "dynatemp_exponent", default_sparams.dynatemp_exponent);
@@ -683,7 +639,7 @@ struct llama_server_context
         slot->sparams.mirostat_eta      = json_value(data, "mirostat_eta",      default_sparams.mirostat_eta);
         slot->sparams.penalize_nl       = json_value(data, "penalize_nl",       default_sparams.penalize_nl);
         slot->params.n_keep             = json_value(data, "n_keep",            slot->params.n_keep);
-        slot->sparams.seed               = json_value(data, "seed",              default_sparams.seed);
+        slot->params.seed               = json_value(data, "seed",              default_params.seed);
         slot->sparams.grammar           = json_value(data, "grammar",           default_sparams.grammar);
         slot->sparams.n_probs           = json_value(data, "n_probs",           default_sparams.n_probs);
         slot->sparams.min_keep          = json_value(data, "min_keep",          default_sparams.min_keep);
@@ -707,7 +663,6 @@ struct llama_server_context
             slot->params.input_prefix = "";
         }
 
-
         if (data.count("input_suffix") != 0)
         {
             slot->params.input_suffix = data["input_suffix"];
@@ -726,10 +681,6 @@ struct llama_server_context
             slot->prompt = "";
         }
 
-        if (json_value(data, "ignore_eos", false)) {
-                slot->sparams.logit_bias.push_back({llama_token_eos(model), -INFINITY});
-        }
-        /*
         slot->sparams.penalty_prompt_tokens.clear();
         slot->sparams.use_penalty_prompt_tokens = false;
         const auto &penalty_prompt = data.find("penalty_prompt");
@@ -765,10 +716,14 @@ struct llama_server_context
                 slot->sparams.use_penalty_prompt_tokens = true;
             }
         }
-      */
 
         slot->sparams.logit_bias.clear();
 
+        if (json_value(data, "ignore_eos", false))
+        {
+            slot->sparams.logit_bias[llama_token_eos(model)] = -INFINITY;
+        }
+
         const auto &logit_bias = data.find("logit_bias");
         if (logit_bias != data.end() && logit_bias->is_array())
         {
@@ -796,21 +751,21 @@ struct llama_server_context
                         llama_token tok = el[0].get<llama_token>();
                         if (tok >= 0 && tok < n_vocab)
                         {
-                            slot->sparams.logit_bias.push_back({tok, bias});
+                            slot->sparams.logit_bias[tok] = bias;
                         }
                     }
                     else if (el[0].is_string())
                     {
-                        auto toks = common_tokenize(model, el[0].get<std::string>(), false);
+                        auto toks = llama_tokenize(model, el[0].get<std::string>(), false);
                         for (auto tok : toks)
                         {
-                            slot->sparams.logit_bias.push_back({tok, bias});
+                            slot->sparams.logit_bias[tok] = bias;
                         }
                     }
                 }
             }
         }
-        
+
         slot->params.antiprompt.clear();
 
         const auto &stop = data.find("stop");
@@ -824,22 +779,24 @@ struct llama_server_context
                 }
             }
         }
-        
-        const auto & samplers = data.find("samplers");
-        if (samplers != data.end() && samplers->is_array()) {
+
+        const auto &samplers_sequence = data.find("samplers");
+        if (samplers_sequence != data.end() && samplers_sequence->is_array())
+        {
             std::vector<std::string> sampler_names;
-                for (const auto & name : *samplers) {
-                    if (name.is_string()) {
-                        sampler_names.emplace_back(name);
-                    }
+            for (const auto &sampler_name : *samplers_sequence)
+            {
+                if (sampler_name.is_string())
+                {
+                    sampler_names.emplace_back(sampler_name);
                 }
-                slot->sparams.samplers = common_sampler_types_from_names(sampler_names, false);
+            }
+            slot->sparams.samplers_sequence = llama_sampling_types_from_names(sampler_names, false);
         }
         else
         {
-                slot->sparams.samplers = default_sparams.samplers;
+            slot->sparams.samplers_sequence = default_sparams.samplers_sequence;
         }
-        
 
         if (multimodal)
         {
@@ -855,11 +812,10 @@ struct llama_server_context
                     img_sl.img_data = clip_image_u8_init();
                     if (!clip_image_load_from_bytes(image_buffer.data(), image_buffer.size(), img_sl.img_data))
                     {
-                        LOG_ERR("%s: failed to load image, slot_id: %d, img_sl_id: %d", 
-                             __func__,
-                             slot->id,
-                             img_sl.id
-                        );
+                        LOG_ERROR("failed to load image", {
+                            {"slot_id",   slot->id},
+                            {"img_sl_id", img_sl.id}
+                        });
                         return false;
                     }
                     LOG_VERBOSE("image loaded", {
@@ -897,12 +853,12 @@ struct llama_server_context
                                     }
                                 }
                                 if (!found) {
-                                    LOG("ERROR: Image with id: %i, not found.\n", img_id);
+                                    LOG_TEE("ERROR: Image with id: %i, not found.\n", img_id);
                                     slot->images.clear();
                                     return false;
                                 }
                             } catch (const std::invalid_argument& e) {
-                                LOG("Invalid image number id in prompt\n");
+                                LOG_TEE("Invalid image number id in prompt\n");
                                 slot->images.clear();
                                 return false;
                             }
@@ -917,10 +873,10 @@ struct llama_server_context
 
         if (slot->ctx_sampling != nullptr)
         {
-            common_sampler_free(slot->ctx_sampling);
+            llama_sampling_free(slot->ctx_sampling);
         }
-        slot->ctx_sampling = common_sampler_init(model, slot->sparams);
-        //llama_set_rng_seed(ctx, slot->params.seed);
+        slot->ctx_sampling = llama_sampling_init(slot->sparams);
+        llama_set_rng_seed(ctx, slot->params.seed);
         slot->command = LOAD_PROMPT;
 
         all_slots_are_idle = false;
@@ -930,7 +886,7 @@ struct llama_server_context
             {"task_id", slot->task_id},
         });
 
-      //  LOG("sampling: \n%s\n", llama_sampling_print(slot->sparams).c_str());
+        LOG_TEE("sampling: \n%s\n", llama_sampling_print(slot->sparams).c_str());
 
         return true;
     }
@@ -946,13 +902,13 @@ struct llama_server_context
         system_tokens.clear();
 
         if (!system_prompt.empty()) {
-            system_tokens = common_tokenize(ctx, system_prompt, add_bos_token);
+            system_tokens = ::llama_tokenize(ctx, system_prompt, add_bos_token);
 
-            common_batch_clear(batch);
+            llama_batch_clear(batch);
 
             for (int i = 0; i < (int)system_tokens.size(); ++i)
             {
-                common_batch_add(batch, system_tokens[i], i, { 0 }, false);
+                llama_batch_add(batch, system_tokens[i], i, { 0 }, false);
             }
 
             for (int32_t i = 0; i < (int32_t) batch.n_tokens; i += params.n_batch)
@@ -966,10 +922,11 @@ struct llama_server_context
                     batch.n_seq_id + i,
                     batch.seq_id   + i,
                     batch.logits   + i,
+                    0, 0, 0, // unused
                 };
                 if (llama_decode(ctx, batch_view) != 0)
                 {
-                    LOG("%s: llama_decode() failed\n", __func__);
+                    LOG_TEE("%s: llama_decode() failed\n", __func__);
                     return;
                 }
             }
@@ -981,7 +938,7 @@ struct llama_server_context
             }
         }
 
-        LOG("system prompt updated\n");
+        LOG_TEE("system prompt updated\n");
         system_need_update = false;
     }
 
@@ -1040,20 +997,18 @@ struct llama_server_context
 
     bool process_token(completion_token_output &result, llama_client_slot &slot) {
         // remember which tokens were sampled - used for repetition penalties during sampling
-        const std::string token_str = common_token_to_piece(ctx, result.tok);
+        const std::string token_str = llama_token_to_piece(ctx, result.tok);
         slot.sampled = result.tok;
 
         // search stop word and delete it
         slot.generated_text += token_str;
         slot.has_next_token = true;
 
-/*
         if (slot.ctx_sampling->params.use_penalty_prompt_tokens && result.tok != -1)
         {
             // we can change penalty_prompt_tokens because it is always created from scratch each request
             slot.ctx_sampling->params.penalty_prompt_tokens.push_back(result.tok);
         }
-        */
 
         // check if there is incomplete UTF-8 character at the end
         bool incomplete = false;
@@ -1162,8 +1117,8 @@ struct llama_server_context
                 continue;
             }
 
-            if (!llava_image_embed_make_with_clip_img(clp_ctx, params.cpuparams.n_threads, img.img_data, &img.image_embedding, &img.image_tokens)) {
-                LOG("Error processing the given image");
+            if (!llava_image_embed_make_with_clip_img(clp_ctx, params.n_threads, img.img_data, &img.image_embedding, &img.image_tokens)) {
+                LOG_TEE("Error processing the given image");
                 return false;
             }
 
@@ -1175,7 +1130,7 @@ struct llama_server_context
 
     void send_error(task_server& task, const std::string &error)
     {
-        LOG("task %i - error: %s\n", task.id, error.c_str());
+        LOG_TEE("task %i - error: %s\n", task.id, error.c_str());
         task_result res;
         res.id = task.id;
         res.multitask_id = task.multitask_id;
@@ -1187,11 +1142,13 @@ struct llama_server_context
 
     json get_formated_generation(llama_client_slot &slot)
     {
-        std::vector<std::string> samplers;
-        samplers.reserve(slot.sparams.samplers.size());
-        for (const auto & sampler : slot.sparams.samplers)
+        const auto eos_bias = slot.sparams.logit_bias.find(llama_token_eos(model));
+        const bool ignore_eos = eos_bias != slot.sparams.logit_bias.end() &&
+                                eos_bias->second < 0.0f && std::isinf(eos_bias->second);
+        std::vector<std::string> samplers_sequence;
+        for (const auto &sampler_type : slot.sparams.samplers_sequence)
         {
-            samplers.emplace_back(common_sampler_type_to_str(sampler));
+            samplers_sequence.emplace_back(llama_sampling_type_to_str(sampler_type));
         }
 
         return json {
@@ -1205,11 +1162,14 @@ struct llama_server_context
             {"top_k",             slot.sparams.top_k},
             {"top_p",             slot.sparams.top_p},
             {"min_p",             slot.sparams.min_p},
-            {"typical_p",         slot.sparams.typ_p},
+            {"tfs_z",             slot.sparams.tfs_z},
+            {"typical_p",         slot.sparams.typical_p},
             {"repeat_last_n",     slot.sparams.penalty_last_n},
             {"repeat_penalty",    slot.sparams.penalty_repeat},
             {"presence_penalty",  slot.sparams.penalty_present},
             {"frequency_penalty", slot.sparams.penalty_freq},
+            {"penalty_prompt_tokens", slot.sparams.penalty_prompt_tokens},
+            {"use_penalty_prompt_tokens", slot.sparams.use_penalty_prompt_tokens},
             {"mirostat",          slot.sparams.mirostat},
             {"mirostat_tau",      slot.sparams.mirostat_tau},
             {"mirostat_eta",      slot.sparams.mirostat_eta},
@@ -1217,13 +1177,13 @@ struct llama_server_context
             {"stop",              slot.params.antiprompt},
             {"n_predict",         slot.params.n_predict},
             {"n_keep",            params.n_keep},
-            {"ignore_eos",        slot.sparams.ignore_eos},
+            {"ignore_eos",        ignore_eos},
             {"stream",            slot.params.stream},
-      //      {"logit_bias",        slot.sparams.logit_bias},
+            {"logit_bias",        slot.sparams.logit_bias},
             {"n_probs",           slot.sparams.n_probs},
             {"min_keep",          slot.sparams.min_keep},
             {"grammar",           slot.sparams.grammar},
-            {"samplers",          samplers}
+            {"samplers",          samplers_sequence}
         };
     }
 
@@ -1246,7 +1206,7 @@ struct llama_server_context
         if (slot.sparams.n_probs > 0)
         {
             std::vector<completion_token_output> probs_output = {};
-            const std::vector<llama_token> to_send_toks = common_tokenize(ctx, tkn.text_to_send, false);
+            const std::vector<llama_token> to_send_toks = llama_tokenize(ctx, tkn.text_to_send, false);
             size_t probs_pos      = std::min(slot.sent_token_probs_index,                       slot.generated_token_probs.size());
             size_t probs_stop_pos = std::min(slot.sent_token_probs_index + to_send_toks.size(), slot.generated_token_probs.size());
             if (probs_pos < probs_stop_pos)
@@ -1298,7 +1258,7 @@ struct llama_server_context
             std::vector<completion_token_output> probs = {};
             if (!slot.params.stream && slot.stopped_word)
             {
-                const std::vector<llama_token> stop_word_toks = common_tokenize(ctx, slot.stopping_word, false);
+                const std::vector<llama_token> stop_word_toks = llama_tokenize(ctx, slot.stopping_word, false);
                 probs = std::vector<completion_token_output>(slot.generated_token_probs.begin(), slot.generated_token_probs.end() - stop_word_toks.size());
             }
             else
@@ -1409,10 +1369,11 @@ struct llama_server_context
                     batch.n_seq_id + i,
                     batch.seq_id   + i,
                     batch.logits   + i,
+                    0, 0, 0, // unused
                 };
                 if (llama_decode(ctx, batch_view))
                 {
-                    LOG("%s : failed to eval\n", __func__);
+                    LOG_TEE("%s : failed to eval\n", __func__);
                     return false;
                 }
             }
@@ -1427,18 +1388,17 @@ struct llama_server_context
                 }
 
                 const int n_embd = llama_n_embd(model);
-                float * embd = img.image_embedding + i * n_embd;
-                llava_embd_batch llava_batch = llava_embd_batch(embd, n_eval, slot.n_past, 0);
-                if (llama_decode(ctx, llava_batch.batch))
+                llama_batch batch_img = { n_eval, nullptr, (img.image_embedding + i * n_embd), nullptr, nullptr, nullptr, nullptr, slot.n_past, 1, 0, };
+                if (llama_decode(ctx, batch_img))
                 {
-                    LOG("%s : failed to eval image\n", __func__);
+                    LOG_TEE("%s : failed to eval image\n", __func__);
                     return false;
                 }
                 slot.n_past += n_eval;
             }
             image_idx++;
 
-            common_batch_clear(batch);
+            llama_batch_clear(batch);
 
             // append prefix of next image
             const auto json_prompt = (image_idx >= (int) slot.images.size()) ?
@@ -1448,7 +1408,7 @@ struct llama_server_context
             std::vector<llama_token> append_tokens = tokenize(json_prompt, false); // has next image
             for (int i = 0; i < (int) append_tokens.size(); ++i)
             {
-                common_batch_add(batch, append_tokens[i], system_tokens.size() + slot.n_past, { slot.id }, true);
+                llama_batch_add(batch, append_tokens[i], system_tokens.size() + slot.n_past, { slot.id }, true);
                 slot.n_past += 1;
             }
         }
@@ -1580,7 +1540,7 @@ struct llama_server_context
             update_system_prompt();
         }
 
-        common_batch_clear(batch);
+        llama_batch_clear(batch);
 
         if (all_slots_are_idle)
         {
@@ -1614,7 +1574,7 @@ struct llama_server_context
                     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());
+                    LOG_TEE("Context exhausted. Slot %d released (%d tokens in cache)\n", slot.id, (int) slot.cache_tokens.size());
 
                     continue;
                     // END LOCALAI changes
@@ -1658,7 +1618,7 @@ struct llama_server_context
 
             // TODO: we always have to take into account the "system_tokens"
             //       this is not great and needs to be improved somehow
-            common_batch_add(batch, slot.sampled, system_tokens.size() + slot_npast, { slot.id }, true);
+            llama_batch_add(batch, slot.sampled, system_tokens.size() + slot_npast, { slot.id }, true);
             slot.n_past += 1;
         }
 
@@ -1752,7 +1712,7 @@ struct llama_server_context
 
                     if (!slot.params.cache_prompt)
                     {
-                        common_sampler_reset(slot.ctx_sampling);
+                        llama_sampling_reset(slot.ctx_sampling);
 
                         slot.n_past = 0;
                         slot.n_past_se = 0;
@@ -1764,7 +1724,7 @@ struct llama_server_context
                         // push the prompt into the sampling context (do not apply grammar)
                         for (auto &token : prompt_tokens)
                         {
-                            common_sampler_accept(slot.ctx_sampling, token, false);
+                            llama_sampling_accept(slot.ctx_sampling, ctx, token, false);
                         }
 
                         slot.n_past = common_part(slot.cache_tokens, prompt_tokens);
@@ -1856,17 +1816,16 @@ struct llama_server_context
                                 ga_i += ga_w/ga_n;
                             }
                         }
-                        common_batch_add(batch, prefix_tokens[slot.n_past], system_tokens.size() + slot_npast, {slot.id }, false);
+                        llama_batch_add(batch, prefix_tokens[slot.n_past], system_tokens.size() + slot_npast, {slot.id }, false);
                         slot_npast++;
                     }
 
                     if (has_images && !ingest_images(slot, n_batch))
                     {
-                        LOG_ERR("%s: failed processing images Slot id : %d, Task id: %d", 
-                            __func__,
-                            slot.id,
-                            slot.task_id
-                        );
+                        LOG_ERROR("failed processing images", {
+                            "slot_id", slot.id,
+                            "task_id", slot.task_id,
+                        });
                         // FIXME @phymbert: to be properly tested
                         //  early returning without changing the slot state will block the slot for ever
                         // no one at the moment is checking the return value
@@ -1906,10 +1865,10 @@ struct llama_server_context
                         const int bd = (slot.ga_w / slot.ga_n) * (slot.ga_n - 1);
                         const int dd = (slot.ga_w / slot.ga_n) - ib * bd - slot.ga_w;
 
-                        LOG("\n");
-                        LOG("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i, slot.n_past_se, ib * bd, slot.ga_i + ib * bd, slot.n_past_se + ib * bd);
-                        LOG("div:   [%6d, %6d] / %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd, slot.ga_i + ib * bd + slot.ga_w, slot.ga_n, (slot.ga_i + ib * bd) / slot.ga_n, (slot.ga_i + ib * bd + slot.ga_w) / slot.ga_n);
-                        LOG("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd + slot.ga_w, slot.n_past_se + ib * bd, dd, slot.ga_i + ib * bd + slot.ga_w + dd, slot.n_past_se + ib * bd + dd);
+                        LOG_TEE("\n");
+                        LOG_TEE("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i, slot.n_past_se, ib * bd, slot.ga_i + ib * bd, slot.n_past_se + ib * bd);
+                        LOG_TEE("div:   [%6d, %6d] / %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd, slot.ga_i + ib * bd + slot.ga_w, slot.ga_n, (slot.ga_i + ib * bd) / slot.ga_n, (slot.ga_i + ib * bd + slot.ga_w) / slot.ga_n);
+                        LOG_TEE("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd + slot.ga_w, slot.n_past_se + ib * bd, dd, slot.ga_i + ib * bd + slot.ga_w + dd, slot.n_past_se + ib * bd + dd);
 
                         llama_kv_cache_seq_add(ctx, slot.id, slot.ga_i, slot.n_past_se, ib * bd);
                         llama_kv_cache_seq_div(ctx, slot.id, slot.ga_i + ib * bd, slot.ga_i + ib * bd + slot.ga_w,slot.ga_n);
@@ -1919,7 +1878,7 @@ struct llama_server_context
 
                         slot.ga_i += slot.ga_w / slot.ga_n;
 
-                        LOG("\nn_past_old = %d, n_past = %d, ga_i = %d\n\n", slot.n_past_se + bd, slot.n_past_se, slot.ga_i);
+                        LOG_TEE("\nn_past_old = %d, n_past = %d, ga_i = %d\n\n", slot.n_past_se + bd, slot.n_past_se, slot.ga_i);
                     }
                     slot.n_past_se += n_tokens;
                 }
@@ -1934,6 +1893,7 @@ struct llama_server_context
                 batch.n_seq_id + i,
                 batch.seq_id   + i,
                 batch.logits   + i,
+                0, 0, 0, // unused
             };
 
             const int ret = llama_decode(ctx, batch_view);
@@ -1943,11 +1903,11 @@ struct llama_server_context
                 if (n_batch == 1 || ret < 0)
                 {
                     // if you get here, it means the KV cache is full - try increasing it via the context size
-                    LOG("%s : failed to decode the batch, n_batch = %d, ret = %d\n", __func__, n_batch, ret);
+                    LOG_TEE("%s : failed to decode the batch, n_batch = %d, ret = %d\n", __func__, n_batch, ret);
                     return false;
                 }
 
-                LOG("%s : failed to find free space in the KV cache, retrying with smaller n_batch = %d\n", __func__, n_batch / 2);
+                LOG_TEE("%s : failed to find free space in the KV cache, retrying with smaller n_batch = %d\n", __func__, n_batch / 2);
 
                 // retry with half the batch size to try to find a free slot in the KV cache
                 n_batch /= 2;
@@ -1972,9 +1932,9 @@ struct llama_server_context
                 }
 
                 completion_token_output result;
-                const llama_token id = common_sampler_sample(slot.ctx_sampling, ctx, slot.i_batch - i);
+                const llama_token id = llama_sampling_sample(slot.ctx_sampling, ctx, NULL, slot.i_batch - i);
 
-                common_sampler_accept(slot.ctx_sampling, id, true);
+                llama_sampling_accept(slot.ctx_sampling, ctx, id, true);
 
                 slot.n_decoded += 1;
                 if (slot.n_decoded == 1)
@@ -1984,14 +1944,19 @@ struct llama_server_context
                     metrics.on_prompt_eval(slot);
                 }
 
+                llama_token_data_array cur_p = { slot.ctx_sampling->cur.data(), slot.ctx_sampling->cur.size(), false };
                 result.tok = id;
-                const auto * cur_p = common_sampler_get_candidates(slot.ctx_sampling);
 
-                for (size_t i = 0; i < (size_t) slot.sparams.n_probs; ++i) {
-                    result.probs.push_back({
-                        cur_p->data[i].id,
-                        i >= cur_p->size ? 0.0f : cur_p->data[i].p,
-                    });
+                const int32_t n_probs = slot.sparams.n_probs;
+                if (slot.sparams.temp <= 0 && n_probs > 0)
+                {
+                    // for llama_sample_token_greedy we need to sort candidates
+                    llama_sample_softmax(ctx, &cur_p);
+                }
+
+                for (size_t i = 0; i < std::min(cur_p.size, (size_t)n_probs); ++i)
+                {
+                    result.probs.push_back({cur_p.data[i].id, cur_p.data[i].p});
                 }
 
                 if (!process_token(result, slot))
@@ -2038,7 +2003,7 @@ static json format_partial_response(
 struct token_translator
 {
     llama_context * ctx;
-    std::string operator()(llama_token tok)                    const { return common_token_to_piece(ctx, tok); }
+    std::string operator()(llama_token tok)                    const { return llama_token_to_piece(ctx, tok); }
     std::string operator()(const completion_token_output &cto) const { return (*this)(cto.tok); }
 };
 
@@ -2103,6 +2068,7 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
     //     slot->params.n_predict        = json_value(data, "n_predict",         default_params.n_predict);
     //     slot->sparams.top_k           = json_value(data, "top_k",             default_sparams.top_k);
     //     slot->sparams.top_p           = json_value(data, "top_p",             default_sparams.top_p);
+    //     slot->sparams.tfs_z           = json_value(data, "tfs_z",             default_sparams.tfs_z);
     //     slot->sparams.typical_p       = json_value(data, "typical_p",         default_sparams.typical_p);
     //     slot->sparams.temp            = json_value(data, "temperature",       default_sparams.temp);
     //     slot->sparams.penalty_last_n  = json_value(data, "repeat_last_n",     default_sparams.penalty_last_n);
@@ -2126,6 +2092,7 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
     data["n_predict"] = predict->tokens() == 0 ? -1 : predict->tokens();
     data["top_k"] = predict->topk();
     data["top_p"] = predict->topp();
+    data["tfs_z"] = predict->tailfreesamplingz();
     data["typical_p"] = predict->typicalp();
     data["temperature"] = predict->temperature();
     data["repeat_last_n"] = predict->repeat();
@@ -2143,9 +2110,6 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
     data["ignore_eos"] = predict->ignoreeos();
     data["embeddings"] = predict->embeddings();
 
-    // Add the correlationid to json data
-    data["correlation_id"] = predict->correlationid();
-
     // for each image in the request, add the image data
     //
     for (int i = 0; i < predict->images_size(); i++) {
@@ -2172,6 +2136,7 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
 //     llama.params.n_predict = predict->tokens() == 0 ? -1 : predict->tokens();
 //     llama.params.sparams.top_k = predict->topk();
 //     llama.params.sparams.top_p = predict->topp();
+//     llama.params.sparams.tfs_z = predict->tailfreesamplingz();
 //     llama.params.sparams.typical_p = predict->typicalp();
 //     llama.params.sparams.penalty_last_n = predict->repeat();
 //     llama.params.sparams.temp = predict->temperature();
@@ -2229,7 +2194,7 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, llama
 // }
 
 static void params_parse(const backend::ModelOptions* request,
-                                common_params & params) {
+                                gpt_params & params) {
    
     // this is comparable to: https://github.com/ggerganov/llama.cpp/blob/d9b33fe95bd257b36c84ee5769cc048230067d6f/examples/server/server.cpp#L1809
 
@@ -2243,7 +2208,7 @@ static void params_parse(const backend::ModelOptions* request,
     params.model_alias =  request->modelfile();
     params.n_ctx = request->contextsize();
     //params.memory_f16 = request->f16memory();
-    params.cpuparams.n_threads = request->threads();
+    params.n_threads = request->threads();
     params.n_gpu_layers = request->ngpulayers();
     params.n_batch = request->nbatch();
     // Set params.n_parallel by environment variable (LLAMA_PARALLEL), defaults to 1
@@ -2293,13 +2258,13 @@ static void params_parse(const backend::ModelOptions* request,
      }
      // get the directory of modelfile
      std::string model_dir = params.model.substr(0, params.model.find_last_of("/\\"));
-     params.lora_adapters.push_back({ model_dir + "/"+request->loraadapter(), scale_factor });
+     params.lora_adapter.push_back(std::make_tuple(model_dir + "/"+request->loraadapter(), scale_factor));
+     params.lora_base  =  model_dir + "/"+request->lorabase();
     }
     params.use_mlock = request->mlock();
     params.use_mmap = request->mmap();
     params.flash_attn = request->flashattention();
     params.no_kv_offload = request->nokvoffload();
-    params.ctx_shift = false; // We control context-shifting in any case (and we disable it as it could just lead to infinite loops)
 
     params.embedding = request->embeddings();
 
@@ -2338,7 +2303,7 @@ public:
 
   grpc::Status LoadModel(ServerContext* context, const backend::ModelOptions* request, backend::Result* result) {
     // Implement LoadModel RPC
-    common_params params;
+    gpt_params params;
     params_parse(request, params);
 
     llama_backend_init();
@@ -2384,11 +2349,6 @@ public:
                 int32_t tokens_evaluated = result.result_json.value("tokens_evaluated", 0);
                 reply.set_prompt_tokens(tokens_evaluated);
 
-                // Log Request Correlation Id
-                LOG_VERBOSE("correlation:", {
-                    { "id", data["correlation_id"] }
-                });
-
                 // Send the reply
                 writer->Write(reply);
 
@@ -2412,12 +2372,6 @@ public:
         std::string completion_text;
         task_result result = llama.queue_results.recv(task_id);
         if (!result.error && result.stop) {
-            
-            // Log Request Correlation Id
-            LOG_VERBOSE("correlation:", {
-                { "id", data["correlation_id"] }
-            });
-
             completion_text = result.result_json.value("content", "");
             int32_t tokens_predicted = result.result_json.value("tokens_predicted", 0);
             int32_t tokens_evaluated = result.result_json.value("tokens_evaluated", 0);
@@ -2457,31 +2411,6 @@ public:
 
         return grpc::Status::OK;
     }
-
-    grpc::Status GetMetrics(ServerContext* context, const backend::MetricsRequest* request, backend::MetricsResponse* response) {
-        llama_client_slot* active_slot = llama.get_active_slot();
-
-        if (active_slot != nullptr) {
-            // Calculate the tokens per second using existing logic
-            double tokens_per_second = 1e3 / active_slot->t_token_generation * active_slot->n_decoded;
-
-            // Populate the response with metrics
-            response->set_slot_id(active_slot->id);
-            response->set_prompt_json_for_slot(active_slot->prompt.dump());
-            response->set_tokens_per_second(tokens_per_second);
-            response->set_tokens_generated(active_slot->n_decoded);
-            response->set_prompt_tokens_processed(active_slot->num_prompt_tokens_processed);
-        } else {
-            // Handle case when no active slot exists
-            response->set_slot_id(0);
-            response->set_prompt_json_for_slot("");
-            response->set_tokens_per_second(0);
-            response->set_tokens_generated(0);
-            response->set_prompt_tokens_processed(0);
-        }
-
-        return grpc::Status::OK;
-    } 
 };
 
 void RunServer(const std::string& server_address) {

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

@@ -1,13 +0,0 @@
-diff --git a/examples/llava/clip.cpp b/examples/llava/clip.cpp
-index 342042ff..224db9b5 100644
---- a/examples/llava/clip.cpp
-+++ b/examples/llava/clip.cpp
-@@ -2419,7 +2419,7 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
-             struct ggml_tensor * patches = ggml_graph_get_tensor(gf, "patches");
-             int* patches_data = (int*)malloc(ggml_nbytes(patches));
-             for (int i = 0; i < num_patches; i++) {
--                patches_data[i] = i + 1;
-+                patches_data[i] = i;
-             }
-             ggml_backend_tensor_set(patches, patches_data, 0, ggml_nbytes(patches));
-             free(patches_data);

backend/cpp/llama/prepare.sh

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

backend/cpp/llama/utils.hpp

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

backend/go/bark/Makefile

@@ -1,25 +0,0 @@
-INCLUDE_PATH := $(abspath ./)
-LIBRARY_PATH := $(abspath ./)
-
-AR?=ar
-
-BUILD_TYPE?=
-# keep standard at C11 and C++11
-CXXFLAGS = -I. -I$(INCLUDE_PATH)/../../../sources/bark.cpp/examples -I$(INCLUDE_PATH)/../../../sources/bark.cpp/spm-headers -I$(INCLUDE_PATH)/../../../sources/bark.cpp -O3 -DNDEBUG -std=c++17 -fPIC
-LDFLAGS  = -L$(LIBRARY_PATH) -L$(LIBRARY_PATH)/../../../sources/bark.cpp/build/examples -lbark -lstdc++ -lm
-
-# warnings
-CXXFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function
-
-gobark.o:
-	$(CXX) $(CXXFLAGS) gobark.cpp -o gobark.o -c $(LDFLAGS)
-
-libbark.a: gobark.o
-	cp $(INCLUDE_PATH)/../../../sources/bark.cpp/build/libbark.a ./
-	$(AR) rcs libbark.a gobark.o
-	$(AR) rcs libbark.a $(LIBRARY_PATH)/../../../sources/bark.cpp/build/encodec.cpp/ggml/src/CMakeFiles/ggml.dir/ggml.c.o
-	$(AR) rcs libbark.a $(LIBRARY_PATH)/../../../sources/bark.cpp/build/encodec.cpp/ggml/src/CMakeFiles/ggml.dir/ggml-alloc.c.o
-	$(AR) rcs libbark.a $(LIBRARY_PATH)/../../../sources/bark.cpp/build/encodec.cpp/ggml/src/CMakeFiles/ggml.dir/ggml-backend.c.o
-
-clean:
-	rm -f gobark.o libbark.a

backend/go/bark/gobark.cpp

@@ -1,85 +0,0 @@
-#include <iostream>
-#include <tuple>
-
-#include "bark.h"
-#include "gobark.h"
-#include "common.h"
-#include "ggml.h"
-
-struct bark_context *c;
-
-void bark_print_progress_callback(struct bark_context *bctx, enum bark_encoding_step step, int progress, void *user_data) {
-    if (step == bark_encoding_step::SEMANTIC) {
-        printf("\rGenerating semantic tokens... %d%%", progress);
-    } else if (step == bark_encoding_step::COARSE) {
-        printf("\rGenerating coarse tokens... %d%%", progress);
-    } else if (step == bark_encoding_step::FINE) {
-        printf("\rGenerating fine tokens... %d%%", progress);
-    }
-    fflush(stdout);
-}
-
-int load_model(char *model) {
-    // initialize bark context
-    struct bark_context_params ctx_params = bark_context_default_params();
-    bark_params params;
-
-    params.model_path = model;
-
-   // ctx_params.verbosity = verbosity;
-    ctx_params.progress_callback = bark_print_progress_callback;
-    ctx_params.progress_callback_user_data = nullptr;
-
-    struct bark_context *bctx = bark_load_model(params.model_path.c_str(), ctx_params, params.seed);
-    if (!bctx) {
-        fprintf(stderr, "%s: Could not load model\n", __func__);
-        return 1;
-    }
-
-    c = bctx;
-
-    return 0;
-}
-
-int tts(char *text,int  threads, char *dst ) {
-
-    ggml_time_init();
-    const int64_t t_main_start_us = ggml_time_us();
-
-    // generate audio
-    if (!bark_generate_audio(c, text, threads)) {
-        fprintf(stderr, "%s: An error occured. If the problem persists, feel free to open an issue to report it.\n", __func__);
-        return 1;
-    }
-
-    const float *audio_data = bark_get_audio_data(c);
-    if (audio_data == NULL) {
-        fprintf(stderr, "%s: Could not get audio data\n", __func__);
-        return 1;
-    }
-
-    const int audio_arr_size = bark_get_audio_data_size(c);
-
-    std::vector<float> audio_arr(audio_data, audio_data + audio_arr_size);
-
-    write_wav_on_disk(audio_arr, dst);
-
-    // report timing
-    {
-        const int64_t t_main_end_us = ggml_time_us();
-        const int64_t t_load_us = bark_get_load_time(c);
-        const int64_t t_eval_us = bark_get_eval_time(c);
-
-        printf("\n\n");
-        printf("%s:     load time = %8.2f ms\n", __func__, t_load_us / 1000.0f);
-        printf("%s:     eval time = %8.2f ms\n", __func__, t_eval_us / 1000.0f);
-        printf("%s:    total time = %8.2f ms\n", __func__, (t_main_end_us - t_main_start_us) / 1000.0f);
-    }
-    
-    return 0;
-}
-
-int unload() {
-    bark_free(c);
-}
-

backend/go/bark/gobark.go

@@ -1,52 +0,0 @@
-package main
-
-// #cgo CXXFLAGS: -I${SRCDIR}/../../../sources/bark.cpp/ -I${SRCDIR}/../../../sources/bark.cpp/encodec.cpp -I${SRCDIR}/../../../sources/bark.cpp/examples -I${SRCDIR}/../../../sources/bark.cpp/spm-headers
-// #cgo LDFLAGS: -L${SRCDIR}/ -L${SRCDIR}/../../../sources/bark.cpp/build/examples -L${SRCDIR}/../../../sources/bark.cpp/build/encodec.cpp/ -lbark -lencodec -lcommon
-// #include <gobark.h>
-// #include <stdlib.h>
-import "C"
-
-import (
-	"fmt"
-	"unsafe"
-
-	"github.com/mudler/LocalAI/pkg/grpc/base"
-	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
-)
-
-type Bark struct {
-	base.SingleThread
-	threads int
-}
-
-func (sd *Bark) Load(opts *pb.ModelOptions) error {
-
-	sd.threads = int(opts.Threads)
-
-	modelFile := C.CString(opts.ModelFile)
-	defer C.free(unsafe.Pointer(modelFile))
-
-	ret := C.load_model(modelFile)
-	if ret != 0 {
-		return fmt.Errorf("inference failed")
-	}
-
-	return nil
-}
-
-func (sd *Bark) TTS(opts *pb.TTSRequest) error {
-	t := C.CString(opts.Text)
-	defer C.free(unsafe.Pointer(t))
-
-	dst := C.CString(opts.Dst)
-	defer C.free(unsafe.Pointer(dst))
-
-	threads := C.int(sd.threads)
-
-	ret := C.tts(t, threads, dst)
-	if ret != 0 {
-		return fmt.Errorf("inference failed")
-	}
-
-	return nil
-}

backend/go/bark/gobark.h

@@ -1,8 +0,0 @@
-#ifdef __cplusplus
-extern "C" {
-#endif
-int load_model(char *model);
-int tts(char *text,int  threads, char *dst );
-#ifdef __cplusplus
-}
-#endif

backend/go/image/stablediffusion-ggml/Makefile

@@ -1,21 +0,0 @@
-INCLUDE_PATH := $(abspath ./)
-LIBRARY_PATH := $(abspath ./)
-
-AR?=ar
-
-BUILD_TYPE?=
-# keep standard at C11 and C++11
-CXXFLAGS = -I. -I$(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
-
-# warnings
-CXXFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function
-
-gosd.o:
-	$(CXX) $(CXXFLAGS) gosd.cpp -o gosd.o -c
-
-libsd.a: gosd.o
-	cp $(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp/build/libstable-diffusion.a ./libsd.a
-	$(AR) rcs libsd.a gosd.o
-
-clean:
-	rm -f gosd.o libsd.a

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

@@ -1,228 +0,0 @@
-#include <stdio.h>
-#include <string.h>
-#include <time.h>
-#include <iostream>
-#include <random>
-#include <string>
-#include <vector>
-#include "gosd.h"
-
-// #include "preprocessing.hpp"
-#include "flux.hpp"
-#include "stable-diffusion.h"
-
-#define STB_IMAGE_IMPLEMENTATION
-#define STB_IMAGE_STATIC
-#include "stb_image.h"
-
-#define STB_IMAGE_WRITE_IMPLEMENTATION
-#define STB_IMAGE_WRITE_STATIC
-#include "stb_image_write.h"
-
-#define STB_IMAGE_RESIZE_IMPLEMENTATION
-#define STB_IMAGE_RESIZE_STATIC
-#include "stb_image_resize.h"
-
-// Names of the sampler method, same order as enum sample_method in stable-diffusion.h
-const char* sample_method_str[] = {
-    "euler_a",
-    "euler",
-    "heun",
-    "dpm2",
-    "dpm++2s_a",
-    "dpm++2m",
-    "dpm++2mv2",
-    "ipndm",
-    "ipndm_v",
-    "lcm",
-};
-
-// Names of the sigma schedule overrides, same order as sample_schedule in stable-diffusion.h
-const char* schedule_str[] = {
-    "default",
-    "discrete",
-    "karras",
-    "exponential",
-    "ays",
-    "gits",
-};
-
-sd_ctx_t* sd_c;
-
-sample_method_t sample_method;
-
-int load_model(char *model, char* options[], int threads, int diff) {
-    fprintf (stderr, "Loading model!\n");
-
-    char *stableDiffusionModel = "";
-    if (diff == 1 ) {
-        stableDiffusionModel = model;
-        model = "";
-    }
-
-    // decode options. Options are in form optname:optvale, or if booleans only optname.
-    char *clip_l_path  = "";
-    char *clip_g_path  = "";
-    char *t5xxl_path  = "";
-    char *vae_path  = "";
-    char *scheduler = "";
-    char *sampler = "";
-
-    // If options is not NULL, parse options
-    for (int i = 0; options[i] != NULL; i++) {
-        char *optname = strtok(options[i], ":");
-        char *optval = strtok(NULL, ":");
-        if (optval == NULL) {
-            optval = "true";
-        }
-
-        if (!strcmp(optname, "clip_l_path")) {
-            clip_l_path = optval;
-        }
-        if (!strcmp(optname, "clip_g_path")) {
-            clip_g_path = optval;
-        }
-        if (!strcmp(optname, "t5xxl_path")) {
-            t5xxl_path = optval;
-        }
-        if (!strcmp(optname, "vae_path")) {
-            vae_path = optval;
-        }
-        if (!strcmp(optname, "scheduler")) {
-            scheduler = optval;
-        }
-        if (!strcmp(optname, "sampler")) {
-            sampler = optval;
-        }
-    }
-
-    int sample_method_found = -1;
-    for (int m = 0; m < N_SAMPLE_METHODS; m++) {
-        if (!strcmp(sampler, sample_method_str[m])) {
-            sample_method_found = m;
-        }
-    }
-    if (sample_method_found == -1) {
-        fprintf(stderr, "Invalid sample method, default to EULER_A!\n");
-        sample_method_found = EULER_A;
-    }
-    sample_method = (sample_method_t)sample_method_found;
-
-    int schedule_found            = -1;
-    for (int d = 0; d < N_SCHEDULES; d++) {
-        if (!strcmp(scheduler, schedule_str[d])) {
-            schedule_found = d;
-                fprintf (stderr, "Found scheduler: %s\n", scheduler);
-
-        }
-    }
-
-    if (schedule_found == -1) {
-        fprintf (stderr, "Invalid scheduler! using DEFAULT\n");
-        schedule_found = DEFAULT;
-    }
-
-    schedule_t schedule = (schedule_t)schedule_found;
-    
-    fprintf (stderr, "Creating context\n");
-    sd_ctx_t* sd_ctx = new_sd_ctx(model,
-                                  clip_l_path,
-                                  clip_g_path,
-                                  t5xxl_path,
-                                  stableDiffusionModel,
-                                  vae_path,
-                                  "",
-                                  "",
-                                  "",
-                                  "",
-                                  "",
-                                  false,
-                                  false,
-                                  false,
-                                  threads,
-                                  SD_TYPE_COUNT,
-                                  STD_DEFAULT_RNG,
-                                  schedule,
-                                  false,
-                                  false,
-                                  false,
-                                  false);
-
-    if (sd_ctx == NULL) {
-        fprintf (stderr, "failed loading model (generic error)\n");
-        return 1;
-    }
-    fprintf (stderr, "Created context: OK\n");
-
-    sd_c = sd_ctx;
-
-    return 0;
-}
-
-int gen_image(char *text, char *negativeText, int width, int height, int steps, int seed , char *dst, float cfg_scale) {
-
-    sd_image_t* results;
-
-    std::vector<int> skip_layers = {7, 8, 9};
-
-    fprintf (stderr, "Generating image\n");
-
-    results = txt2img(sd_c,
-                            text,
-                            negativeText,
-                            -1, //clip_skip
-                            cfg_scale, // sfg_scale
-                            3.5f,
-                            width,
-                            height,
-                            sample_method, 
-                            steps,
-                            seed,
-                            1,
-                            NULL,
-                            0.9f,
-                            20.f,
-                            false,
-                            "",
-                            skip_layers.data(),
-                            skip_layers.size(),
-                            0,
-                            0.01,
-                            0.2);
-
-    if (results == NULL) {
-        fprintf (stderr, "NO results\n");
-        return 1;
-    }
-
-    if (results[0].data == NULL) {
-        fprintf (stderr, "Results with no data\n");
-        return 1;
-    }
-
-    fprintf (stderr, "Writing PNG\n");
-
-    fprintf (stderr, "DST: %s\n", dst);
-    fprintf (stderr, "Width: %d\n", results[0].width);
-    fprintf (stderr, "Height: %d\n", results[0].height);
-    fprintf (stderr, "Channel: %d\n", results[0].channel);
-    fprintf (stderr, "Data: %p\n", results[0].data);
-
-    stbi_write_png(dst, results[0].width, results[0].height, results[0].channel,
-                       results[0].data, 0, NULL);
-    fprintf (stderr, "Saved resulting image to '%s'\n", dst);
-
-    // TODO: free results. Why does it crash?
-
-    free(results[0].data);
-    results[0].data = NULL;
-    free(results);
-    fprintf (stderr, "gen_image is done", dst);
-
-    return 0;
-}
-
-int unload() {
-    free_sd_ctx(sd_c);
-}
-

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

@@ -1,96 +0,0 @@
-package main
-
-// #cgo CXXFLAGS: -I${SRCDIR}/../../../../sources/stablediffusion-ggml.cpp/thirdparty -I${SRCDIR}/../../../../sources/stablediffusion-ggml.cpp -I${SRCDIR}/../../../../sources/stablediffusion-ggml.cpp/ggml/include
-// #cgo LDFLAGS: -L${SRCDIR}/ -L${SRCDIR}/../../../../sources/stablediffusion-ggml.cpp/build/ggml/src/ggml-cpu -L${SRCDIR}/../../../../sources/stablediffusion-ggml.cpp/build/ggml/src -lsd -lstdc++ -lm -lggml -lggml-base -lggml-cpu -lgomp
-// #include <gosd.h>
-// #include <stdlib.h>
-import "C"
-
-import (
-	"fmt"
-	"os"
-	"path/filepath"
-	"strings"
-	"unsafe"
-
-	"github.com/mudler/LocalAI/pkg/grpc/base"
-	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
-	"github.com/mudler/LocalAI/pkg/utils"
-)
-
-type SDGGML struct {
-	base.SingleThread
-	threads      int
-	sampleMethod string
-	cfgScale     float32
-}
-
-func (sd *SDGGML) Load(opts *pb.ModelOptions) error {
-
-	sd.threads = int(opts.Threads)
-
-	modelFile := C.CString(opts.ModelFile)
-	defer C.free(unsafe.Pointer(modelFile))
-
-	var options **C.char
-	// prepare the options array to pass to C
-
-	size := C.size_t(unsafe.Sizeof((*C.char)(nil)))
-	length := C.size_t(len(opts.Options))
-	options = (**C.char)(C.malloc(length * size))
-	view := (*[1 << 30]*C.char)(unsafe.Pointer(options))[0:len(opts.Options):len(opts.Options)]
-
-	var diffusionModel int
-
-	var oo []string
-	for _, op := range opts.Options {
-		if op == "diffusion_model" {
-			diffusionModel = 1
-			continue
-		}
-
-		// If it's an option path, we resolve absolute path from the model path
-		if strings.Contains(op, ":") && strings.Contains(op, "path") {
-			data := strings.Split(op, ":")
-			data[1] = filepath.Join(opts.ModelPath, data[1])
-			if err := utils.VerifyPath(data[1], opts.ModelPath); err == nil {
-				oo = append(oo, strings.Join(data, ":"))
-			}
-		} else {
-			oo = append(oo, op)
-		}
-	}
-
-	fmt.Fprintf(os.Stderr, "Options: %+v\n", oo)
-
-	for i, x := range oo {
-		view[i] = C.CString(x)
-	}
-
-	sd.cfgScale = opts.CFGScale
-
-	ret := C.load_model(modelFile, options, C.int(opts.Threads), C.int(diffusionModel))
-	if ret != 0 {
-		return fmt.Errorf("could not load model")
-	}
-
-	return nil
-}
-
-func (sd *SDGGML) GenerateImage(opts *pb.GenerateImageRequest) error {
-	t := C.CString(opts.PositivePrompt)
-	defer C.free(unsafe.Pointer(t))
-
-	dst := C.CString(opts.Dst)
-	defer C.free(unsafe.Pointer(dst))
-
-	negative := C.CString(opts.NegativePrompt)
-	defer C.free(unsafe.Pointer(negative))
-
-	ret := C.gen_image(t, negative, C.int(opts.Width), C.int(opts.Height), C.int(opts.Step), C.int(opts.Seed), dst, C.float(sd.cfgScale))
-	if ret != 0 {
-		return fmt.Errorf("inference failed")
-	}
-
-	return nil
-}

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

@@ -1,8 +0,0 @@
-#ifdef __cplusplus
-extern "C" {
-#endif
-int load_model(char *model, char* options[], int threads, int diffusionModel);
-int gen_image(char *text, char *negativeText, int width, int height, int steps, int seed, char *dst, float cfg_scale);
-#ifdef __cplusplus
-}
-#endif

backend/go/llm/bert/bert.go

@@ -0,0 +1,34 @@
+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 (
+	bert "github.com/go-skynet/go-bert.cpp"
+
+	"github.com/mudler/LocalAI/pkg/grpc/base"
+	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
+)
+
+type Embeddings struct {
+	base.SingleThread
+	bert *bert.Bert
+}
+
+func (llm *Embeddings) Load(opts *pb.ModelOptions) error {
+	model, err := bert.New(opts.ModelFile)
+	llm.bert = model
+	return err
+}
+
+func (llm *Embeddings) Embeddings(opts *pb.PredictOptions) ([]float32, error) {
+
+	if len(opts.EmbeddingTokens) > 0 {
+		tokens := []int{}
+		for _, t := range opts.EmbeddingTokens {
+			tokens = append(tokens, int(t))
+		}
+		return llm.bert.TokenEmbeddings(tokens, bert.SetThreads(int(opts.Threads)))
+	}
+
+	return llm.bert.Embeddings(opts.Embeddings, bert.SetThreads(int(opts.Threads)))
+}

backend/go/llm/bert/main.go

@@ -1,6 +1,7 @@
 package main
 
 // Note: this is started internally by LocalAI and a server is allocated for each model
+
 import (
 	"flag"
 
@@ -14,7 +15,7 @@ var (
 func main() {
 	flag.Parse()
 
-	if err := grpc.StartServer(*addr, &SDGGML{}); err != nil {
+	if err := grpc.StartServer(*addr, &Embeddings{}); err != nil {
 		panic(err)
 	}
 }

backend/go/llm/gpt4all/gpt4all.go

@@ -0,0 +1,62 @@
+package main
+
+// This is a wrapper to statisfy the GRPC service interface
+// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
+import (
+	"fmt"
+
+	"github.com/mudler/LocalAI/pkg/grpc/base"
+	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
+	gpt4all "github.com/nomic-ai/gpt4all/gpt4all-bindings/golang"
+)
+
+type LLM struct {
+	base.SingleThread
+
+	gpt4all *gpt4all.Model
+}
+
+func (llm *LLM) Load(opts *pb.ModelOptions) error {
+	model, err := gpt4all.New(opts.ModelFile,
+		gpt4all.SetThreads(int(opts.Threads)),
+		gpt4all.SetLibrarySearchPath(opts.LibrarySearchPath))
+	llm.gpt4all = model
+	return err
+}
+
+func buildPredictOptions(opts *pb.PredictOptions) []gpt4all.PredictOption {
+	predictOptions := []gpt4all.PredictOption{
+		gpt4all.SetTemperature(float64(opts.Temperature)),
+		gpt4all.SetTopP(float64(opts.TopP)),
+		gpt4all.SetTopK(int(opts.TopK)),
+		gpt4all.SetTokens(int(opts.Tokens)),
+	}
+
+	if opts.Batch != 0 {
+		predictOptions = append(predictOptions, gpt4all.SetBatch(int(opts.Batch)))
+	}
+	return predictOptions
+}
+
+func (llm *LLM) Predict(opts *pb.PredictOptions) (string, error) {
+	return llm.gpt4all.Predict(opts.Prompt, buildPredictOptions(opts)...)
+}
+
+func (llm *LLM) PredictStream(opts *pb.PredictOptions, results chan string) error {
+	predictOptions := buildPredictOptions(opts)
+
+	go func() {
+		llm.gpt4all.SetTokenCallback(func(token string) bool {
+			results <- token
+			return true
+		})
+		_, err := llm.gpt4all.Predict(opts.Prompt, predictOptions...)
+		if err != nil {
+			fmt.Println("err: ", err)
+		}
+		llm.gpt4all.SetTokenCallback(nil)
+		close(results)
+	}()
+
+	return nil
+}

backend/go/llm/gpt4all/main.go

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

backend/go/llm/rwkv/main.go

@@ -1,6 +1,7 @@
 package main
 
 // Note: this is started internally by LocalAI and a server is allocated for each model
+
 import (
 	"flag"
 
@@ -14,7 +15,7 @@ var (
 func main() {
 	flag.Parse()
 
-	if err := grpc.StartServer(*addr, &Bark{}); err != nil {
+	if err := grpc.StartServer(*addr, &LLM{}); err != nil {
 		panic(err)
 	}
 }

backend/go/llm/rwkv/rwkv.go

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

backend/go/transcribe/transcript.go

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

backend/go/transcribe/whisper.go

@@ -0,0 +1,26 @@
+package main
+
+// This is a wrapper to statisfy the GRPC service interface
+// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
+import (
+	"github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
+	"github.com/mudler/LocalAI/core/schema"
+	"github.com/mudler/LocalAI/pkg/grpc/base"
+	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
+)
+
+type Whisper struct {
+	base.SingleThread
+	whisper whisper.Model
+}
+
+func (sd *Whisper) Load(opts *pb.ModelOptions) error {
+	// Note: the Model here is a path to a directory containing the model files
+	w, err := whisper.New(opts.ModelFile)
+	sd.whisper = w
+	return err
+}
+
+func (sd *Whisper) AudioTranscription(opts *pb.TranscriptRequest) (schema.TranscriptionResult, error) {
+	return Transcript(sd.whisper, opts.Dst, opts.Language, opts.Translate, uint(opts.Threads))
+}

backend/go/transcribe/whisper/whisper.go

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

backend/go/vad/silero/vad.go

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

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

backend/python/autogptq/requirements-intel.txt

@@ -2,4 +2,4 @@
 intel-extension-for-pytorch
 torch
 optimum[openvino]
-setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
+setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406

backend/python/autogptq/requirements.txt

@@ -1,6 +1,7 @@
 accelerate
 auto-gptq==0.7.1
-grpcio==1.68.1
+grpcio==1.65.0
 protobuf
+torch
 certifi
 transformers

backend/python/bark/requirements-cpu.txt

@@ -1,4 +0,0 @@
-transformers
-accelerate
-torch==2.4.1
-torchaudio==2.4.1

backend/python/bark/requirements-cublas11.txt

@@ -1,5 +0,0 @@
---extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.4.1+cu118
-torchaudio==2.4.1+cu118
-transformers
-accelerate

backend/python/bark/requirements-cublas12.txt

@@ -1,4 +0,0 @@
-torch==2.4.1
-torchaudio==2.4.1
-transformers
-accelerate

backend/python/bark/requirements-hipblas.txt

@@ -1,5 +1,3 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch==2.4.1+rocm6.0
-torchaudio==2.4.1+rocm6.0
-transformers
-accelerate
+torch
+torchaudio

backend/python/bark/requirements-intel.txt

@@ -3,6 +3,4 @@ intel-extension-for-pytorch
 torch
 torchaudio
 optimum[openvino]
-setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
-transformers
-accelerate
+setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406

backend/python/bark/requirements.txt

@@ -1,4 +1,6 @@
+accelerate
 bark==0.1.5
-grpcio==1.68.1
+grpcio==1.65.0
 protobuf
-certifi
+certifi
+transformers

backend/python/common/libbackend.sh

@@ -18,23 +18,10 @@
 # source $(dirname $0)/../common/libbackend.sh
 #
 function init() {
-    # Name of the backend (directory name)
     BACKEND_NAME=${PWD##*/}
-
-    # Path where all backends files are
     MY_DIR=$(realpath `dirname $0`)
-
-    # Build type
     BUILD_PROFILE=$(getBuildProfile)
 
-    # Environment directory
-    EDIR=${MY_DIR}
-
-    # Allow to specify a custom env dir for shared environments
-    if [ "x${ENV_DIR}" != "x" ]; then
-        EDIR=${ENV_DIR}
-    fi
-
     # If a backend has defined a list of valid build profiles...
     if [ ! -z "${LIMIT_TARGETS}" ]; then
         isValidTarget=$(checkTargets ${LIMIT_TARGETS})
@@ -87,14 +74,13 @@ function getBuildProfile() {
 # This function is idempotent, so you can call it as many times as you want and it will
 # always result in an activated virtual environment
 function ensureVenv() {
-    if [ ! -d "${EDIR}/venv" ]; then
-        uv venv ${EDIR}/venv
+    if [ ! -d "${MY_DIR}/venv" ]; then
+        uv venv ${MY_DIR}/venv
         echo "virtualenv created"
     fi
-
-    # Source if we are not already in a Virtual env
-    if [ "x${VIRTUAL_ENV}" != "x${EDIR}/venv" ]; then
-        source ${EDIR}/venv/bin/activate
+    
+    if [ "x${VIRTUAL_ENV}" != "x${MY_DIR}/venv" ]; then
+        source ${MY_DIR}/venv/bin/activate
         echo "virtualenv activated"
     fi
 
@@ -127,24 +113,13 @@ function installRequirements() {
 
     # These are the requirements files we will attempt to install, in order
     declare -a requirementFiles=(
-        "${EDIR}/requirements-install.txt"
-        "${EDIR}/requirements.txt"
-        "${EDIR}/requirements-${BUILD_TYPE}.txt"
+        "${MY_DIR}/requirements-install.txt"
+        "${MY_DIR}/requirements.txt"
+        "${MY_DIR}/requirements-${BUILD_TYPE}.txt"
     )
 
     if [ "x${BUILD_TYPE}" != "x${BUILD_PROFILE}" ]; then
-        requirementFiles+=("${EDIR}/requirements-${BUILD_PROFILE}.txt")
-    fi
-
-    # if BUILD_TYPE is empty, we are a CPU build, so we should try to install the CPU requirements
-    if [ "x${BUILD_TYPE}" == "x" ]; then
-        requirementFiles+=("${EDIR}/requirements-cpu.txt")
-    fi
-
-    requirementFiles+=("${EDIR}/requirements-after.txt")
-
-    if [ "x${BUILD_TYPE}" != "x${BUILD_PROFILE}" ]; then
-        requirementFiles+=("${EDIR}/requirements-${BUILD_PROFILE}-after.txt")
+        requirementFiles+=("${MY_DIR}/requirements-${BUILD_PROFILE}.txt")
     fi
 
     for reqFile in ${requirementFiles[@]}; do

backend/python/common/template/Makefile

@@ -1,9 +1,8 @@
 .DEFAULT_GOAL := install
 
 .PHONY: install
-install:
+install: protogen
 	bash install.sh
-	$(MAKE) protogen
 
 .PHONY: protogen
 protogen: backend_pb2_grpc.py backend_pb2.py
@@ -13,7 +12,7 @@ protogen-clean:
 	$(RM) backend_pb2_grpc.py backend_pb2.py
 
 backend_pb2_grpc.py backend_pb2.py:
-	bash protogen.sh
+	python3 -m grpc_tools.protoc -I../.. --python_out=. --grpc_python_out=. backend.proto
 
 .PHONY: clean
 clean: protogen-clean

backend/python/common/template/protogen.sh

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

backend/python/common/template/requirements.txt

@@ -1,3 +1,2 @@
-grpcio==1.68.1
-protobuf
-grpcio-tools
+grpcio==1.65.0
+protobuf

backend/python/coqui/requirements-cpu.txt

@@ -1,4 +0,0 @@
-transformers
-accelerate
-torch==2.4.1
-coqui-tts

backend/python/coqui/requirements-cublas11.txt

@@ -1,6 +0,0 @@
---extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.4.1+cu118
-torchaudio==2.4.1+cu118
-transformers
-accelerate
-coqui-tts

backend/python/coqui/requirements-cublas12.txt

@@ -1,5 +0,0 @@
-torch==2.4.1
-torchaudio==2.4.1
-transformers
-accelerate
-coqui-tts

backend/python/coqui/requirements-hipblas.txt

@@ -1,6 +1,3 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch==2.4.1+rocm6.0
-torchaudio==2.4.1+rocm6.0
-transformers
-accelerate
-coqui-tts
+torch
+torchaudio

backend/python/coqui/requirements-intel.txt

@@ -3,7 +3,4 @@ intel-extension-for-pytorch
 torch
 torchaudio
 optimum[openvino]
-setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
-transformers
-accelerate
-coqui-tts
+setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406

backend/python/coqui/requirements.txt

@@ -1,4 +1,6 @@
-grpcio==1.68.1
+accelerate
+TTS==0.22.0
+grpcio==1.65.0
 protobuf
 certifi
-packaging==24.1
+transformers

backend/python/coqui/test.py

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

backend/python/diffusers/backend.py

@@ -18,13 +18,13 @@ import backend_pb2_grpc
 import grpc
 
 from diffusers import StableDiffusion3Pipeline, StableDiffusionXLPipeline, StableDiffusionDepth2ImgPipeline, DPMSolverMultistepScheduler, StableDiffusionPipeline, DiffusionPipeline, \
-    EulerAncestralDiscreteScheduler, FluxPipeline, FluxTransformer2DModel
+    EulerAncestralDiscreteScheduler
 from diffusers import StableDiffusionImg2ImgPipeline, AutoPipelineForText2Image, ControlNetModel, StableVideoDiffusionPipeline
 from diffusers.pipelines.stable_diffusion import safety_checker
 from diffusers.utils import load_image, export_to_video
 from compel import Compel, ReturnedEmbeddingsType
-from optimum.quanto import freeze, qfloat8, quantize
-from transformers import CLIPTextModel, T5EncoderModel
+
+from transformers import CLIPTextModel
 from safetensors.torch import load_file
 
 _ONE_DAY_IN_SECONDS = 60 * 60 * 24
@@ -163,12 +163,10 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             modelFile = request.Model
 
             self.cfg_scale = 7
-            self.PipelineType = request.PipelineType
-
             if request.CFGScale != 0:
                 self.cfg_scale = request.CFGScale
 
-            clipmodel = "Lykon/dreamshaper-8"
+            clipmodel = "runwayml/stable-diffusion-v1-5"
             if request.CLIPModel != "":
                 clipmodel = request.CLIPModel
             clipsubfolder = "text_encoder"
@@ -246,35 +244,6 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                         torch_dtype=torchType,
                         use_safetensors=True,
                         variant=variant)
-            elif request.PipelineType == "FluxPipeline":
-                if fromSingleFile:
-                    self.pipe = FluxPipeline.from_single_file(modelFile,
-                                                              torch_dtype=torchType,
-                                                              use_safetensors=True)
-                else:
-                    self.pipe = FluxPipeline.from_pretrained(
-                        request.Model,
-                        torch_dtype=torch.bfloat16)
-                if request.LowVRAM:
-                    self.pipe.enable_model_cpu_offload()
-            elif request.PipelineType == "FluxTransformer2DModel":
-                    dtype = torch.bfloat16
-                    # specify from environment or default to "ChuckMcSneed/FLUX.1-dev"
-                    bfl_repo = os.environ.get("BFL_REPO", "ChuckMcSneed/FLUX.1-dev")
-
-                    transformer = FluxTransformer2DModel.from_single_file(modelFile, torch_dtype=dtype)
-                    quantize(transformer, weights=qfloat8)
-                    freeze(transformer)
-                    text_encoder_2 = T5EncoderModel.from_pretrained(bfl_repo, subfolder="text_encoder_2", torch_dtype=dtype)
-                    quantize(text_encoder_2, weights=qfloat8)
-                    freeze(text_encoder_2)
-
-                    self.pipe = FluxPipeline.from_pretrained(bfl_repo, transformer=None, text_encoder_2=None, torch_dtype=dtype)
-                    self.pipe.transformer = transformer
-                    self.pipe.text_encoder_2 = text_encoder_2
-
-                    if request.LowVRAM:
-                        self.pipe.enable_model_cpu_offload()
 
             if CLIPSKIP and request.CLIPSkip != 0:
                 self.clip_skip = request.CLIPSkip
@@ -301,34 +270,22 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                 self.pipe.controlnet = self.controlnet
             else:
                 self.controlnet = None
-
-            if request.LoraAdapter and not os.path.isabs(request.LoraAdapter):
+            # Assume directory from request.ModelFile.
+            # Only if request.LoraAdapter it's not an absolute path
+            if request.LoraAdapter and request.ModelFile != "" and not os.path.isabs(request.LoraAdapter) and request.LoraAdapter:
+                # get base path of modelFile
+                modelFileBase = os.path.dirname(request.ModelFile)
                 # modify LoraAdapter to be relative to modelFileBase
-                request.LoraAdapter = os.path.join(request.ModelPath, request.LoraAdapter)
-
+                request.LoraAdapter = os.path.join(modelFileBase, request.LoraAdapter)
             device = "cpu" if not request.CUDA else "cuda"
             self.device = device
             if request.LoraAdapter:
                 # Check if its a local file and not a directory ( we load lora differently for a safetensor file )
                 if os.path.exists(request.LoraAdapter) and not os.path.isdir(request.LoraAdapter):
+                    # self.load_lora_weights(request.LoraAdapter, 1, device, torchType)
                     self.pipe.load_lora_weights(request.LoraAdapter)
                 else:
                     self.pipe.unet.load_attn_procs(request.LoraAdapter)
-            if len(request.LoraAdapters) > 0:
-                i = 0
-                adapters_name = []
-                adapters_weights = []
-                for adapter in request.LoraAdapters:
-                    if not os.path.isabs(adapter):
-                        adapter = os.path.join(request.ModelPath, adapter)
-                    self.pipe.load_lora_weights(adapter, adapter_name=f"adapter_{i}")
-                    adapters_name.append(f"adapter_{i}")
-                    i += 1
-
-                for adapters_weight in request.LoraScales:
-                    adapters_weights.append(adapters_weight)
-
-                self.pipe.set_adapters(adapters_name, adapter_weights=adapters_weights)
 
             if request.CUDA:
                 self.pipe.to('cuda')
@@ -409,6 +366,8 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         # create a dictionary of values for the parameters
         options = {
             "negative_prompt": request.negative_prompt,
+            "width": request.width,
+            "height": request.height,
             "num_inference_steps": steps,
         }
 
@@ -426,13 +385,13 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         keys = options.keys()
 
         if request.EnableParameters != "":
-            keys = [key.strip() for key in request.EnableParameters.split(",")]
+            keys = request.EnableParameters.split(",")
 
         if request.EnableParameters == "none":
             keys = []
 
         # create a dictionary of parameters by using the keys from EnableParameters and the values from defaults
-        kwargs = {key: options.get(key) for key in keys if key in options}
+        kwargs = {key: options[key] for key in keys}
 
         # Set seed
         if request.seed > 0:
@@ -440,19 +399,6 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                 request.seed
             )
 
-        if self.PipelineType == "FluxPipeline":
-            kwargs["max_sequence_length"] = 256
-
-        if request.width:
-            kwargs["width"] = request.width
-
-        if request.height:
-            kwargs["height"] = request.height
-
-        if self.PipelineType == "FluxTransformer2DModel":
-            kwargs["output_type"] = "pil"
-            kwargs["generator"] = torch.Generator("cpu").manual_seed(0)
-
         if self.img2vid:
             # Load the conditioning image
             image = load_image(request.src)
@@ -468,7 +414,6 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             export_to_video(video_frames, request.dst)
             return backend_pb2.Result(message="Media generated successfully", success=True)
 
-        print(f"Generating image with {kwargs=}", file=sys.stderr)
         image = {}
         if COMPEL:
             conditioning, pooled = self.compel.build_conditioning_tensor(prompt)

backend/python/diffusers/requirements-cpu.txt

@@ -1,9 +0,0 @@
-diffusers
-opencv-python
-transformers
-accelerate
-compel
-peft
-sentencepiece
-torch==2.4.1
-optimum-quanto

backend/python/diffusers/requirements-cublas11.txt

@@ -1,10 +0,0 @@
---extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.4.1+cu118
-diffusers
-opencv-python
-transformers
-accelerate
-compel
-peft
-sentencepiece
-optimum-quanto

backend/python/diffusers/requirements-cublas12.txt

@@ -1,9 +0,0 @@
-torch==2.4.1
-diffusers
-opencv-python
-transformers
-accelerate
-compel
-peft
-sentencepiece
-optimum-quanto

backend/python/diffusers/requirements-hipblas.txt

@@ -1,11 +1,3 @@
 --extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch==2.3.1+rocm6.0
-torchvision==0.18.1+rocm6.0
-diffusers
-opencv-python
-transformers
-accelerate
-compel
-peft
-sentencepiece
-optimum-quanto
+torch
+torchvision

backend/python/diffusers/requirements-intel.txt

@@ -3,12 +3,4 @@ intel-extension-for-pytorch
 torch
 torchvision
 optimum[openvino]
-setuptools==75.1.0 # https://github.com/mudler/LocalAI/issues/2406
-diffusers
-opencv-python
-transformers
-accelerate
-compel
-peft
-sentencepiece
-optimum-quanto
+setuptools==70.3.0 # https://github.com/mudler/LocalAI/issues/2406

backend/python/diffusers/requirements.txt

@@ -1,5 +1,13 @@
 setuptools
-grpcio==1.68.1
+accelerate
+compel
+peft
+diffusers
+grpcio==1.65.0
+opencv-python
 pillow
 protobuf
+sentencepiece
+torch
+transformers
 certifi

backend/python/diffusers/test.py

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

backend/python/exllama/.gitignore

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