⬆️ Update ggerganov/llama.cpp (#1655)

Signed-off-by: GitHub <noreply@github.com>
Co-authored-by: mudler <mudler@users.noreply.github.com>

.dockerignore

@@ -1 +1,5 @@
+.idea
 models
+examples/chatbot-ui/models
+examples/rwkv/models
+examples/**/models

.env

@@ -1,4 +1,89 @@
-THREADS=14
-CONTEXT_SIZE=512
+## Set number of threads.
+## Note: prefer the number of physical cores. Overbooking the CPU degrades performance notably.
+# THREADS=14
+
+## Specify a different bind address (defaults to ":8080")
+# ADDRESS=127.0.0.1:8080
+
+## Default models context size
+# CONTEXT_SIZE=512
+#
+## Define galleries.
+## models will to install will be visible in `/models/available`
+# GALLERIES=[{"name":"model-gallery", "url":"github:go-skynet/model-gallery/index.yaml"}]
+
+## CORS settings
+# CORS=true
+# CORS_ALLOW_ORIGINS=*
+
+## Default path for models
+#
 MODELS_PATH=/models
-# DEBUG=true
+
+## Enable debug mode
+# DEBUG=true
+
+## Disables COMPEL (Diffusers)
+# COMPEL=0
+
+## Enable/Disable single backend (useful if only one GPU is available)
+# SINGLE_ACTIVE_BACKEND=true
+
+## Specify a build type. Available: cublas, openblas, clblas.
+## cuBLAS: This is a GPU-accelerated version of the complete standard BLAS (Basic Linear Algebra Subprograms) library. It's provided by Nvidia and is part of their CUDA toolkit.
+## OpenBLAS: This is an open-source implementation of the BLAS library that aims to provide highly optimized code for various platforms. It includes support for multi-threading and can be compiled to use hardware-specific features for additional performance. OpenBLAS can run on many kinds of hardware, including CPUs from Intel, AMD, and ARM.
+## clBLAS:   This is an open-source implementation of the BLAS library that uses OpenCL, a framework for writing programs that execute across heterogeneous platforms consisting of CPUs, GPUs, and other processors. clBLAS is designed to take advantage of the parallel computing power of GPUs but can also run on any hardware that supports OpenCL. This includes hardware from different vendors like Nvidia, AMD, and Intel.
+# BUILD_TYPE=openblas
+
+## Uncomment and set to true to enable rebuilding from source
+# REBUILD=true
+
+## Enable go tags, available: stablediffusion, tts
+## stablediffusion: image generation with stablediffusion
+## tts: enables text-to-speech with go-piper 
+## (requires REBUILD=true)
+#
+# GO_TAGS=stablediffusion
+
+## Path where to store generated images
+# IMAGE_PATH=/tmp
+
+## Specify a default upload limit in MB (whisper)
+# UPLOAD_LIMIT
+
+## List of external GRPC backends (note on the container image this variable is already set to use extra backends available in extra/)
+# EXTERNAL_GRPC_BACKENDS=my-backend:127.0.0.1:9000,my-backend2:/usr/bin/backend.py
+
+### Advanced settings ###
+### Those are not really used by LocalAI, but from components in the stack ###
+##
+### Preload libraries
+# LD_PRELOAD=
+
+### Huggingface cache for models
+# HUGGINGFACE_HUB_CACHE=/usr/local/huggingface
+
+### Python backends GRPC max workers
+### Default number of workers for GRPC Python backends.
+### This actually controls wether a backend can process multiple requests or not.
+# PYTHON_GRPC_MAX_WORKERS=1
+
+### Define the number of parallel LLAMA.cpp workers (Defaults to 1)
+# LLAMACPP_PARALLEL=1
+
+### Enable to run parallel requests
+# PARALLEL_REQUESTS=true
+
+### Watchdog settings
+###
+# Enables watchdog to kill backends that are inactive for too much time
+# WATCHDOG_IDLE=true
+#
+# Enables watchdog to kill backends that are busy for too much time
+# WATCHDOG_BUSY=true
+#
+# Time in duration format (e.g. 1h30m) after which a backend is considered idle
+# WATCHDOG_IDLE_TIMEOUT=5m
+#
+# Time in duration format (e.g. 1h30m) after which a backend is considered busy
+# WATCHDOG_BUSY_TIMEOUT=5m

.gitattributes

@@ -0,0 +1 @@
+*.sh text eol=lf

.github/FUNDING.yml

@@ -0,0 +1,5 @@
+# These are supported funding model platforms
+
+github: [mudler]
+custom: 
+- https://www.buymeacoffee.com/mudler

.github/ISSUE_TEMPLATE/bug_report.md

@@ -0,0 +1,29 @@
+---
+name: Bug report
+about: Create a report to help us improve
+title: ''
+labels: bug, unconfirmed, up-for-grabs
+---
+
+<!-- Thanks for helping us to improve LocalAI! We welcome all bug reports. Please fill out each area of the template so we can better help you. Comments like this will be hidden when you post but you can delete them if you wish. -->
+
+**LocalAI version:**
+<!-- Container Image or LocalAI tag/commit -->
+
+**Environment, CPU architecture, OS, and Version:**
+<!-- Provide the output from "uname -a", HW specs, if it's a VM  -->
+
+**Describe the bug**
+<!-- A clear and concise description of what the bug is. -->
+
+**To Reproduce**
+<!-- Steps to reproduce the behavior, including the LocalAI command used, if any -->
+
+**Expected behavior**
+<!-- A clear and concise description of what you expected to happen. -->
+
+**Logs**
+<!-- If applicable, add logs while running LocalAI in debug mode (`--debug` or `DEBUG=true`) to help explain your problem.  -->
+
+**Additional context**
+<!-- Add any other context about the problem here. -->

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,8 @@
+blank_issues_enabled: false
+contact_links:
+  - name: Community Support
+    url: https://github.com/go-skynet/LocalAI/discussions
+    about: Please ask and answer questions here.
+  - name: Discord
+    url: https://discord.gg/uJAeKSAGDy
+    about: Join our community on Discord!

.github/ISSUE_TEMPLATE/feature_request.md

@@ -0,0 +1,20 @@
+---
+name: Feature request
+about: Suggest an idea for this project
+title: ''
+labels: enhancement, up-for-grabs
+---
+
+<!-- Thanks for helping us to improve LocalAI! We welcome all feature requests. Please fill out each area of the template so we can better help you. Comments like this will be hidden when you post but you can delete them if you wish. -->
+
+**Is your feature request related to a problem? Please describe.**
+<!-- A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]  -->
+
+**Describe the solution you'd like**
+<!-- A clear and concise description of what you want to happen.  -->
+
+**Describe alternatives you've considered**
+<!-- A clear and concise description of any alternative solutions or features you've considered.  -->
+
+**Additional context**
+<!-- Add any other context or screenshots about the feature request here. -->

.github/PULL_REQUEST_TEMPLATE.md

@@ -0,0 +1,31 @@
+**Description**
+
+This PR fixes #
+
+**Notes for Reviewers**
+
+
+**[Signed commits](../CONTRIBUTING.md#signing-off-on-commits-developer-certificate-of-origin)**
+- [ ] Yes, I signed my commits.
+ 
+<!--
+Thank you for contributing to LocalAI! 
+
+Contributing Conventions
+-------------------------
+
+The draft above helps to give a quick overview of your PR.
+
+Remember to remove this comment and to at least:
+
+1. Include descriptive PR titles with [<component-name>] prepended. We use [conventional commits](https://www.conventionalcommits.org/en/v1.0.0/).
+2. Build and test your changes before submitting a PR (`make build`). 
+3. Sign your commits
+4. **Tag maintainer:** for a quicker response, tag the relevant maintainer (see below).
+5. **X/Twitter handle:** we announce bigger features on X/Twitter. If your PR gets announced, and you'd like a mention, we'll gladly shout you out!
+
+By following the community's contribution conventions upfront, the review process will 
+be accelerated and your PR merged more quickly.
+
+If no one reviews your PR within a few days, please @-mention @mudler.
+-->

.github/bump_deps.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+set -xe
+REPO=$1
+BRANCH=$2
+VAR=$3
+
+LAST_COMMIT=$(curl -s -H "Accept: application/vnd.github.VERSION.sha" "https://api.github.com/repos/$REPO/commits/$BRANCH")
+
+sed -i Makefile -e "s/$VAR?=.*/$VAR?=$LAST_COMMIT/"

.github/bump_docs.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+set -xe
+REPO=$1
+
+LATEST_TAG=$(curl -s "https://api.github.com/repos/$REPO/releases/latest" | jq -r '.name')
+
+cat <<< $(jq ".version = \"$LATEST_TAG\"" docs/data/version.json) > docs/data/version.json

.github/release.yml

@@ -0,0 +1,24 @@
+# .github/release.yml
+
+changelog:
+  exclude:
+    labels:
+      - ignore-for-release
+  categories:
+    - title: Breaking Changes 🛠
+      labels:
+        - Semver-Major
+        - breaking-change
+    - title: "Bug fixes :bug:"
+      labels:
+        - bug
+    - title: Exciting New Features 🎉
+      labels:
+        - Semver-Minor
+        - enhancement
+    - title: 👒 Dependencies
+      labels:
+        - dependencies
+    - title: Other Changes
+      labels:
+        - "*"

.github/stale.yml

@@ -0,0 +1,18 @@
+# Number of days of inactivity before an issue becomes stale
+daysUntilStale: 45
+# Number of days of inactivity before a stale issue is closed
+daysUntilClose: 10
+# Issues with these labels will never be considered stale
+exemptLabels:
+  - issue/willfix
+# Label to use when marking an issue as stale
+staleLabel: issue/stale
+# Comment to post when marking an issue as stale. Set to `false` to disable
+markComment: >
+  This issue has been automatically marked as stale because it has not had
+  recent activity. It will be closed if no further activity occurs. Thank you
+  for your contributions.
+# Comment to post when closing a stale issue. Set to `false` to disable
+closeComment: >
+  This issue is being automatically closed due to inactivity.
+  However, you may choose to reopen this issue.

.github/workflows/bump_deps.yaml

@@ -0,0 +1,63 @@
+name: Bump dependencies
+on:
+  schedule:
+    - cron: 0 20 * * *
+  workflow_dispatch:
+jobs:
+  bump:
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - repository: "go-skynet/go-llama.cpp"
+            variable: "GOLLAMA_VERSION"
+            branch: "master"
+          - 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: "go-skynet/go-bert.cpp"
+            variable: "BERT_VERSION"
+            branch: "master"
+          - repository: "go-skynet/bloomz.cpp"
+            variable: "BLOOMZ_VERSION"
+            branch: "main"
+          - repository: "nomic-ai/gpt4all"
+            variable: "GPT4ALL_VERSION"
+            branch: "main"
+          - repository: "mudler/go-ggllm.cpp"
+            variable: "GOGGLLM_VERSION"
+            branch: "master"
+          - repository: "mudler/go-stable-diffusion"
+            variable: "STABLEDIFFUSION_VERSION"
+            branch: "master"
+          - repository: "mudler/go-piper"
+            variable: "PIPER_VERSION"
+            branch: "master"
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - name: Bump dependencies 🔧
+        run: |
+          bash .github/bump_deps.sh ${{ matrix.repository }} ${{ matrix.branch }} ${{ matrix.variable }}
+      - name: Create Pull Request
+        uses: peter-evans/create-pull-request@v5
+        with:
+          token: ${{ secrets.UPDATE_BOT_TOKEN }}
+          push-to-fork: ci-forks/LocalAI
+          commit-message: ':arrow_up: Update ${{ matrix.repository }}'
+          title: ':arrow_up: Update ${{ matrix.repository }}'
+          branch: "update/${{ matrix.variable }}"
+          body: Bump of ${{ matrix.repository }} version
+          signoff: true
+
+
+

.github/workflows/bump_docs.yaml

@@ -0,0 +1,31 @@
+name: Bump dependencies
+on:
+  schedule:
+    - cron: 0 20 * * *
+  workflow_dispatch:
+jobs:
+  bump:
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - repository: "mudler/LocalAI"
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - name: Bump dependencies 🔧
+        run: |
+          bash .github/bump_docs.sh ${{ matrix.repository }}
+      - name: Create Pull Request
+        uses: peter-evans/create-pull-request@v5
+        with:
+          token: ${{ secrets.UPDATE_BOT_TOKEN }}
+          push-to-fork: ci-forks/LocalAI
+          commit-message: ':arrow_up: Update docs version ${{ matrix.repository }}'
+          title: ':arrow_up: Update docs version ${{ matrix.repository }}'
+          branch: "update/docs"
+          body: Bump of ${{ matrix.repository }} version inside docs
+          signoff: true
+
+
+

.github/workflows/disabled/test-gpu.yml

@@ -0,0 +1,63 @@
+---
+name: 'GPU tests'
+
+on:
+  pull_request:
+  push:
+    branches:
+      - master
+    tags:
+      - '*'
+
+concurrency:
+  group: ci-gpu-tests-${{ github.head_ref || github.ref }}-${{ github.repository }}
+  cancel-in-progress: true
+
+jobs:
+  ubuntu-latest:
+    runs-on: gpu
+    strategy:
+      matrix:
+        go-version: ['1.21.x']
+    steps:
+      - name: Clone
+        uses: actions/checkout@v4
+        with: 
+          submodules: true
+      - name: Setup Go ${{ matrix.go-version }}
+        uses: actions/setup-go@v4
+        with:
+          go-version: ${{ matrix.go-version }}
+      # You can test your matrix by printing the current Go version
+      - name: Display Go version
+        run: go version
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo DEBIAN_FRONTEND=noninteractive apt-get install -y make wget
+      - name: Build
+        run: |
+          if [ ! -e /run/systemd/system ]; then
+            sudo mkdir /run/systemd/system
+          fi
+          sudo mkdir -p /host/tests/${{ github.head_ref || github.ref }}
+          sudo chmod -R 777 /host/tests/${{ github.head_ref || github.ref }}
+          make \
+            TEST_DIR="/host/tests/${{ github.head_ref || github.ref }}" \
+            BUILD_TYPE=cublas \
+            prepare-e2e run-e2e-image test-e2e
+      - name: Release space from worker ♻
+        if: always()
+        run: |
+          sudo rm -rf build || true
+          sudo rm -rf bin || true
+          sudo rm -rf dist || true
+          sudo docker logs $(sudo docker ps -q --filter ancestor=localai-tests) > logs.txt
+          sudo cat logs.txt || true
+          sudo rm -rf logs.txt
+          make clean || true
+          make \
+            TEST_DIR="/host/tests/${{ github.head_ref || github.ref }}" \
+            teardown-e2e || true
+          sudo rm -rf /host/tests/${{ github.head_ref || github.ref }} || true
+          docker system prune -f -a --volumes || true

.github/workflows/image-pr.yml

@@ -0,0 +1,86 @@
+---
+name: 'build container images tests'
+
+on:
+  pull_request:
+
+concurrency:
+  group: ci-${{ github.head_ref || github.ref }}-${{ github.repository }}
+  cancel-in-progress: true
+
+jobs:
+  extras-image-build:
+    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 }}
+    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: ${{ github.event_name != 'pull_request' && 2 || 4 }}
+      matrix:
+        include:
+          - build-type: ''
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-ffmpeg'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+          - build-type: 'cublas'
+            cuda-major-version: "12"
+            cuda-minor-version: "1"
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda12-ffmpeg'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+  core-image-build:
+    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 }}
+    secrets:
+      dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
+      dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
+      quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
+      quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
+    strategy:
+      matrix:
+        include:
+          - build-type: ''
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-ffmpeg-core'
+            ffmpeg: 'true'
+            image-type: 'core'
+            runs-on: 'ubuntu-latest'
+          - build-type: 'cublas'
+            cuda-major-version: "12"
+            cuda-minor-version: "1"
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda12-ffmpeg-core'
+            ffmpeg: 'true'
+            image-type: 'core'
+            runs-on: 'ubuntu-latest'

.github/workflows/image.yml

@@ -2,77 +2,159 @@
 name: 'build container images'
 
 on:
-  pull_request:
   push:
     branches:
       - master
     tags:
       - '*'
 
+concurrency:
+  group: ci-${{ github.head_ref || github.ref }}-${{ github.repository }}
+  cancel-in-progress: true
+
 jobs:
-  docker:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v3
-
-      - name: Prepare
-        id: prep
-        run: |
-          DOCKER_IMAGE=quay.io/go-skynet/local-ai
-          VERSION=master
-          SHORTREF=${GITHUB_SHA::8}
-
-          # If this is git tag, use the tag name as a docker tag
-          if [[ $GITHUB_REF == refs/tags/* ]]; then
-            VERSION=${GITHUB_REF#refs/tags/}
-          fi
-          TAGS="${DOCKER_IMAGE}:${VERSION},${DOCKER_IMAGE}:${SHORTREF}"
-
-          # If the VERSION looks like a version number, assume that
-          # this is the most recent version of the image and also
-          # tag it 'latest'.
-          if [[ $VERSION =~ ^v[0-9]{1,3}\.[0-9]{1,3}\.[0-9]{1,3}$ ]]; then
-            TAGS="$TAGS,${DOCKER_IMAGE}:latest"
-          fi
-
-          # Set output parameters.
-          echo ::set-output name=tags::${TAGS}
-          echo ::set-output name=docker_image::${DOCKER_IMAGE}
-
-      - name: Set up QEMU
-        uses: docker/setup-qemu-action@master
-        with:
-          platforms: all
-
-      - name: Set up Docker Buildx
-        id: buildx
-        uses: docker/setup-buildx-action@master
-
-      - name: Login to DockerHub
-        if: github.event_name != 'pull_request'
-        uses: docker/login-action@v2
-        with:
-          registry: quay.io
-          username: ${{ secrets.QUAY_USERNAME }}
-          password: ${{ secrets.QUAY_PASSWORD }}
-      - name: Build
-        if: github.event_name != 'pull_request'
-        uses: docker/build-push-action@v4
-        with:
-          builder: ${{ steps.buildx.outputs.name }}
-          context: .
-          file: ./Dockerfile
-          platforms: linux/amd64,linux/arm64
-          push: true
-          tags: ${{ steps.prep.outputs.tags }}
-      - name: Build PRs
-        if: github.event_name == 'pull_request'
-        uses: docker/build-push-action@v4
-        with:
-          builder: ${{ steps.buildx.outputs.name }}
-          context: .
-          file: ./Dockerfile
-          platforms: linux/amd64
-          push: false
-          tags: ${{ steps.prep.outputs.tags }}
+  extras-image-build:
+    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 }}
+    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: ${{ github.event_name != 'pull_request' && 2 || 4 }}
+      matrix:
+        include:
+          - build-type: ''
+            #platforms: 'linux/amd64,linux/arm64'
+            platforms: 'linux/amd64'
+            tag-latest: 'auto'
+            tag-suffix: ''
+            ffmpeg: ''
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+          - build-type: ''
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-ffmpeg'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+          - build-type: 'cublas'
+            cuda-major-version: "11"
+            cuda-minor-version: "7"
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda11'
+            ffmpeg: ''
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+          - build-type: 'cublas'
+            cuda-major-version: "12"
+            cuda-minor-version: "1"
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda12'
+            ffmpeg: ''
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+          - build-type: 'cublas'
+            cuda-major-version: "11"
+            cuda-minor-version: "7"
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda11-ffmpeg'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+          - build-type: 'cublas'
+            cuda-major-version: "12"
+            cuda-minor-version: "1"
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda12-ffmpeg'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+          - build-type: ''
+            #platforms: 'linux/amd64,linux/arm64'
+            platforms: 'linux/amd64'
+            tag-latest: 'auto'
+            tag-suffix: ''
+            ffmpeg: ''
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+  core-image-build:
+    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 }}
+    secrets:
+      dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
+      dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
+      quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
+      quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
+    strategy:
+      matrix:
+        include:
+          - build-type: ''
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-ffmpeg-core'
+            ffmpeg: 'true'
+            image-type: 'core'
+            runs-on: 'ubuntu-latest'
+          - build-type: 'cublas'
+            cuda-major-version: "11"
+            cuda-minor-version: "7"
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda11-core'
+            ffmpeg: ''
+            image-type: 'core'
+            runs-on: 'ubuntu-latest'
+          - build-type: 'cublas'
+            cuda-major-version: "12"
+            cuda-minor-version: "1"
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda12-core'
+            ffmpeg: ''
+            image-type: 'core'
+            runs-on: 'ubuntu-latest'
+          - build-type: 'cublas'
+            cuda-major-version: "11"
+            cuda-minor-version: "7"
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda11-ffmpeg-core'
+            ffmpeg: 'true'
+            image-type: 'core'
+            runs-on: 'ubuntu-latest'
+          - build-type: 'cublas'
+            cuda-major-version: "12"
+            cuda-minor-version: "1"
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda12-ffmpeg-core'
+            ffmpeg: 'true'
+            image-type: 'core'
+            runs-on: 'ubuntu-latest'

.github/workflows/image_build.yml

@@ -0,0 +1,160 @@
+---
+name: 'build container images (reusable)'
+
+on:
+  workflow_call:
+    inputs:
+      build-type:
+        description: 'Build type'
+        default: ''
+        type: string
+      cuda-major-version:
+        description: 'CUDA major version'
+        default: "11"
+        type: string
+      cuda-minor-version:
+        description: 'CUDA minor version'
+        default: "7"
+        type: string
+      platforms:
+        description: 'Platforms'
+        default: ''
+        type: string
+      tag-latest:
+        description: 'Tag latest'
+        default: ''
+        type: string
+      tag-suffix:
+        description: 'Tag suffix'
+        default: ''
+        type: string
+      ffmpeg:
+        description: 'FFMPEG'
+        default: ''
+        type: string
+      image-type:
+        description: 'Image type'
+        default: ''
+        type: string
+      runs-on:
+        description: 'Runs on'
+        required: true
+        default: ''
+        type: string
+    secrets:
+      dockerUsername:
+        required: true
+      dockerPassword:
+        required: true
+      quayUsername:
+        required: true
+      quayPassword:
+        required: true
+jobs:
+  reusable_image-build:
+    runs-on: ${{ inputs.runs-on }}
+    steps:
+      - name: Force Install GIT latest
+        run: |
+          sudo apt-get update \
+          && sudo apt-get install -y software-properties-common \
+          && sudo apt-get update \
+          && sudo add-apt-repository -y ppa:git-core/ppa \
+          && sudo apt-get update \
+          && sudo apt-get install -y git
+      - name: Checkout
+        uses: actions/checkout@v4
+      # - name: Release space from worker
+      #   run: |
+      #     echo "Listing top largest packages"
+      #     pkgs=$(dpkg-query -Wf '${Installed-Size}\t${Package}\t${Status}\n' | awk '$NF == "installed"{print $1 "\t" $2}' | sort -nr)
+      #     head -n 30 <<< "${pkgs}"
+      #     echo
+      #     df -h
+      #     echo
+      #     sudo apt-get remove -y '^llvm-.*|^libllvm.*' || true
+      #     sudo apt-get remove --auto-remove android-sdk-platform-tools || true
+      #     sudo apt-get purge --auto-remove android-sdk-platform-tools || true
+      #     sudo rm -rf /usr/local/lib/android
+      #     sudo apt-get remove -y '^dotnet-.*|^aspnetcore-.*' || true
+      #     sudo rm -rf /usr/share/dotnet
+      #     sudo apt-get remove -y '^mono-.*' || true
+      #     sudo apt-get remove -y '^ghc-.*' || true
+      #     sudo apt-get remove -y '.*jdk.*|.*jre.*' || true
+      #     sudo apt-get remove -y 'php.*' || true
+      #     sudo apt-get remove -y hhvm powershell firefox monodoc-manual msbuild || true
+      #     sudo apt-get remove -y '^google-.*' || true
+      #     sudo apt-get remove -y azure-cli || true
+      #     sudo apt-get remove -y '^mongo.*-.*|^postgresql-.*|^mysql-.*|^mssql-.*' || true
+      #     sudo apt-get remove -y '^gfortran-.*' || true
+      #     sudo apt-get remove -y microsoft-edge-stable || true
+      #     sudo apt-get remove -y firefox || true
+      #     sudo apt-get remove -y powershell || true
+      #     sudo apt-get remove -y r-base-core || true
+      #     sudo apt-get autoremove -y
+      #     sudo apt-get clean
+      #     echo
+      #     echo "Listing top largest packages"
+      #     pkgs=$(dpkg-query -Wf '${Installed-Size}\t${Package}\t${Status}\n' | awk '$NF == "installed"{print $1 "\t" $2}' | sort -nr)
+      #     head -n 30 <<< "${pkgs}"
+      #     echo
+      #     sudo rm -rfv build || true
+      #     df -h
+      - name: Docker meta
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: |
+            quay.io/go-skynet/local-ai
+            localai/localai
+          tags: |
+            type=ref,event=branch
+            type=semver,pattern={{raw}}
+            type=sha
+          flavor: |
+            latest=${{ inputs.tag-latest }}
+            suffix=${{ inputs.tag-suffix }}
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@master
+        with:
+          platforms: all
+
+      - name: Set up Docker Buildx
+        id: buildx
+        uses: docker/setup-buildx-action@master
+
+      - name: Login to DockerHub
+        if: github.event_name != 'pull_request'
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.dockerUsername }}
+          password: ${{ secrets.dockerPassword }}
+
+      - name: Login to DockerHub
+        if: github.event_name != 'pull_request'
+        uses: docker/login-action@v3
+        with:
+          registry: quay.io
+          username: ${{ secrets.quayUsername }}
+          password: ${{ secrets.quayPassword }}
+
+      - name: Build and push
+        uses: docker/build-push-action@v5
+        with:
+          builder: ${{ steps.buildx.outputs.name }}
+          build-args: |
+            BUILD_TYPE=${{ inputs.build-type }}
+            CUDA_MAJOR_VERSION=${{ inputs.cuda-major-version }}
+            CUDA_MINOR_VERSION=${{ inputs.cuda-minor-version }}
+            FFMPEG=${{ inputs.ffmpeg }}
+            IMAGE_TYPE=${{ inputs.image-type }}
+          context: .
+          file: ./Dockerfile
+          platforms: ${{ inputs.platforms }}
+          push: ${{ github.event_name != 'pull_request' }}
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}
+      - name: job summary
+        run: |
+          echo "Built image: ${{ steps.meta.outputs.labels }}" >> $GITHUB_STEP_SUMMARY

.github/workflows/release.yaml

@@ -0,0 +1,112 @@
+name: Build and Release
+
+on: push
+
+permissions:
+  contents: write
+
+concurrency:
+  group: ci-releases-${{ github.head_ref || github.ref }}-${{ github.repository }}
+  cancel-in-progress: true
+
+jobs:
+  build-linux:
+    strategy:
+      matrix:
+        include:
+          - build: 'avx2'
+            defines: ''
+          - build: 'avx'
+            defines: '-DLLAMA_AVX2=OFF'
+          - build: 'avx512'
+            defines: '-DLLAMA_AVX512=ON'
+    runs-on: ubuntu-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v4
+        with:
+          submodules: true
+      - uses: actions/setup-go@v4
+        with:
+          go-version: '>=1.21.0'
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install build-essential ffmpeg
+
+      - name: Cache grpc
+        id: cache-grpc
+        uses: actions/cache@v3
+        with:
+          path: grpc
+          key: ${{ runner.os }}-grpc
+      - name: Build grpc
+        if: steps.cache-grpc.outputs.cache-hit != 'true'
+        run: |
+          git clone --recurse-submodules -b v1.58.0 --depth 1 --shallow-submodules https://github.com/grpc/grpc && \
+          cd grpc && mkdir -p cmake/build && cd cmake/build && cmake -DgRPC_INSTALL=ON \
+            -DgRPC_BUILD_TESTS=OFF \
+            ../.. && sudo make -j12
+      - name: Install gRPC
+        run: |
+          cd grpc && cd cmake/build && sudo make -j12 install
+
+      - name: Build
+        id: build
+        env:
+          CMAKE_ARGS: "${{ matrix.defines }}"
+          BUILD_ID: "${{ matrix.build }}"
+        run: |
+          STATIC=true make dist
+      - uses: actions/upload-artifact@v3
+        with:
+          name: ${{ matrix.build }}
+          path: release/
+      - name: Release
+        uses: softprops/action-gh-release@v1
+        if: startsWith(github.ref, 'refs/tags/')
+        with:
+          files: |
+            release/*
+
+  build-macOS:
+    strategy:
+      matrix:
+        include:
+          - build: 'avx2'
+            defines: ''
+          - build: 'avx'
+            defines: '-DLLAMA_AVX2=OFF'
+          - build: 'avx512'
+            defines: '-DLLAMA_AVX512=ON'
+    runs-on: macOS-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v4
+        with:
+          submodules: true
+      - uses: actions/setup-go@v4
+        with:
+          go-version: '>=1.21.0'
+      - name: Dependencies
+        run: |
+          brew install protobuf grpc
+      - name: Build
+        id: build
+        env:
+          CMAKE_ARGS: "${{ matrix.defines }}"
+          BUILD_ID: "${{ matrix.build }}"
+        run: |
+          export C_INCLUDE_PATH=/usr/local/include
+          export CPLUS_INCLUDE_PATH=/usr/local/include
+          make dist
+      - uses: actions/upload-artifact@v3
+        with:
+          name: ${{ matrix.build }}
+          path: release/
+      - name: Release
+        uses: softprops/action-gh-release@v1
+        if: startsWith(github.ref, 'refs/tags/')
+        with:
+          files: |
+            release/*

.github/workflows/release.yml.disabled

@@ -1,26 +0,0 @@
-name: goreleaser
-
-on:
-  push:
-    tags:
-      - 'v*'
-
-jobs:
-  goreleaser:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v3
-        with:
-          fetch-depth: 0
-      - name: Set up Go
-        uses: actions/setup-go@v3
-        with:
-          go-version: 1.18
-      - name: Run GoReleaser
-        uses: goreleaser/goreleaser-action@v4
-        with:
-          version: latest
-          args: release --clean
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/test-extra.yml

@@ -0,0 +1,277 @@
+---
+name: 'Tests extras backends'
+
+on:
+  pull_request:
+  push:
+    branches:
+      - master
+    tags:
+      - '*'
+
+concurrency:
+  group: ci-tests-extra-${{ github.head_ref || github.ref }}-${{ github.repository }}
+  cancel-in-progress: true
+
+jobs:
+  tests-transformers:
+    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
+          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 && \
+              gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list' && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list' && \
+             sudo apt-get update && \
+             sudo apt-get install -y conda
+          sudo apt-get install -y ca-certificates cmake curl patch
+          sudo apt-get install -y libopencv-dev && sudo ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2
+          
+          sudo rm -rfv /usr/bin/conda || true
+
+      - name: Test transformers
+        run: |
+           export PATH=$PATH:/opt/conda/bin
+           make -C backend/python/transformers
+           make -C backend/python/transformers test
+
+  tests-sentencetransformers:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v4
+        with: 
+          submodules: true
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install build-essential ffmpeg
+          curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmor > conda.gpg && \
+             sudo install -o root -g root -m 644 conda.gpg /usr/share/keyrings/conda-archive-keyring.gpg && \
+              gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list' && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list' && \
+             sudo apt-get update && \
+             sudo apt-get install -y conda
+          sudo apt-get install -y ca-certificates cmake curl patch
+          sudo apt-get install -y libopencv-dev && sudo ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2
+          
+          sudo rm -rfv /usr/bin/conda || true
+
+      - name: Test sentencetransformers
+        run: |
+           export PATH=$PATH:/opt/conda/bin
+           make -C backend/python/sentencetransformers
+           make -C backend/python/sentencetransformers test
+
+  tests-diffusers:
+    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
+          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 && \
+              gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list' && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list' && \
+             sudo apt-get update && \
+             sudo apt-get install -y conda
+          sudo apt-get install -y ca-certificates cmake curl patch
+          sudo apt-get install -y libopencv-dev && sudo ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2
+          
+          sudo rm -rfv /usr/bin/conda || true
+
+      - name: Test diffusers
+        run: |
+           export PATH=$PATH:/opt/conda/bin
+           make -C backend/python/diffusers
+           make -C backend/python/diffusers test
+
+
+  tests-transformers-musicgen:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v4
+        with: 
+          submodules: true
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install build-essential ffmpeg
+          curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmor > conda.gpg && \
+             sudo install -o root -g root -m 644 conda.gpg /usr/share/keyrings/conda-archive-keyring.gpg && \
+              gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list' && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list' && \
+             sudo apt-get update && \
+             sudo apt-get install -y conda
+          sudo apt-get install -y ca-certificates cmake curl patch
+          sudo apt-get install -y libopencv-dev && sudo ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2
+          
+          sudo rm -rfv /usr/bin/conda || true
+
+      - name: Test transformers-musicgen
+        run: |
+           export PATH=$PATH:/opt/conda/bin
+           make -C backend/python/transformers-musicgen
+           make -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
+          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 && \
+              gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list' && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list' && \
+             sudo apt-get update && \
+             sudo apt-get install -y conda
+          sudo apt-get install -y ca-certificates cmake curl patch
+          sudo apt-get install -y libopencv-dev && sudo ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2
+          
+          sudo rm -rfv /usr/bin/conda || true
+
+      - name: Test petals
+        run: |
+           export PATH=$PATH:/opt/conda/bin
+           make -C backend/python/petals
+           make -C backend/python/petals test
+
+           
+
+  tests-bark:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v4
+        with: 
+          submodules: true
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install build-essential ffmpeg
+          curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmor > conda.gpg && \
+             sudo install -o root -g root -m 644 conda.gpg /usr/share/keyrings/conda-archive-keyring.gpg && \
+              gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list' && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list' && \
+             sudo apt-get update && \
+             sudo apt-get install -y conda
+          sudo apt-get install -y ca-certificates cmake curl patch
+          sudo apt-get install -y libopencv-dev && sudo ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2
+          
+          sudo rm -rfv /usr/bin/conda || true
+
+      - name: Test bark
+        run: |
+           export PATH=$PATH:/opt/conda/bin
+           make -C backend/python/bark
+           make -C backend/python/bark test
+
+           
+  # Below tests needs GPU. Commented out for now
+  # TODO: Re-enable as soon as we have GPU nodes
+  # tests-vllm:
+  #   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
+  #         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 && \
+  #             gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && \
+  #            sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list' && \
+  #            sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list' && \
+  #            sudo apt-get update && \
+  #            sudo apt-get install -y conda
+  #         sudo apt-get install -y ca-certificates cmake curl patch
+  #         sudo apt-get install -y libopencv-dev && sudo ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2
+  #         sudo rm -rfv /usr/bin/conda || true
+  #     - name: Test vllm
+  #       run: |
+  #          export PATH=$PATH:/opt/conda/bin
+  #          make -C backend/python/vllm
+  #          make -C backend/python/vllm test
+  tests-vallex:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v4
+        with: 
+          submodules: true
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install build-essential ffmpeg
+          curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmor > conda.gpg && \
+             sudo install -o root -g root -m 644 conda.gpg /usr/share/keyrings/conda-archive-keyring.gpg && \
+              gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list' && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list' && \
+             sudo apt-get update && \
+             sudo apt-get install -y conda
+          sudo apt-get install -y ca-certificates cmake curl patch
+          sudo apt-get install -y libopencv-dev && sudo ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2    
+          sudo rm -rfv /usr/bin/conda || true
+      - name: Test vall-e-x
+        run: |
+           export PATH=$PATH:/opt/conda/bin
+           make -C backend/python/vall-e-x
+           make -C backend/python/vall-e-x test
+
+  tests-coqui:
+    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
+          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 && \
+              gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list' && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list' && \
+             sudo apt-get update && \
+             sudo apt-get install -y conda
+          sudo apt-get install -y ca-certificates cmake curl patch espeak espeak-ng          
+          sudo rm -rfv /usr/bin/conda || true
+
+      - name: Test coqui
+        run: |
+           export PATH=$PATH:/opt/conda/bin
+           make -C backend/python/coqui
+           make -C backend/python/coqui test

.github/workflows/test.yml

@@ -0,0 +1,133 @@
+---
+name: 'tests'
+
+on:
+  pull_request:
+  push:
+    branches:
+      - master
+    tags:
+      - '*'
+
+concurrency:
+  group: ci-tests-${{ github.head_ref || github.ref }}-${{ github.repository }}
+  cancel-in-progress: true
+
+jobs:
+  tests-linux:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        go-version: ['1.21.x']
+    steps:
+      - name: Release space from worker
+        run: |
+          echo "Listing top largest packages"
+          pkgs=$(dpkg-query -Wf '${Installed-Size}\t${Package}\t${Status}\n' | awk '$NF == "installed"{print $1 "\t" $2}' | sort -nr)
+          head -n 30 <<< "${pkgs}"
+          echo
+          df -h
+          echo
+          sudo apt-get remove -y '^llvm-.*|^libllvm.*' || true
+          sudo apt-get remove --auto-remove android-sdk-platform-tools || true
+          sudo apt-get purge --auto-remove android-sdk-platform-tools || true
+          sudo rm -rf /usr/local/lib/android
+          sudo apt-get remove -y '^dotnet-.*|^aspnetcore-.*' || true
+          sudo rm -rf /usr/share/dotnet
+          sudo apt-get remove -y '^mono-.*' || true
+          sudo apt-get remove -y '^ghc-.*' || true
+          sudo apt-get remove -y '.*jdk.*|.*jre.*' || true
+          sudo apt-get remove -y 'php.*' || true
+          sudo apt-get remove -y hhvm powershell firefox monodoc-manual msbuild || true
+          sudo apt-get remove -y '^google-.*' || true
+          sudo apt-get remove -y azure-cli || true
+          sudo apt-get remove -y '^mongo.*-.*|^postgresql-.*|^mysql-.*|^mssql-.*' || true
+          sudo apt-get remove -y '^gfortran-.*' || true
+          sudo apt-get autoremove -y
+          sudo apt-get clean
+          echo
+          echo "Listing top largest packages"
+          pkgs=$(dpkg-query -Wf '${Installed-Size}\t${Package}\t${Status}\n' | awk '$NF == "installed"{print $1 "\t" $2}' | sort -nr)
+          head -n 30 <<< "${pkgs}"
+          echo
+          sudo rm -rfv build || true
+          df -h
+      - name: Clone
+        uses: actions/checkout@v4
+        with: 
+          submodules: true
+      - name: Setup Go ${{ matrix.go-version }}
+        uses: actions/setup-go@v4
+        with:
+          go-version: ${{ matrix.go-version }}
+      # You can test your matrix by printing the current Go version
+      - name: Display Go version
+        run: go version
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install build-essential ffmpeg
+          curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmor > conda.gpg && \
+             sudo install -o root -g root -m 644 conda.gpg /usr/share/keyrings/conda-archive-keyring.gpg && \
+              gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list' && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list' && \
+             sudo apt-get update && \
+             sudo apt-get install -y conda
+          sudo apt-get install -y ca-certificates cmake curl patch
+          sudo apt-get install -y libopencv-dev && sudo ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2
+          
+          sudo rm -rfv /usr/bin/conda || true
+          PATH=$PATH:/opt/conda/bin make -C backend/python/sentencetransformers
+
+          # Pre-build piper before we start tests in order to have shared libraries in place
+          make sources/go-piper && \
+          GO_TAGS="tts" make -C sources/go-piper piper.o && \
+          sudo cp -rfv sources/go-piper/piper-phonemize/pi/lib/. /usr/lib/ && \
+          # Pre-build stable diffusion before we install a newer version of abseil (not compatible with stablediffusion-ncn)
+          GO_TAGS="stablediffusion tts" GRPC_BACKENDS=backend-assets/grpc/stablediffusion make build
+      - name: Cache grpc
+        id: cache-grpc
+        uses: actions/cache@v3
+        with:
+          path: grpc
+          key: ${{ runner.os }}-grpc
+      - name: Build grpc
+        if: steps.cache-grpc.outputs.cache-hit != 'true'
+        run: |
+          git clone --recurse-submodules -b v1.58.0 --depth 1 --shallow-submodules https://github.com/grpc/grpc && \
+          cd grpc && mkdir -p cmake/build && cd cmake/build && cmake -DgRPC_INSTALL=ON \
+            -DgRPC_BUILD_TESTS=OFF \
+            ../.. && sudo make -j12
+      - name: Install gRPC
+        run: |
+          cd grpc && cd cmake/build && sudo make -j12 install
+      - name: Test
+        run: |
+          GO_TAGS="stablediffusion tts" make test
+
+  tests-apple:
+    runs-on: macOS-latest
+    strategy:
+      matrix:
+        go-version: ['1.21.x']
+    steps:
+      - name: Clone
+        uses: actions/checkout@v4
+        with: 
+          submodules: true
+      - name: Setup Go ${{ matrix.go-version }}
+        uses: actions/setup-go@v4
+        with:
+          go-version: ${{ matrix.go-version }}
+      # You can test your matrix by printing the current Go version
+      - name: Display Go version
+        run: go version
+      - name: Dependencies
+        run: |
+          brew install protobuf grpc
+      - name: Test
+        run: |
+          export C_INCLUDE_PATH=/usr/local/include
+          export CPLUS_INCLUDE_PATH=/usr/local/include
+          CMAKE_ARGS="-DLLAMA_F16C=OFF -DLLAMA_AVX512=OFF -DLLAMA_AVX2=OFF -DLLAMA_FMA=OFF" make test

.gitignore

@@ -1,11 +1,39 @@
 # go-llama build artifacts
-go-llama
-go-gpt4all-j
+/sources/
+__pycache__/
+*.a
+get-sources
+prepare-sources
+/backend/cpp/llama/grpc-server
+/backend/cpp/llama/llama.cpp
+
+go-ggml-transformers
+go-gpt2
+go-rwkv
+whisper.cpp
+/bloomz
+go-bert
 
 # LocalAI build binary
 LocalAI
 local-ai
+# prevent above rules from omitting the helm chart
+!charts/*
+# prevent above rules from omitting the api/localai folder
+!api/localai
 
 # Ignore models
-models/*.bin
-models/ggml-*
+models/*
+test-models/
+test-dir/
+
+release/
+
+# just in case
+.DS_Store
+.idea
+
+# Generated during build
+backend-assets/
+prepare
+/ggml-metal.metal

.gitmodules

@@ -0,0 +1,6 @@
+[submodule "docs/themes/hugo-theme-relearn"]
+	path = docs/themes/hugo-theme-relearn
+	url = https://github.com/McShelby/hugo-theme-relearn.git
+[submodule "docs/themes/lotusdocs"]
+	path = docs/themes/lotusdocs
+	url = https://github.com/colinwilson/lotusdocs

.goreleaser.yaml

@@ -1,15 +0,0 @@
-# Make sure to check the documentation at http://goreleaser.com
-project_name: local-ai
-builds:
-  - ldflags:
-      - -w -s
-    env:
-      - CGO_ENABLED=0
-    goos:
-      - linux
-      - darwin
-      - windows
-    goarch:
-      - amd64
-      - arm64
-    binary: '{{ .ProjectName }}'

.vscode/launch.json

@@ -1,16 +1,33 @@
 {
     "version": "0.2.0",
     "configurations": [
-    
-    {
-        "name": "Launch Go",
-        "type": "go",
-        "request": "launch",
-        "mode": "debug",
-        "program": "${workspaceFolder}/main.go",
-        "args": [
-            "api"
-        ]
-    }
+        {
+            "name": "Python: Current File",
+            "type": "python",
+            "request": "launch",
+            "program": "${file}",
+            "console": "integratedTerminal",
+            "justMyCode": false,
+            "cwd": "${workspaceFolder}/examples/langchain-chroma",
+            "env": {
+                "OPENAI_API_BASE": "http://localhost:8080/v1",
+                "OPENAI_API_KEY": "abc"
+            }
+        },
+        {
+            "name": "Launch LocalAI API",
+            "type": "go",
+            "request": "launch",
+            "mode": "debug",
+            "program": "${workspaceFolder}/main.go",
+            "args": [
+                "api"
+            ],
+            "env": {
+                "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

@@ -0,0 +1,72 @@
+# Contributing to localAI
+
+Thank you for your interest in contributing to LocalAI! We appreciate your time and effort in helping to improve our project. Before you get started, please take a moment to review these guidelines.
+
+## Table of Contents
+
+- [Getting Started](#getting-started)
+  - [Prerequisites](#prerequisites)
+  - [Setting up the Development Environment](#setting-up-the-development-environment)
+- [Contributing](#contributing)
+  - [Submitting an Issue](#submitting-an-issue)
+  - [Creating a Pull Request (PR)](#creating-a-pull-request-pr)
+- [Coding Guidelines](#coding-guidelines)
+- [Testing](#testing)
+- [Documentation](#documentation)
+- [Community and Communication](#community-and-communication)
+
+
+
+## Getting Started
+
+### Prerequisites
+
+- Golang [1.21]
+- Git
+- macOS/Linux
+
+### Setting up the Development Environment and running localAI in the local environment
+
+1. Clone the repository: `git clone https://github.com/go-skynet/LocalAI.git`
+2. Navigate to the project directory: `cd LocalAI`
+3. Install the required dependencies: `make prepare`
+4. Run LocalAI: `make run`
+
+## Contributing
+
+We welcome contributions from everyone! To get started, follow these steps:
+
+### Submitting an Issue
+
+If you find a bug, have a feature request, or encounter any issues, please check the [issue tracker](https://github.com/go-skynet/LocalAI/issues) to see if a similar issue has already been reported. If not, feel free to [create a new issue](https://github.com/go-skynet/LocalAI/issues/new) and provide as much detail as possible.
+
+### Creating a Pull Request (PR)
+
+1. Fork the repository.
+2. Create a new branch with a descriptive name: `git checkout -b [branch name]`
+3. Make your changes and commit them.
+4. Push the changes to your fork: `git push origin [branch name]`
+5. Create a new pull request from your branch to the main project's `main` or `master` branch.
+6. Provide a clear description of your changes in the pull request.
+7. Make any requested changes during the review process.
+8. Once your PR is approved, it will be merged into the main project.
+
+## 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.
+
+## Testing
+
+`make test` cannot handle all the model now. Please be sure to add a test case for the new features or the part was changed.
+
+## Documentation
+
+- We are welcome the contribution of the documents, please open new PR in the official document repo [localai-website](https://github.com/go-skynet/localai-website)
+
+## Community and Communication
+
+- 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

@@ -1,12 +1,207 @@
-ARG GO_VERSION=1.20
-ARG DEBIAN_VERSION=11
-FROM golang:$GO_VERSION as builder
-WORKDIR /build
-RUN apt-get update && apt-get install -y cmake
-COPY . .
-ARG BUILD_TYPE=
-RUN make build${BUILD_TYPE}
+ARG GO_VERSION=1.21-bullseye
+ARG IMAGE_TYPE=extras
+# extras or core
 
-FROM debian:$DEBIAN_VERSION
-COPY --from=builder /build/local-ai /usr/bin/local-ai
-ENTRYPOINT [ "/usr/bin/local-ai" ]
+
+FROM golang:$GO_VERSION as requirements-core
+
+ARG BUILD_TYPE
+ARG CUDA_MAJOR_VERSION=11
+ARG CUDA_MINOR_VERSION=7
+ARG TARGETARCH
+ARG TARGETVARIANT
+
+ENV BUILD_TYPE=${BUILD_TYPE}
+
+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,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,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"
+
+ARG GO_TAGS="stablediffusion tinydream tts"
+
+RUN apt-get update && \
+    apt-get install -y ca-certificates curl patch pip cmake && apt-get clean
+
+
+COPY --chmod=644 custom-ca-certs/* /usr/local/share/ca-certificates/
+RUN update-ca-certificates
+
+# Use the variables in subsequent instructions
+RUN echo "Target Architecture: $TARGETARCH"
+RUN echo "Target Variant: $TARGETVARIANT"
+
+# CuBLAS requirements
+RUN if [ "${BUILD_TYPE}" = "cublas" ]; then \
+    apt-get install -y software-properties-common && \
+    apt-add-repository contrib && \
+    curl -O https://developer.download.nvidia.com/compute/cuda/repos/debian11/x86_64/cuda-keyring_1.0-1_all.deb && \
+    dpkg -i cuda-keyring_1.0-1_all.deb && \
+    rm -f cuda-keyring_1.0-1_all.deb && \
+    apt-get update && \
+    apt-get install -y cuda-nvcc-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} libcublas-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} libcusparse-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} libcusolver-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION}  && apt-get clean \
+    ; fi
+ENV PATH /usr/local/cuda/bin:${PATH}
+
+# OpenBLAS requirements and stable diffusion
+RUN apt-get install -y \
+    libopenblas-dev \
+    libopencv-dev \ 
+    && apt-get clean
+
+# Set up OpenCV
+RUN ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2
+
+WORKDIR /build
+
+RUN test -n "$TARGETARCH" \
+    || (echo 'warn: missing $TARGETARCH, either set this `ARG` manually, or run using `docker buildkit`')
+
+# Extras requirements
+FROM requirements-core as requirements-extras
+
+RUN curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmor > conda.gpg && \
+    install -o root -g root -m 644 conda.gpg /usr/share/keyrings/conda-archive-keyring.gpg && \
+    gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && \
+    echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list && \
+    echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list && \
+    apt-get update && \
+    apt-get install -y conda && apt-get clean
+
+ENV PATH="/root/.cargo/bin:${PATH}"
+RUN pip install --upgrade pip
+RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
+RUN apt-get install -y espeak-ng espeak && apt-get clean
+
+###################################
+###################################
+
+FROM requirements-${IMAGE_TYPE} as builder
+
+ARG GO_TAGS="stablediffusion tts"
+ARG GRPC_BACKENDS
+ARG BUILD_GRPC=true
+ENV GRPC_BACKENDS=${GRPC_BACKENDS}
+ENV GO_TAGS=${GO_TAGS}
+ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
+ENV NVIDIA_REQUIRE_CUDA="cuda>=${CUDA_MAJOR_VERSION}.0"
+ENV NVIDIA_VISIBLE_DEVICES=all
+
+WORKDIR /build
+
+COPY . .
+COPY .git .
+RUN make prepare
+
+# stablediffusion does not tolerate a newer version of abseil, build it first
+RUN GRPC_BACKENDS=backend-assets/grpc/stablediffusion make build
+
+RUN if [ "${BUILD_GRPC}" = "true" ]; then \
+    git clone --recurse-submodules -b v1.58.0 --depth 1 --shallow-submodules https://github.com/grpc/grpc && \
+    cd grpc && mkdir -p cmake/build && cd cmake/build && cmake -DgRPC_INSTALL=ON \
+      -DgRPC_BUILD_TESTS=OFF \
+       ../.. && make -j12 install \
+    ; fi
+
+# Rebuild with defaults backends
+RUN make build
+
+RUN if [ ! -d "/build/sources/go-piper/piper-phonemize/pi/lib/" ]; then \
+    mkdir -p /build/sources/go-piper/piper-phonemize/pi/lib/ \
+    touch /build/sources/go-piper/piper-phonemize/pi/lib/keep \
+    ; fi
+
+###################################
+###################################
+
+FROM requirements-${IMAGE_TYPE}
+
+ARG FFMPEG
+ARG BUILD_TYPE
+ARG TARGETARCH
+ARG IMAGE_TYPE=extras
+
+ENV BUILD_TYPE=${BUILD_TYPE}
+ENV REBUILD=false
+ENV HEALTHCHECK_ENDPOINT=http://localhost:8080/readyz
+
+ARG CUDA_MAJOR_VERSION=11
+ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
+ENV NVIDIA_REQUIRE_CUDA="cuda>=${CUDA_MAJOR_VERSION}.0"
+ENV NVIDIA_VISIBLE_DEVICES=all
+ENV PIP_CACHE_PURGE=true
+
+# Add FFmpeg
+RUN if [ "${FFMPEG}" = "true" ]; then \
+    apt-get install -y ffmpeg && apt-get clean \
+    ; fi
+
+WORKDIR /build
+
+# we start fresh & re-copy all assets because `make build` does not clean up nicely after itself
+# so when `entrypoint.sh` runs `make build` again (which it does by default), the build would fail
+# see https://github.com/go-skynet/LocalAI/pull/658#discussion_r1241971626 and
+# https://github.com/go-skynet/LocalAI/pull/434
+COPY . .
+
+COPY --from=builder /build/sources ./sources/
+COPY --from=builder /build/grpc ./grpc/
+
+RUN make prepare-sources && cd /build/grpc/cmake/build && make install && rm -rf grpc
+
+# Copy the binary
+COPY --from=builder /build/local-ai ./
+
+# Copy shared libraries for piper
+COPY --from=builder /build/sources/go-piper/piper-phonemize/pi/lib/* /usr/lib/
+
+# do not let stablediffusion rebuild (requires an older version of absl)
+COPY --from=builder /build/backend-assets/grpc/stablediffusion ./backend-assets/grpc/stablediffusion
+
+## Duplicated from Makefile to avoid having a big layer that's hard to push
+RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
+	PATH=$PATH:/opt/conda/bin make -C backend/python/autogptq \
+    ; fi
+RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
+	PATH=$PATH:/opt/conda/bin make -C backend/python/bark \
+    ; fi
+RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
+	PATH=$PATH:/opt/conda/bin make -C backend/python/diffusers \
+    ; fi
+RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
+	PATH=$PATH:/opt/conda/bin make -C backend/python/vllm \
+    ; fi
+RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
+	PATH=$PATH:/opt/conda/bin make -C backend/python/mamba \
+    ; fi
+RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
+	PATH=$PATH:/opt/conda/bin make -C backend/python/sentencetransformers \
+    ; fi
+RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
+	PATH=$PATH:/opt/conda/bin make -C backend/python/transformers \
+    ; fi
+RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
+	PATH=$PATH:/opt/conda/bin make -C backend/python/vall-e-x \
+    ; fi
+RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
+	PATH=$PATH:/opt/conda/bin make -C backend/python/exllama \
+    ; fi
+RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
+    PATH=$PATH:/opt/conda/bin make -C backend/python/exllama2 \
+    ; fi
+RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
+	PATH=$PATH:/opt/conda/bin make -C backend/python/petals \
+    ; fi
+RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
+	PATH=$PATH:/opt/conda/bin make -C backend/python/transformers-musicgen \
+    ; fi
+RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
+	PATH=$PATH:/opt/conda/bin make -C backend/python/coqui \
+    ; fi
+
+# Make sure the models directory exists
+RUN mkdir -p /build/models
+
+# Define the health check command
+HEALTHCHECK --interval=1m --timeout=10m --retries=10 \
+  CMD curl -f $HEALTHCHECK_ENDPOINT || exit 1
+
+EXPOSE 8080
+ENTRYPOINT [ "/build/entrypoint.sh" ]

Entitlements.plist

@@ -0,0 +1,10 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+    <key>com.apple.security.network.client</key>
+    <true/>
+    <key>com.apple.security.network.server</key>
+    <true/>
+</dict>
+</plist>

LICENSE

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

Makefile

@@ -2,69 +2,392 @@ GOCMD=go
 GOTEST=$(GOCMD) test
 GOVET=$(GOCMD) vet
 BINARY_NAME=local-ai
-GOLLAMA_VERSION?=llama.cpp-5ecff35
 
+# llama.cpp versions
+GOLLAMA_VERSION?=aeba71ee842819da681ea537e78846dc75949ac0
+
+GOLLAMA_STABLE_VERSION?=50cee7712066d9e38306eccadcfbb44ea87df4b7
+
+CPPLLAMA_VERSION?=6db2b41a76ee78d5efdd5c3cddd5d7ad3f646855
+
+# gpt4all version
+GPT4ALL_REPO?=https://github.com/nomic-ai/gpt4all
+GPT4ALL_VERSION?=27a8b020c36b0df8f8b82a252d261cda47cf44b8
+
+# go-ggml-transformers version
+GOGGMLTRANSFORMERS_VERSION?=ffb09d7dd71e2cbc6c5d7d05357d230eea6f369a
+
+# go-rwkv version
+RWKV_REPO?=https://github.com/donomii/go-rwkv.cpp
+RWKV_VERSION?=633c5a3485c403cb2520693dc0991a25dace9f0f
+
+# whisper.cpp version
+WHISPER_CPP_VERSION?=37a709f6558c6d9783199e2b8cbb136e1c41d346
+
+# bert.cpp version
+BERT_VERSION?=6abe312cded14042f6b7c3cd8edf082713334a4d
+
+# go-piper version
+PIPER_VERSION?=d6b6275ba037dabdba4a8b65dfdf6b2a73a67f07
+
+# stablediffusion version
+STABLEDIFFUSION_VERSION?=902db5f066fd137697e3b69d0fa10d4782bd2c2f
+
+# tinydream version
+TINYDREAM_VERSION?=772a9c0d9aaf768290e63cca3c904fe69faf677a
+
+export BUILD_TYPE?=
+export STABLE_BUILD_TYPE?=$(BUILD_TYPE)
+export CMAKE_ARGS?=
+
+CGO_LDFLAGS?=
+CUDA_LIBPATH?=/usr/local/cuda/lib64/
+GO_TAGS?=
+BUILD_ID?=git
+
+TEST_DIR=/tmp/test
+
+RANDOM := $(shell bash -c 'echo $$RANDOM')
+
+VERSION?=$(shell git describe --always --tags || echo "dev" )
+# go tool nm ./local-ai | grep Commit
+LD_FLAGS?=
+override LD_FLAGS += -X "github.com/go-skynet/LocalAI/internal.Version=$(VERSION)"
+override LD_FLAGS += -X "github.com/go-skynet/LocalAI/internal.Commit=$(shell git rev-parse HEAD)"
+
+OPTIONAL_TARGETS?=
+
+OS := $(shell uname -s)
+ARCH := $(shell uname -m)
 GREEN  := $(shell tput -Txterm setaf 2)
 YELLOW := $(shell tput -Txterm setaf 3)
 WHITE  := $(shell tput -Txterm setaf 7)
 CYAN   := $(shell tput -Txterm setaf 6)
 RESET  := $(shell tput -Txterm sgr0)
 
+# Default Docker bridge IP
+E2E_BRIDGE_IP?=172.17.0.1
+
+ifndef UNAME_S
+UNAME_S := $(shell uname -s)
+endif
+
+ifeq ($(OS),Darwin)
+	CGO_LDFLAGS += -lcblas -framework Accelerate
+	ifeq ($(OSX_SIGNING_IDENTITY),)
+		OSX_SIGNING_IDENTITY := $(shell security find-identity -v -p codesigning | grep '"' | head -n 1 | sed -E 's/.*"(.*)"/\1/')
+	endif
+
+	# on OSX, if BUILD_TYPE is blank, we should default to use Metal
+	ifeq ($(BUILD_TYPE),)
+		BUILD_TYPE=metal
+	# disable metal if on Darwin and any other value is explicitly passed.
+	else ifneq ($(BUILD_TYPE),metal)
+		CMAKE_ARGS+=-DLLAMA_METAL=OFF
+	endif
+endif
+
+ifeq ($(BUILD_TYPE),openblas)
+	CGO_LDFLAGS+=-lopenblas
+	export WHISPER_OPENBLAS=1
+endif
+
+ifeq ($(BUILD_TYPE),cublas)
+	CGO_LDFLAGS+=-lcublas -lcudart -L$(CUDA_LIBPATH)
+	export LLAMA_CUBLAS=1
+	export WHISPER_CUBLAS=1
+endif
+
+ifeq ($(BUILD_TYPE),hipblas)
+	ROCM_HOME ?= /opt/rocm
+	export CXX=$(ROCM_HOME)/llvm/bin/clang++
+	export CC=$(ROCM_HOME)/llvm/bin/clang
+	# llama-ggml has no hipblas support, so override it here.
+	export STABLE_BUILD_TYPE=
+	export WHISPER_HIPBLAS=1
+	GPU_TARGETS ?= gfx900,gfx90a,gfx1030,gfx1031,gfx1100
+	AMDGPU_TARGETS ?= "$(GPU_TARGETS)"
+	CMAKE_ARGS+=-DLLAMA_HIPBLAS=ON -DAMDGPU_TARGETS="$(AMDGPU_TARGETS)" -DGPU_TARGETS="$(GPU_TARGETS)"
+	CGO_LDFLAGS += -O3 --rtlib=compiler-rt -unwindlib=libgcc -lhipblas -lrocblas --hip-link
+endif
+
+ifeq ($(BUILD_TYPE),metal)
+	CGO_LDFLAGS+=-framework Foundation -framework Metal -framework MetalKit -framework MetalPerformanceShaders
+	export LLAMA_METAL=1
+	export WHISPER_METAL=1
+endif
+
+ifeq ($(BUILD_TYPE),clblas)
+	CGO_LDFLAGS+=-lOpenCL -lclblast
+	export WHISPER_CLBLAST=1
+endif
+
+# glibc-static or glibc-devel-static required
+ifeq ($(STATIC),true)
+	LD_FLAGS=-linkmode external -extldflags -static
+endif
+
+ifeq ($(findstring stablediffusion,$(GO_TAGS)),stablediffusion)
+#	OPTIONAL_TARGETS+=go-stable-diffusion/libstablediffusion.a
+	OPTIONAL_GRPC+=backend-assets/grpc/stablediffusion
+endif
+
+ifeq ($(findstring tinydream,$(GO_TAGS)),tinydream)
+#	OPTIONAL_TARGETS+=go-tiny-dream/libtinydream.a
+	OPTIONAL_GRPC+=backend-assets/grpc/tinydream
+endif
+
+ifeq ($(findstring tts,$(GO_TAGS)),tts)
+#	OPTIONAL_TARGETS+=go-piper/libpiper_binding.a
+#	OPTIONAL_TARGETS+=backend-assets/espeak-ng-data
+	PIPER_CGO_CXXFLAGS+=-I$(CURDIR)/sources/go-piper/piper/src/cpp -I$(CURDIR)/sources/go-piper/piper/build/fi/include -I$(CURDIR)/sources/go-piper/piper/build/pi/include -I$(CURDIR)/sources/go-piper/piper/build/si/include
+	PIPER_CGO_LDFLAGS+=-L$(CURDIR)/sources/go-piper/piper/build/fi/lib -L$(CURDIR)/sources/go-piper/piper/build/pi/lib -L$(CURDIR)/sources/go-piper/piper/build/si/lib -lfmt -lspdlog -lucd
+	OPTIONAL_GRPC+=backend-assets/grpc/piper
+endif
+
+ALL_GRPC_BACKENDS=backend-assets/grpc/langchain-huggingface backend-assets/grpc/falcon-ggml backend-assets/grpc/bert-embeddings backend-assets/grpc/llama backend-assets/grpc/llama-cpp backend-assets/grpc/llama-ggml backend-assets/grpc/gpt4all backend-assets/grpc/dolly backend-assets/grpc/gpt2 backend-assets/grpc/gptj backend-assets/grpc/gptneox backend-assets/grpc/mpt backend-assets/grpc/replit backend-assets/grpc/starcoder backend-assets/grpc/rwkv backend-assets/grpc/whisper $(OPTIONAL_GRPC)
+GRPC_BACKENDS?=$(ALL_GRPC_BACKENDS) $(OPTIONAL_GRPC)
+
+# If empty, then we build all
+ifeq ($(GRPC_BACKENDS),)
+	GRPC_BACKENDS=$(ALL_GRPC_BACKENDS)
+endif
+
+ifeq ($(BUILD_API_ONLY),true)
+	GRPC_BACKENDS=
+endif
+
 .PHONY: all test build vendor
 
 all: help
 
+## GPT4ALL
+sources/gpt4all:
+	git clone --recurse-submodules $(GPT4ALL_REPO) sources/gpt4all
+	cd sources/gpt4all && git checkout -b build $(GPT4ALL_VERSION) && git submodule update --init --recursive --depth 1
+
+## go-piper
+sources/go-piper:
+	git clone --recurse-submodules https://github.com/mudler/go-piper sources/go-piper
+	cd sources/go-piper && git checkout -b build $(PIPER_VERSION) && git submodule update --init --recursive --depth 1
+
+## BERT embeddings
+sources/go-bert:
+	git clone --recurse-submodules https://github.com/go-skynet/go-bert.cpp sources/go-bert
+	cd sources/go-bert && git checkout -b build $(BERT_VERSION) && git submodule update --init --recursive --depth 1
+
+## stable diffusion
+sources/go-stable-diffusion:
+	git clone --recurse-submodules https://github.com/mudler/go-stable-diffusion sources/go-stable-diffusion
+	cd sources/go-stable-diffusion && git checkout -b build $(STABLEDIFFUSION_VERSION) && git submodule update --init --recursive --depth 1
+
+sources/go-stable-diffusion/libstablediffusion.a:
+	$(MAKE) -C sources/go-stable-diffusion libstablediffusion.a
+
+## tiny-dream
+sources/go-tiny-dream:
+	git clone --recurse-submodules https://github.com/M0Rf30/go-tiny-dream sources/go-tiny-dream
+	cd sources/go-tiny-dream && git checkout -b build $(TINYDREAM_VERSION) && git submodule update --init --recursive --depth 1
+
+sources/go-tiny-dream/libtinydream.a:
+	$(MAKE) -C sources/go-tiny-dream libtinydream.a
+
+## RWKV
+sources/go-rwkv:
+	git clone --recurse-submodules $(RWKV_REPO) sources/go-rwkv
+	cd sources/go-rwkv && git checkout -b build $(RWKV_VERSION) && git submodule update --init --recursive --depth 1
+
+sources/go-rwkv/librwkv.a: sources/go-rwkv
+	cd sources/go-rwkv && cd rwkv.cpp &&	cmake . -DRWKV_BUILD_SHARED_LIBRARY=OFF &&	cmake --build . && 	cp librwkv.a ..
+
+sources/go-bert/libgobert.a: sources/go-bert
+	$(MAKE) -C sources/go-bert libgobert.a
+
+backend-assets/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/espeak-ng-data: sources/go-piper
+	mkdir -p backend-assets/espeak-ng-data
+	$(MAKE) -C sources/go-piper piper.o
+	@cp -rf sources/go-piper/piper-phonemize/pi/share/espeak-ng-data/. backend-assets/espeak-ng-data
+
+sources/gpt4all/gpt4all-bindings/golang/libgpt4all.a: sources/gpt4all
+	$(MAKE) -C sources/gpt4all/gpt4all-bindings/golang/ libgpt4all.a
+
+## CEREBRAS GPT
+sources/go-ggml-transformers:
+	git clone --recurse-submodules https://github.com/go-skynet/go-ggml-transformers.cpp sources/go-ggml-transformers
+	cd sources/go-ggml-transformers && git checkout -b build $(GOGPT2_VERSION) && git submodule update --init --recursive --depth 1
+
+sources/go-ggml-transformers/libtransformers.a: sources/go-ggml-transformers
+	$(MAKE) -C sources/go-ggml-transformers BUILD_TYPE=$(BUILD_TYPE) libtransformers.a
+
+sources/whisper.cpp:
+	git clone https://github.com/ggerganov/whisper.cpp.git sources/whisper.cpp
+	cd sources/whisper.cpp && git checkout -b build $(WHISPER_CPP_VERSION) && git submodule update --init --recursive --depth 1
+
+sources/whisper.cpp/libwhisper.a: sources/whisper.cpp
+	cd sources/whisper.cpp && make libwhisper.a
+
+sources/go-llama:
+	git clone --recurse-submodules https://github.com/go-skynet/go-llama.cpp sources/go-llama
+	cd sources/go-llama && git checkout -b build $(GOLLAMA_VERSION) && git submodule update --init --recursive --depth 1
+
+sources/go-llama-ggml:
+	git clone --recurse-submodules https://github.com/go-skynet/go-llama.cpp sources/go-llama-ggml
+	cd sources/go-llama-ggml && git checkout -b build $(GOLLAMA_STABLE_VERSION) && git submodule update --init --recursive --depth 1
+
+sources/go-llama/libbinding.a: sources/go-llama
+	$(MAKE) -C sources/go-llama BUILD_TYPE=$(BUILD_TYPE) libbinding.a
+
+sources/go-llama-ggml/libbinding.a: sources/go-llama-ggml
+	$(MAKE) -C sources/go-llama-ggml BUILD_TYPE=$(STABLE_BUILD_TYPE) libbinding.a
+
+sources/go-piper/libpiper_binding.a: sources/go-piper
+	$(MAKE) -C sources/go-piper libpiper_binding.a example/main
+
+backend/cpp/llama/llama.cpp:
+	LLAMA_VERSION=$(CPPLLAMA_VERSION) $(MAKE) -C backend/cpp/llama llama.cpp	
+
+get-sources: backend/cpp/llama/llama.cpp sources/go-llama sources/go-llama-ggml sources/go-ggml-transformers sources/gpt4all sources/go-piper sources/go-rwkv sources/whisper.cpp sources/go-bert sources/go-stable-diffusion sources/go-tiny-dream
+	touch $@
+
+replace:
+	$(GOCMD) mod edit -replace github.com/nomic-ai/gpt4all/gpt4all-bindings/golang=$(CURDIR)/sources/gpt4all/gpt4all-bindings/golang
+	$(GOCMD) mod edit -replace github.com/go-skynet/go-ggml-transformers.cpp=$(CURDIR)/sources/go-ggml-transformers
+	$(GOCMD) mod edit -replace github.com/donomii/go-rwkv.cpp=$(CURDIR)/sources/go-rwkv
+	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp=$(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
+	$(GOCMD) mod edit -replace github.com/mudler/go-stable-diffusion=$(CURDIR)/sources/go-stable-diffusion
+	$(GOCMD) mod edit -replace github.com/M0Rf30/go-tiny-dream=$(CURDIR)/sources/go-tiny-dream
+	$(GOCMD) mod edit -replace github.com/mudler/go-piper=$(CURDIR)/sources/go-piper
+
+prepare-sources: get-sources replace
+	$(GOCMD) mod download
+	touch $@
+
+## GENERIC
+rebuild: ## Rebuilds the project
+	$(GOCMD) clean -cache
+	$(MAKE) -C sources/go-llama clean
+	$(MAKE) -C sources/go-llama-ggml clean
+	$(MAKE) -C sources/gpt4all/gpt4all-bindings/golang/ clean
+	$(MAKE) -C sources/go-ggml-transformers clean
+	$(MAKE) -C sources/go-rwkv clean
+	$(MAKE) -C sources/whisper.cpp clean
+	$(MAKE) -C sources/go-stable-diffusion clean
+	$(MAKE) -C sources/go-bert clean
+	$(MAKE) -C sources/go-piper clean
+	$(MAKE) -C sources/go-tiny-dream clean
+	$(MAKE) build
+
+prepare: prepare-sources $(OPTIONAL_TARGETS)
+	touch $@
+
+clean: ## Remove build related file
+	$(GOCMD) clean -cache
+	rm -f prepare
+	rm -rf ./sources
+	rm -rf $(BINARY_NAME)
+	rm -rf release/
+	rm -rf backend-assets
+	$(MAKE) -C backend/cpp/grpc clean
+	$(MAKE) -C backend/cpp/llama clean
+
 ## Build:
 
-build: prepare ## Build the project
-	C_INCLUDE_PATH=$(shell pwd)/go-llama.cpp:$(shell pwd)/go-gpt4all-j LIBRARY_PATH=$(shell pwd)/go-llama.cpp:$(shell pwd)/go-gpt4all-j $(GOCMD) build -o $(BINARY_NAME) ./
+build: backend-assets grpcs prepare ## 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)${RESET})
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o $(BINARY_NAME) ./
 
-buildgeneric: prepare-generic ## Build the project
-	C_INCLUDE_PATH=$(shell pwd)/go-llama.cpp:$(shell pwd)/go-gpt4all-j LIBRARY_PATH=$(shell pwd)/go-llama.cpp:$(shell pwd)/go-gpt4all-j $(GOCMD) build -o $(BINARY_NAME) ./
+dist: build
+	mkdir -p release
+	cp $(BINARY_NAME) release/$(BINARY_NAME)-$(BUILD_ID)-$(OS)-$(ARCH)
 
-go-gpt4all-j:
-	git clone --recurse-submodules https://github.com/go-skynet/go-gpt4all-j.cpp go-gpt4all-j
-# This is hackish, but needed as both go-llama and go-gpt4allj have their own version of ggml..
-	@find ./go-gpt4all-j -type f -name "*.c" -exec sed -i'' -e 's/ggml_/ggml_gptj_/g' {} +
-	@find ./go-gpt4all-j -type f -name "*.cpp" -exec sed -i'' -e 's/ggml_/ggml_gptj_/g' {} +
-	@find ./go-gpt4all-j -type f -name "*.h" -exec sed -i'' -e 's/ggml_/ggml_gptj_/g' {} +
-	@find ./go-gpt4all-j -type f -name "*.cpp" -exec sed -i'' -e 's/gpt_/gptj_/g' {} +
-	@find ./go-gpt4all-j -type f -name "*.h" -exec sed -i'' -e 's/gpt_/gptj_/g' {} +
+osx-signed: build
+	codesign --deep --force --sign "$(OSX_SIGNING_IDENTITY)" --entitlements "./Entitlements.plist" "./$(BINARY_NAME)"
 
-go-gpt4all-j/libgptj.a: go-gpt4all-j
-	$(MAKE) -C go-gpt4all-j libgptj.a
+## Run
+run: prepare ## run local-ai
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GOCMD) run ./
 
-go-gpt4all-j/libgptj.a-generic: go-gpt4all-j
-	$(MAKE) -C go-gpt4all-j generic-libgptj.a
+test-models/testmodel:
+	mkdir test-models
+	mkdir test-dir
+	wget -q https://huggingface.co/nnakasato/ggml-model-test/resolve/main/ggml-model-q4.bin -O test-models/testmodel
+	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
 
-go-llama:
-	git clone -b $(GOLLAMA_VERSION) --recurse-submodules https://github.com/go-skynet/go-llama.cpp go-llama
-	$(MAKE) -C go-llama libbinding.a
+prepare-test: grpcs
+	cp -rf backend-assets api
+	cp tests/models_fixtures/* test-models
 
-go-llama-generic:
-	git clone -b $(GOLLAMA_VERSION) --recurse-submodules https://github.com/go-skynet/go-llama.cpp go-llama
-	$(MAKE) -C go-llama generic-libbinding.a
+test: prepare test-models/testmodel grpcs
+	@echo 'Running tests'
+	export GO_TAGS="tts stablediffusion"
+	$(MAKE) prepare-test
+	HUGGINGFACE_GRPC=$(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="!gpt4all && !llama && !llama-gguf"  --flake-attempts 5 --fail-fast -v -r ./api ./pkg
+	$(MAKE) test-gpt4all
+	$(MAKE) test-llama
+	$(MAKE) test-llama-gguf
+	$(MAKE) test-tts
+	$(MAKE) test-stablediffusion
 
-prepare: go-llama go-gpt4all-j/libgptj.a
-	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(shell pwd)/go-llama
-	$(GOCMD) mod edit -replace github.com/go-skynet/go-gpt4all-j.cpp=$(shell pwd)/go-gpt4all-j
+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 $(TEST_DIR)/ggllm-test-model.bin
+	docker build --build-arg BUILD_GRPC=true --build-arg GRPC_BACKENDS="$(GRPC_BACKENDS)" --build-arg IMAGE_TYPE=core --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg CUDA_MAJOR_VERSION=11 --build-arg CUDA_MINOR_VERSION=7 --build-arg FFMPEG=true -t localai-tests .
 
-prepare-generic: go-llama-generic go-gpt4all-j/libgptj.a-generic
-	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(shell pwd)/go-llama
-	$(GOCMD) mod edit -replace github.com/go-skynet/go-gpt4all-j.cpp=$(shell pwd)/go-gpt4all-j
-	
-clean: ## Remove build related file
-	rm -fr ./go-llama
-	rm -rf ./go-gpt4all-j
-	rm -rf $(BINARY_NAME)
+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:
-run: prepare
-	$(GOCMD) run ./ api
+test-e2e:
+	@echo 'Running e2e tests'
+	BUILD_TYPE=$(BUILD_TYPE) \
+	LOCALAI_API=http://$(E2E_BRIDGE_IP):5390/v1 \
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts 5 -v -r ./tests/e2e
 
-## Test:
-test: ## Run the tests of the project
-	$(GOTEST) -v -race ./... $(OUTPUT_OPTIONS)
+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 \
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="gpt4all" --flake-attempts 5 -v -r ./api ./pkg
+
+test-llama: prepare-test
+	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama" --flake-attempts 5 -v -r ./api ./pkg
+
+test-llama-gguf: prepare-test
+	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama-gguf" --flake-attempts 5 -v -r ./api ./pkg
+
+test-tts: prepare-test
+	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="tts" --flake-attempts 1 -v -r ./api ./pkg
+
+test-stablediffusion: prepare-test
+	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stablediffusion" --flake-attempts 1 -v -r ./api ./pkg
+
+test-container:
+	docker build --target requirements -t local-ai-test-container .
+	docker run -ti --rm --entrypoint /bin/bash -ti -v $(abspath ./):/build local-ai-test-container
 
 ## Help:
 help: ## Show this help.
@@ -77,3 +400,171 @@ help: ## Show this help.
 		if (/^[a-zA-Z_-]+:.*?##.*$$/) {printf "    ${YELLOW}%-20s${GREEN}%s${RESET}\n", $$1, $$2} \
 		else if (/^## .*$$/) {printf "  ${CYAN}%s${RESET}\n", substr($$1,4)} \
 		}' $(MAKEFILE_LIST)
+
+protogen: protogen-go protogen-python
+
+protogen-go:
+	protoc -Ibackend/ --go_out=pkg/grpc/proto/ --go_opt=paths=source_relative --go-grpc_out=pkg/grpc/proto/ --go-grpc_opt=paths=source_relative \
+    backend/backend.proto
+
+protogen-python:
+	python3 -m grpc_tools.protoc -Ibackend/ --python_out=backend/python/sentencetransformers/ --grpc_python_out=backend/python/sentencetransformers/ backend/backend.proto
+	python3 -m grpc_tools.protoc -Ibackend/ --python_out=backend/python/transformers/ --grpc_python_out=backend/python/transformers/ backend/backend.proto
+	python3 -m grpc_tools.protoc -Ibackend/ --python_out=backend/python/transformers-musicgen/ --grpc_python_out=backend/python/transformers-musicgen/ backend/backend.proto
+	python3 -m grpc_tools.protoc -Ibackend/ --python_out=backend/python/autogptq/ --grpc_python_out=backend/python/autogptq/ backend/backend.proto
+	python3 -m grpc_tools.protoc -Ibackend/ --python_out=backend/python/exllama/ --grpc_python_out=backend/python/exllama/ backend/backend.proto
+	python3 -m grpc_tools.protoc -Ibackend/ --python_out=backend/python/bark/ --grpc_python_out=backend/python/bark/ backend/backend.proto
+	python3 -m grpc_tools.protoc -Ibackend/ --python_out=backend/python/diffusers/ --grpc_python_out=backend/python/diffusers/ backend/backend.proto
+	python3 -m grpc_tools.protoc -Ibackend/ --python_out=backend/python/coqui/ --grpc_python_out=backend/python/coqui/ backend/backend.proto
+	python3 -m grpc_tools.protoc -Ibackend/ --python_out=backend/python/vall-e-x/ --grpc_python_out=backend/python/vall-e-x/ backend/backend.proto
+	python3 -m grpc_tools.protoc -Ibackend/ --python_out=backend/python/vllm/ --grpc_python_out=backend/python/vllm/ backend/backend.proto
+	python3 -m grpc_tools.protoc -Ibackend/ --python_out=backend/python/petals/ --grpc_python_out=backend/python/petals/ backend/backend.proto
+	python3 -m grpc_tools.protoc -Ibackend/ --python_out=backend/python/mamba/ --grpc_python_out=backend/python/mamba/ backend/backend.proto
+	python3 -m grpc_tools.protoc -Ibackend/ --python_out=backend/python/exllama2/ --grpc_python_out=backend/python/exllama2/ backend/backend.proto
+
+## GRPC
+# Note: it is duplicated in the Dockerfile
+prepare-extra-conda-environments:
+	$(MAKE) -C backend/python/autogptq
+	$(MAKE) -C backend/python/bark
+	$(MAKE) -C backend/python/coqui
+	$(MAKE) -C backend/python/diffusers
+	$(MAKE) -C backend/python/vllm
+	$(MAKE) -C backend/python/mamba
+	$(MAKE) -C backend/python/sentencetransformers
+	$(MAKE) -C backend/python/transformers
+	$(MAKE) -C backend/python/transformers-musicgen
+	$(MAKE) -C backend/python/vall-e-x
+	$(MAKE) -C backend/python/exllama
+	$(MAKE) -C backend/python/petals
+	$(MAKE) -C backend/python/exllama2
+
+prepare-test-extra:
+	$(MAKE) -C backend/python/transformers
+	$(MAKE) -C backend/python/diffusers
+
+test-extra: prepare-test-extra
+	$(MAKE) -C backend/python/transformers test
+	$(MAKE) -C backend/python/diffusers test
+
+backend-assets:
+	mkdir -p backend-assets
+ifeq ($(BUILD_API_ONLY),true)
+	touch backend-assets/keep
+endif
+
+backend-assets/grpc:
+	mkdir -p backend-assets/grpc
+
+backend-assets/grpc/llama: backend-assets/grpc sources/go-llama/libbinding.a
+	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(CURDIR)/sources/go-llama
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-llama LIBRARY_PATH=$(CURDIR)/sources/go-llama \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/llama ./backend/go/llm/llama/
+# TODO: every binary should have its own folder instead, so can have different  implementations
+ifeq ($(BUILD_TYPE),metal)
+	cp backend/cpp/llama/llama.cpp/ggml-metal.metal backend-assets/grpc/
+endif
+
+## BACKEND CPP LLAMA START
+# Sets the variables in case it has to build the gRPC locally.
+INSTALLED_PACKAGES=$(CURDIR)/backend/cpp/grpc/installed_packages
+INSTALLED_LIB_CMAKE=$(INSTALLED_PACKAGES)/lib/cmake
+ADDED_CMAKE_ARGS=-Dabsl_DIR=${INSTALLED_LIB_CMAKE}/absl \
+                 -DProtobuf_DIR=${INSTALLED_LIB_CMAKE}/protobuf \
+                 -Dutf8_range_DIR=${INSTALLED_LIB_CMAKE}/utf8_range \
+                 -DgRPC_DIR=${INSTALLED_LIB_CMAKE}/grpc \
+                 -DCMAKE_CXX_STANDARD_INCLUDE_DIRECTORIES=${INSTALLED_PACKAGES}/include
+
+backend/cpp/llama/grpc-server:
+ifdef BUILD_GRPC_FOR_BACKEND_LLAMA
+	$(MAKE) -C backend/cpp/grpc build
+	export _PROTOBUF_PROTOC=${INSTALLED_PACKAGES}/bin/proto && \
+	export _GRPC_CPP_PLUGIN_EXECUTABLE=${INSTALLED_PACKAGES}/bin/grpc_cpp_plugin && \
+	export PATH="${INSTALLED_PACKAGES}/bin:${PATH}" && \
+	CMAKE_ARGS="${CMAKE_ARGS} ${ADDED_CMAKE_ARGS}" LLAMA_VERSION=$(CPPLLAMA_VERSION) $(MAKE) -C backend/cpp/llama grpc-server
+else
+	echo "BUILD_GRPC_FOR_BACKEND_LLAMA is not defined."
+	LLAMA_VERSION=$(CPPLLAMA_VERSION) $(MAKE) -C backend/cpp/llama grpc-server			
+endif
+## BACKEND CPP LLAMA END
+
+##
+backend-assets/grpc/llama-cpp: backend-assets/grpc backend/cpp/llama/grpc-server
+	cp -rfv backend/cpp/llama/grpc-server backend-assets/grpc/llama-cpp
+# TODO: every binary should have its own folder instead, so can have different metal implementations
+ifeq ($(BUILD_TYPE),metal)
+	cp backend/cpp/llama/llama.cpp/build/bin/ggml-metal.metal backend-assets/grpc/
+endif
+
+backend-assets/grpc/llama-ggml: backend-assets/grpc sources/go-llama-ggml/libbinding.a
+	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(CURDIR)/sources/go-llama-ggml
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-llama-ggml LIBRARY_PATH=$(CURDIR)/sources/go-llama-ggml \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/llama-ggml ./backend/go/llm/llama-ggml/
+
+backend-assets/grpc/gpt4all: backend-assets/grpc backend-assets/gpt4all sources/gpt4all/gpt4all-bindings/golang/libgpt4all.a
+	CGO_LDFLAGS="$(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 "$(GO_TAGS)" -o backend-assets/grpc/gpt4all ./backend/go/llm/gpt4all/
+
+backend-assets/grpc/dolly: backend-assets/grpc sources/go-ggml-transformers/libtransformers.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-ggml-transformers LIBRARY_PATH=$(CURDIR)/sources/go-ggml-transformers \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/dolly ./backend/go/llm/dolly/
+
+backend-assets/grpc/gpt2: backend-assets/grpc sources/go-ggml-transformers/libtransformers.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-ggml-transformers LIBRARY_PATH=$(CURDIR)/sources/go-ggml-transformers \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gpt2 ./backend/go/llm/gpt2/
+
+backend-assets/grpc/gptj: backend-assets/grpc sources/go-ggml-transformers/libtransformers.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-ggml-transformers LIBRARY_PATH=$(CURDIR)/sources/go-ggml-transformers \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gptj ./backend/go/llm/gptj/
+
+backend-assets/grpc/gptneox: backend-assets/grpc sources/go-ggml-transformers/libtransformers.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-ggml-transformers LIBRARY_PATH=$(CURDIR)/sources/go-ggml-transformers \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gptneox ./backend/go/llm/gptneox/
+
+backend-assets/grpc/mpt: backend-assets/grpc sources/go-ggml-transformers/libtransformers.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-ggml-transformers LIBRARY_PATH=$(CURDIR)/sources/go-ggml-transformers \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/mpt ./backend/go/llm/mpt/
+
+backend-assets/grpc/replit: backend-assets/grpc sources/go-ggml-transformers/libtransformers.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-ggml-transformers LIBRARY_PATH=$(CURDIR)/sources/go-ggml-transformers \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/replit ./backend/go/llm/replit/
+
+backend-assets/grpc/falcon-ggml: backend-assets/grpc sources/go-ggml-transformers/libtransformers.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-ggml-transformers LIBRARY_PATH=$(CURDIR)/sources/go-ggml-transformers \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/falcon-ggml ./backend/go/llm/falcon-ggml/
+
+backend-assets/grpc/starcoder: backend-assets/grpc sources/go-ggml-transformers/libtransformers.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-ggml-transformers LIBRARY_PATH=$(CURDIR)/sources/go-ggml-transformers \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/starcoder ./backend/go/llm/starcoder/
+
+backend-assets/grpc/rwkv: backend-assets/grpc sources/go-rwkv/librwkv.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-rwkv LIBRARY_PATH=$(CURDIR)/sources/go-rwkv \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/rwkv ./backend/go/llm/rwkv
+
+backend-assets/grpc/bert-embeddings: backend-assets/grpc sources/go-bert/libgobert.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-bert LIBRARY_PATH=$(CURDIR)/sources/go-bert \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/bert-embeddings ./backend/go/llm/bert/
+
+backend-assets/grpc/langchain-huggingface: backend-assets/grpc
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/langchain-huggingface ./backend/go/llm/langchain/
+
+backend-assets/grpc/stablediffusion: backend-assets/grpc
+	if [ ! -f backend-assets/grpc/stablediffusion ]; then \
+		$(MAKE) sources/go-stable-diffusion/libstablediffusion.a; \
+		CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-stable-diffusion/ LIBRARY_PATH=$(CURDIR)/sources/go-stable-diffusion/ \
+		$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/stablediffusion ./backend/go/image/stablediffusion; \
+	fi
+
+backend-assets/grpc/tinydream: backend-assets/grpc sources/go-tiny-dream/libtinydream.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" LIBRARY_PATH=$(CURDIR)/go-tiny-dream \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/tinydream ./backend/go/image/tinydream
+
+backend-assets/grpc/piper: backend-assets/grpc backend-assets/espeak-ng-data sources/go-piper/libpiper_binding.a
+	CGO_CXXFLAGS="$(PIPER_CGO_CXXFLAGS)" CGO_LDFLAGS="$(PIPER_CGO_LDFLAGS)" LIBRARY_PATH=$(CURDIR)/sources/go-piper \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/piper ./backend/go/tts/
+
+backend-assets/grpc/whisper: backend-assets/grpc sources/whisper.cpp/libwhisper.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/whisper.cpp LIBRARY_PATH=$(CURDIR)/sources/whisper.cpp \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/whisper ./backend/go/transcribe/
+
+grpcs: prepare $(GRPC_BACKENDS)

README.md

@@ -1,219 +1,186 @@
 <h1 align="center">
   <br>
-  <img height="300" src="https://user-images.githubusercontent.com/2420543/233147843-88697415-6dbf-4368-a862-ab217f9f7342.jpeg"> <br>
+  <img height="300" src="https://github.com/go-skynet/LocalAI/assets/2420543/0966aa2a-166e-4f99-a3e5-6c915fc997dd"> <br>
     LocalAI
 <br>
 </h1>
 
-> :warning: This project has been renamed from `llama-cli` to `LocalAI` to reflect the fact that we are focusing on a fast drop-in OpenAI API rather on the CLI interface. We think that there are already many projects that can be used as a CLI interface already, for instance  [llama.cpp](https://github.com/ggerganov/llama.cpp) and [gpt4all](https://github.com/nomic-ai/gpt4all). If you are were using `llama-cli` for CLI interactions and want to keep using it, use older versions or please open up an issue - contributions are welcome!
+<p align="center">
+<a href="https://github.com/go-skynet/LocalAI/fork" target="blank">
+<img src="https://img.shields.io/github/forks/go-skynet/LocalAI?style=for-the-badge" alt="LocalAI forks"/>
+</a>
+<a href="https://github.com/go-skynet/LocalAI/stargazers" target="blank">
+<img src="https://img.shields.io/github/stars/go-skynet/LocalAI?style=for-the-badge" alt="LocalAI stars"/>
+</a>
+<a href="https://github.com/go-skynet/LocalAI/pulls" target="blank">
+<img src="https://img.shields.io/github/issues-pr/go-skynet/LocalAI?style=for-the-badge" alt="LocalAI pull-requests"/>
+</a>
+<a href='https://github.com/go-skynet/LocalAI/releases'>
+<img src='https://img.shields.io/github/release/go-skynet/LocalAI?&label=Latest&style=for-the-badge'>
+</a>
+</p>
 
-LocalAI is a straightforward, drop-in replacement API compatible with OpenAI for local CPU inferencing, based on [llama.cpp](https://github.com/ggerganov/llama.cpp), [gpt4all](https://github.com/nomic-ai/gpt4all) and [ggml](https://github.com/ggerganov/ggml), including support GPT4ALL-J which is Apache 2.0 Licensed and can be used for commercial purposes.
+[<img src="https://img.shields.io/badge/dockerhub-images-important.svg?logo=Docker">](https://hub.docker.com/r/localai/localai)
+[<img src="https://img.shields.io/badge/quay.io-images-important.svg?">](https://quay.io/repository/go-skynet/local-ai?tab=tags&tag=latest)
 
-- OpenAI compatible API
-- Supports multiple-models
-- Once loaded the first time, it keep models loaded in memory for faster inference
-- Provides a simple command line interface that allows text generation directly from the terminal
-- Support for prompt templates
-- Doesn't shell-out, but uses C bindings for a faster inference and better performance. Uses [go-llama.cpp](https://github.com/go-skynet/go-llama.cpp) and [go-gpt4all-j.cpp](https://github.com/go-skynet/go-gpt4all-j.cpp).
+> :bulb: Get help - [❓FAQ](https://localai.io/faq/) [💭Discussions](https://github.com/go-skynet/LocalAI/discussions) [:speech_balloon: Discord](https://discord.gg/uJAeKSAGDy) [:book: Documentation website](https://localai.io/)
+>
+> [💻 Quickstart](https://localai.io/basics/getting_started/) [📣 News](https://localai.io/basics/news/) [ 🛫 Examples ](https://github.com/go-skynet/LocalAI/tree/master/examples/) [ 🖼️ Models ](https://localai.io/models/) [ 🚀 Roadmap ](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap)
 
-## Model compatibility
+[![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)
 
-It is compatible with the models supported by [llama.cpp](https://github.com/ggerganov/llama.cpp) and also [GPT4ALL-J](https://github.com/nomic-ai/gpt4all).
+<p align="center">
+<a href="https://twitter.com/LocalAI_API" target="blank">
+<img src="https://img.shields.io/twitter/follow/LocalAI_API?label=Follow: LocalAI_API&style=social" alt="Follow LocalAI_API"/>
+</a>
+<a href="https://discord.gg/uJAeKSAGDy" target="blank">
+<img src="https://dcbadge.vercel.app/api/server/uJAeKSAGDy?style=flat-square&theme=default-inverted" alt="Join LocalAI Discord Community"/>
+</a>
 
-Note: You might need to convert older models to the new format, see [here](https://github.com/ggerganov/llama.cpp#using-gpt4all) for instance to run `gpt4all`.
+**LocalAI** is the free, Open Source OpenAI alternative. LocalAI act as a drop-in replacement REST API that’s compatible with OpenAI API specifications for local inferencing. It allows you to run LLMs, generate images, audio (and not only) locally or on-prem with consumer grade hardware, supporting multiple model families. Does not require GPU.
 
-## Usage
+## 🔥🔥 Hot topics / Roadmap
 
-> `LocalAI` comes by default as a container image. You can check out all the available images with corresponding tags [here](https://quay.io/repository/go-skynet/local-ai?tab=tags&tag=latest).
+[Roadmap](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap)
 
-The easiest way to run LocalAI is by using `docker-compose`:
+- Mamba support: https://github.com/mudler/LocalAI/pull/1589
+- Start and share models with config file: https://github.com/mudler/LocalAI/pull/1522
+- 🐸 Coqui: https://github.com/mudler/LocalAI/pull/1489
+- Inline templates: https://github.com/mudler/LocalAI/pull/1452
+- Mixtral: https://github.com/mudler/LocalAI/pull/1449
+- Img2vid https://github.com/mudler/LocalAI/pull/1442
+- Musicgen https://github.com/mudler/LocalAI/pull/1387
 
-```bash
+Hot topics (looking for contributors):
+- Backends v2: https://github.com/mudler/LocalAI/issues/1126
+- Improving UX v2: https://github.com/mudler/LocalAI/issues/1373
 
-git clone https://github.com/go-skynet/LocalAI
+If you want to help and contribute, issues up for grabs: https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3A%22up+for+grabs%22
 
-cd LocalAI
+## 💻 [Getting started](https://localai.io/basics/getting_started/index.html)
 
-# copy your models to models/
-cp your-model.bin models/
-
-# (optional) Edit the .env file to set things like context size and threads
-# vim .env
-
-# start with docker-compose
-docker compose up -d --build
-
-# Now API is accessible at localhost:8080
-curl http://localhost:8080/v1/models
-# {"object":"list","data":[{"id":"your-model.bin","object":"model"}]}
-
-curl http://localhost:8080/v1/completions -H "Content-Type: application/json" -d '{
-     "model": "your-model.bin",            
-     "prompt": "A long time ago in a galaxy far, far away",
-     "temperature": 0.7
-   }'
-```
-
-## Prompt templates 
-
-The API doesn't inject a default prompt for talking to the model. You have to use a prompt similar to what's described in the standford-alpaca docs: https://github.com/tatsu-lab/stanford_alpaca#data-release.
-
-<details>
-You can use a default template for every model present in your model path, by creating a corresponding file with the `.tmpl` suffix next to your model. For instance, if the model is called `foo.bin`, you can create a sibiling file, `foo.bin.tmpl` which will be used as a default prompt, for instance this can be used with alpaca:
+For a detailed step-by-step introduction, refer to the [Getting Started](https://localai.io/basics/getting_started/index.html) guide. For those in a hurry, here's a straightforward one-liner to launch a LocalAI instance with [phi-2](https://huggingface.co/microsoft/phi-2) using `docker`:
 
 ```
-Below is an instruction that describes a task. Write a response that appropriately completes the request.
-
-### Instruction:
-{{.Input}}
-
-### Response:
+docker run -ti -p 8080:8080 localai/localai:v2.5.1-ffmpeg-core phi-2
 ```
 
-See the [prompt-templates](https://github.com/go-skynet/LocalAI/tree/master/prompt-templates) directory in this repository for templates for most popular models.
+## 🚀 [Features](https://localai.io/features/)
 
-</details>
+- 📖 [Text generation with GPTs](https://localai.io/features/text-generation/) (`llama.cpp`, `gpt4all.cpp`, ... [:book: and more](https://localai.io/model-compatibility/index.html#model-compatibility-table))
+- 🗣 [Text to Audio](https://localai.io/features/text-to-audio/)
+- 🔈 [Audio to Text](https://localai.io/features/audio-to-text/) (Audio transcription with `whisper.cpp`)
+- 🎨 [Image generation with stable diffusion](https://localai.io/features/image-generation)
+- 🔥 [OpenAI functions](https://localai.io/features/openai-functions/) 🆕
+- 🧠 [Embeddings generation for vector databases](https://localai.io/features/embeddings/)
+- ✍️ [Constrained grammars](https://localai.io/features/constrained_grammars/)
+- 🖼️ [Download Models directly from Huggingface ](https://localai.io/models/)
+- 🆕 [Vision API](https://localai.io/features/gpt-vision/)
 
-## API
+## 💻 Usage
 
-`LocalAI` provides an API for running text generation as a service, that follows the OpenAI reference and can be used as a drop-in. The models once loaded the first time will be kept in memory.
+Check out the [Getting started](https://localai.io/basics/getting_started/index.html) section in our documentation.
 
-<details>
-Example of starting the API with `docker`:
+### 🔗 Community and integrations
 
-```bash
-docker run -p 8080:8080 -ti --rm quay.io/go-skynet/local-ai:latest --models-path /path/to/models --context-size 700 --threads 4
-```
+Build and deploy custom containers:
+- https://github.com/sozercan/aikit
 
-And you'll see:
-```
-┌───────────────────────────────────────────────────┐ 
-│                   Fiber v2.42.0                   │ 
-│               http://127.0.0.1:8080               │ 
-│       (bound on host 0.0.0.0 and port 8080)       │ 
-│                                                   │ 
-│ Handlers ............. 1  Processes ........... 1 │ 
-│ Prefork ....... Disabled  PID ................. 1 │ 
-└───────────────────────────────────────────────────┘ 
-```
+WebUIs:
+- https://github.com/Jirubizu/localai-admin
+- https://github.com/go-skynet/LocalAI-frontend
 
-Note: Models have to end up with `.bin` so can be listed by the `/models` endpoint.
+Model galleries
+- https://github.com/go-skynet/model-gallery
+  
+Auto Docker / Model setup
+- https://io.midori-ai.xyz/howtos/easy-localai-installer/
+- https://io.midori-ai.xyz/howtos/easy-model-installer/
 
-You can control the API server options with command line arguments:
+Other:
+- Helm chart https://github.com/go-skynet/helm-charts
+- VSCode extension https://github.com/badgooooor/localai-vscode-plugin
+- Local Smart assistant https://github.com/mudler/LocalAGI
+- Home Assistant https://github.com/sammcj/homeassistant-localai / https://github.com/drndos/hass-openai-custom-conversation
+- Discord bot https://github.com/mudler/LocalAGI/tree/main/examples/discord
+- Slack bot https://github.com/mudler/LocalAGI/tree/main/examples/slack
+- Telegram bot https://github.com/mudler/LocalAI/tree/master/examples/telegram-bot
+- Examples: https://github.com/mudler/LocalAI/tree/master/examples/
+
+### 🔗 Resources
+
+- 🆕 New! [LLM finetuning guide](https://localai.io/advanced/fine-tuning/)
+- [How to build locally](https://localai.io/basics/build/index.html)
+- [How to install in Kubernetes](https://localai.io/basics/getting_started/index.html#run-localai-in-kubernetes)
+- [Projects integrating LocalAI](https://localai.io/integrations/)
+- [How tos section](https://io.midori-ai.xyz/howtos/) (curated by our community)
+
+## :book: 🎥 [Media, Blogs, Social](https://localai.io/basics/news/#media-blogs-social)
+
+- [Create a slackbot for teams and OSS projects that answer to documentation](https://mudler.pm/posts/smart-slackbot-for-teams/)
+- [LocalAI meets k8sgpt](https://www.youtube.com/watch?v=PKrDNuJ_dfE)
+- [Question Answering on Documents locally with LangChain, LocalAI, Chroma, and GPT4All](https://mudler.pm/posts/localai-question-answering/)
+- [Tutorial to use k8sgpt with LocalAI](https://medium.com/@tyler_97636/k8sgpt-localai-unlock-kubernetes-superpowers-for-free-584790de9b65)
+
+## Citation
+
+If you utilize this repository, data in a downstream project, please consider citing it with:
 
 ```
-local-api --models-path <model_path> [--address <address>] [--threads <num_threads>]
+@misc{localai,
+  author = {Ettore Di Giacinto},
+  title = {LocalAI: The free, Open source OpenAI alternative},
+  year = {2023},
+  publisher = {GitHub},
+  journal = {GitHub repository},
+  howpublished = {\url{https://github.com/go-skynet/LocalAI}},
 ```
 
-The API takes takes the following parameters:
+## ❤️ Sponsors
 
-| Parameter    | Environment Variable | Default Value | Description                            |
-| ------------ | -------------------- | ------------- | -------------------------------------- |
-| models-path        | MODELS_PATH           |               | The path where you have models (ending with `.bin`).      |
-| threads      | THREADS              | CPU cores     | The number of threads to use for text generation. |
-| address      | ADDRESS              | :8080         | The address and port to listen on. |
-| context-size | CONTEXT_SIZE         | 512           | Default token context size. |
+> Do you find LocalAI useful?
 
-Once the server is running, you can start making requests to it using HTTP, using the OpenAI API. 
+Support the project by becoming [a backer or sponsor](https://github.com/sponsors/mudler). Your logo will show up here with a link to your website.
 
-</details>
+A huge thank you to our generous sponsors who support this project:
 
-### Supported OpenAI API endpoints
+| ![Spectro Cloud logo_600x600px_transparent bg](https://github.com/go-skynet/LocalAI/assets/2420543/68a6f3cb-8a65-4a4d-99b5-6417a8905512) |
+|:-----------------------------------------------:|
+|  [Spectro Cloud](https://www.spectrocloud.com/)  |
+|  Spectro Cloud kindly supports LocalAI by providing GPU and computing resources to run tests on lamdalabs!  |
 
-You can check out the [OpenAI API reference](https://platform.openai.com/docs/api-reference/chat/create). 
+And a huge shout-out to individuals sponsoring the project by donating hardware or backing the project.
 
-Following the list of endpoints/parameters supported.
+- [Sponsor list](https://github.com/sponsors/mudler)
+- JDAM00 (donating HW for the CI)
 
-#### Chat completions
+## 🌟 Star history
 
-For example, to generate a chat completion, you can send a POST request to the `/v1/chat/completions` endpoint with the instruction as the request body:
+[![LocalAI Star history Chart](https://api.star-history.com/svg?repos=go-skynet/LocalAI&type=Date)](https://star-history.com/#go-skynet/LocalAI&Date)
 
-```
-curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
-     "model": "ggml-koala-7b-model-q4_0-r2.bin",
-     "messages": [{"role": "user", "content": "Say this is a test!"}],
-     "temperature": 0.7
-   }'
-```
+## 📖 License
 
-Available additional parameters: `top_p`, `top_k`, `max_tokens`
+LocalAI is a community-driven project created by [Ettore Di Giacinto](https://github.com/mudler/).
 
-#### Completions
+MIT - Author Ettore Di Giacinto
 
-For example, to generate a comletion, you can send a POST request to the `/v1/completions` endpoint with the instruction as the request body:
-```
-curl http://localhost:8080/v1/completions -H "Content-Type: application/json" -d '{
-     "model": "ggml-koala-7b-model-q4_0-r2.bin",
-     "prompt": "A long time ago in a galaxy far, far away",
-     "temperature": 0.7
-   }'
-```
+## 🙇 Acknowledgements
 
-Available additional parameters: `top_p`, `top_k`, `max_tokens`
-
-#### List models
-
-You can list all the models available with:
-
-```
-curl http://localhost:8080/v1/models
-```
-
-## Using other models
-
-gpt4all (https://github.com/nomic-ai/gpt4all) works as well, however the original model needs to be converted (same applies for old alpaca models, too):
-
-```bash
-wget -O tokenizer.model https://huggingface.co/decapoda-research/llama-30b-hf/resolve/main/tokenizer.model
-mkdir models
-cp gpt4all.. models/
-git clone https://gist.github.com/eiz/828bddec6162a023114ce19146cb2b82
-pip install sentencepiece
-python 828bddec6162a023114ce19146cb2b82/gistfile1.txt models tokenizer.model
-# There will be a new model with the ".tmp" extension, you have to use that one!
-```
-
-### Windows compatibility
-
-It should work, however you need to make sure you give enough resources to the container. See https://github.com/go-skynet/LocalAI/issues/2
-
-### Kubernetes
-
-You can run the API in Kubernetes, see an example deployment in [kubernetes](https://github.com/go-skynet/LocalAI/tree/master/kubernetes)
-
-### Build locally
-
-Pre-built images might fit well for most of the modern hardware, however you can and might need to build the images manually.
-
-In order to build the `LocalAI` container image locally you can use `docker`:
-
-```
-# build the image
-docker build -t LocalAI .
-docker run LocalAI
-```
-
-Or build the binary with `make`:
-
-```
-make build
-```
-
-## Short-term roadmap
-
-- [x] Mimic OpenAI API (https://github.com/go-skynet/LocalAI/issues/10)
-- Binary releases (https://github.com/go-skynet/LocalAI/issues/6)
-- Upstream our golang bindings to llama.cpp (https://github.com/ggerganov/llama.cpp/issues/351)
-- [x] Multi-model support
-- Have a webUI!
-
-## License
-
-MIT
-
-## Acknowledgements
+LocalAI couldn't have been built without the help of great software already available from the community. Thank you!
 
 - [llama.cpp](https://github.com/ggerganov/llama.cpp)
 - https://github.com/tatsu-lab/stanford_alpaca
 - https://github.com/cornelk/llama-go for the initial ideas
-- https://github.com/antimatter15/alpaca.cpp for the light model version (this is compatible and tested only with that checkpoint model!)
+- 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
+- https://github.com/cmp-nct/ggllm.cpp
+
+## 🤗 Contributors
+
+This is a community project, a special thanks to our contributors! 🤗
+<a href="https://github.com/go-skynet/LocalAI/graphs/contributors">
+  <img src="https://contrib.rocks/image?repo=go-skynet/LocalAI" />
+</a>

api/api.go

@@ -4,288 +4,121 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"os"
 	"strings"
-	"sync"
 
-	model "github.com/go-skynet/LocalAI/pkg/model"
-	gptj "github.com/go-skynet/go-gpt4all-j.cpp"
-	llama "github.com/go-skynet/go-llama.cpp"
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/localai"
+	"github.com/go-skynet/LocalAI/api/openai"
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/api/schema"
+	"github.com/go-skynet/LocalAI/internal"
+	"github.com/go-skynet/LocalAI/metrics"
+	"github.com/go-skynet/LocalAI/pkg/assets"
+	"github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/go-skynet/LocalAI/pkg/startup"
+
 	"github.com/gofiber/fiber/v2"
 	"github.com/gofiber/fiber/v2/middleware/cors"
+	"github.com/gofiber/fiber/v2/middleware/logger"
 	"github.com/gofiber/fiber/v2/middleware/recover"
+	"github.com/rs/zerolog"
 	"github.com/rs/zerolog/log"
 )
 
-type OpenAIResponse struct {
-	Created int      `json:"created,omitempty"`
-	Object  string   `json:"chat.completion,omitempty"`
-	ID      string   `json:"id,omitempty"`
-	Model   string   `json:"model,omitempty"`
-	Choices []Choice `json:"choices,omitempty"`
-}
+func Startup(opts ...options.AppOption) (*options.Option, *config.ConfigLoader, error) {
+	options := options.NewOptions(opts...)
 
-type Choice struct {
-	Index        int      `json:"index,omitempty"`
-	FinishReason string   `json:"finish_reason,omitempty"`
-	Message      *Message `json:"message,omitempty"`
-	Text         string   `json:"text,omitempty"`
-}
-
-type Message struct {
-	Role    string `json:"role,omitempty"`
-	Content string `json:"content,omitempty"`
-}
-
-type OpenAIModel struct {
-	ID     string `json:"id"`
-	Object string `json:"object"`
-}
-
-type OpenAIRequest struct {
-	Model string `json:"model"`
-
-	// Prompt is read only by completion API calls
-	Prompt string `json:"prompt"`
-
-	// Messages is read only by chat/completion API calls
-	Messages []Message `json:"messages"`
-
-	Echo bool `json:"echo"`
-	// Common options between all the API calls
-	TopP        float64 `json:"top_p"`
-	TopK        int     `json:"top_k"`
-	Temperature float64 `json:"temperature"`
-	Maxtokens   int     `json:"max_tokens"`
-
-	N int `json:"n"`
-
-	// Custom parameters - not present in the OpenAI API
-	Batch     int  `json:"batch"`
-	F16       bool `json:"f16kv"`
-	IgnoreEOS bool `json:"ignore_eos"`
-
-	Seed int `json:"seed"`
-}
-
-// https://platform.openai.com/docs/api-reference/completions
-func openAIEndpoint(chat bool, loader *model.ModelLoader, threads, ctx int, f16 bool, mutexMap *sync.Mutex, mutexes map[string]*sync.Mutex) func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-		var err error
-		var model *llama.LLama
-		var gptModel *gptj.GPTJ
-
-		input := new(OpenAIRequest)
-		// Get input data from the request body
-		if err := c.BodyParser(input); err != nil {
-			return err
-		}
-		modelFile := input.Model
-		received, _ := json.Marshal(input)
-
-		log.Debug().Msgf("Request received: %s", string(received))
-
-		// Set model from bearer token, if available
-		bearer := strings.TrimLeft(c.Get("authorization"), "Bearer ")
-		bearerExists := bearer != "" && loader.ExistsInModelPath(bearer)
-		if modelFile == "" && !bearerExists {
-			return fmt.Errorf("no model specified")
-		}
-
-		if bearerExists { // model specified in bearer token takes precedence
-			log.Debug().Msgf("Using model from bearer token: %s", bearer)
-			modelFile = bearer
-		}
-
-		// Try to load the model with both
-		var llamaerr error
-		llamaOpts := []llama.ModelOption{}
-		if ctx != 0 {
-			llamaOpts = append(llamaOpts, llama.SetContext(ctx))
-		}
-		if f16 {
-			llamaOpts = append(llamaOpts, llama.EnableF16Memory)
-		}
-
-		model, llamaerr = loader.LoadLLaMAModel(modelFile, llamaOpts...)
-		if llamaerr != nil {
-			gptModel, err = loader.LoadGPTJModel(modelFile)
-			if err != nil {
-				return fmt.Errorf("llama: %s gpt: %s", llamaerr.Error(), err.Error()) // llama failed first, so we want to catch both errors
-			}
-		}
-
-		// This is still needed, see: https://github.com/ggerganov/llama.cpp/discussions/784
-		mutexMap.Lock()
-		l, ok := mutexes[modelFile]
-		if !ok {
-			m := &sync.Mutex{}
-			mutexes[modelFile] = m
-			l = m
-		}
-		mutexMap.Unlock()
-		l.Lock()
-		defer l.Unlock()
-
-		// Set the parameters for the language model prediction
-		topP := input.TopP
-		if topP == 0 {
-			topP = 0.7
-		}
-		topK := input.TopK
-		if topK == 0 {
-			topK = 80
-		}
-
-		temperature := input.Temperature
-		if temperature == 0 {
-			temperature = 0.9
-		}
-
-		tokens := input.Maxtokens
-		if tokens == 0 {
-			tokens = 512
-		}
-
-		predInput := input.Prompt
-		if chat {
-			mess := []string{}
-			// TODO: encode roles
-			for _, i := range input.Messages {
-				mess = append(mess, i.Content)
-			}
-
-			predInput = strings.Join(mess, "\n")
-		}
-
-		// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
-		templatedInput, err := loader.TemplatePrefix(modelFile, struct {
-			Input string
-		}{Input: predInput})
-		if err == nil {
-			predInput = templatedInput
-			log.Debug().Msgf("Template found, input modified to: %s", predInput)
-		}
-
-		result := []Choice{}
-
-		n := input.N
-
-		if input.N == 0 {
-			n = 1
-		}
-
-		var predFunc func() (string, error)
-		switch {
-		case gptModel != nil:
-			predFunc = func() (string, error) {
-				// Generate the prediction using the language model
-				predictOptions := []gptj.PredictOption{
-					gptj.SetTemperature(temperature),
-					gptj.SetTopP(topP),
-					gptj.SetTopK(topK),
-					gptj.SetTokens(tokens),
-					gptj.SetThreads(threads),
-				}
-
-				if input.Batch != 0 {
-					predictOptions = append(predictOptions, gptj.SetBatch(input.Batch))
-				}
-
-				if input.Seed != 0 {
-					predictOptions = append(predictOptions, gptj.SetSeed(input.Seed))
-				}
-
-				return gptModel.Predict(
-					predInput,
-					predictOptions...,
-				)
-			}
-		case model != nil:
-			predFunc = func() (string, error) {
-				// Generate the prediction using the language model
-				predictOptions := []llama.PredictOption{
-					llama.SetTemperature(temperature),
-					llama.SetTopP(topP),
-					llama.SetTopK(topK),
-					llama.SetTokens(tokens),
-					llama.SetThreads(threads),
-				}
-
-				if input.Batch != 0 {
-					predictOptions = append(predictOptions, llama.SetBatch(input.Batch))
-				}
-
-				if input.F16 {
-					predictOptions = append(predictOptions, llama.EnableF16KV)
-				}
-
-				if input.IgnoreEOS {
-					predictOptions = append(predictOptions, llama.IgnoreEOS)
-				}
-
-				if input.Seed != 0 {
-					predictOptions = append(predictOptions, llama.SetSeed(input.Seed))
-				}
-
-				return model.Predict(
-					predInput,
-					predictOptions...,
-				)
-			}
-		}
-
-		for i := 0; i < n; i++ {
-			prediction, err := predFunc()
-			if err != nil {
-				return err
-			}
-
-			if input.Echo {
-				prediction = predInput + prediction
-			}
-
-			if chat {
-				result = append(result, Choice{Message: &Message{Role: "assistant", Content: prediction}})
-			} else {
-				result = append(result, Choice{Text: prediction})
-			}
-		}
-
-		jsonResult, _ := json.Marshal(result)
-		log.Debug().Msgf("Response: %s", jsonResult)
-
-		// Return the prediction in the response body
-		return c.JSON(OpenAIResponse{
-			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
-			Choices: result,
-		})
+	zerolog.SetGlobalLevel(zerolog.InfoLevel)
+	if options.Debug {
+		zerolog.SetGlobalLevel(zerolog.DebugLevel)
 	}
-}
 
-func listModels(loader *model.ModelLoader) func(ctx *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-		models, err := loader.ListModels()
+	log.Info().Msgf("Starting LocalAI using %d threads, with models path: %s", options.Threads, options.Loader.ModelPath)
+	log.Info().Msgf("LocalAI version: %s", internal.PrintableVersion())
+
+	startup.PreloadModelsConfigurations(options.ModelLibraryURL, options.Loader.ModelPath, options.ModelsURL...)
+
+	cl := config.NewConfigLoader()
+	if err := cl.LoadConfigs(options.Loader.ModelPath); err != nil {
+		log.Error().Msgf("error loading config files: %s", err.Error())
+	}
+
+	if options.ConfigFile != "" {
+		if err := cl.LoadConfigFile(options.ConfigFile); err != nil {
+			log.Error().Msgf("error loading config file: %s", err.Error())
+		}
+	}
+
+	if err := cl.Preload(options.Loader.ModelPath); err != nil {
+		log.Error().Msgf("error downloading models: %s", err.Error())
+	}
+
+	if options.PreloadJSONModels != "" {
+		if err := localai.ApplyGalleryFromString(options.Loader.ModelPath, options.PreloadJSONModels, cl, options.Galleries); err != nil {
+			return nil, nil, err
+		}
+	}
+
+	if options.PreloadModelsFromPath != "" {
+		if err := localai.ApplyGalleryFromFile(options.Loader.ModelPath, options.PreloadModelsFromPath, cl, options.Galleries); err != nil {
+			return nil, nil, err
+		}
+	}
+
+	if options.Debug {
+		for _, v := range cl.ListConfigs() {
+			cfg, _ := cl.GetConfig(v)
+			log.Debug().Msgf("Model: %s (config: %+v)", v, cfg)
+		}
+	}
+
+	if options.AssetsDestination != "" {
+		// Extract files from the embedded FS
+		err := assets.ExtractFiles(options.BackendAssets, options.AssetsDestination)
+		log.Debug().Msgf("Extracting backend assets files to %s", options.AssetsDestination)
 		if err != nil {
-			return err
+			log.Warn().Msgf("Failed extracting backend assets files: %s (might be required for some backends to work properly, like gpt4all)", err)
 		}
-
-		dataModels := []OpenAIModel{}
-		for _, m := range models {
-			dataModels = append(dataModels, OpenAIModel{ID: m, Object: "model"})
-		}
-		return c.JSON(struct {
-			Object string        `json:"object"`
-			Data   []OpenAIModel `json:"data"`
-		}{
-			Object: "list",
-			Data:   dataModels,
-		})
 	}
+
+	// turn off any process that was started by GRPC if the context is canceled
+	go func() {
+		<-options.Context.Done()
+		log.Debug().Msgf("Context canceled, shutting down")
+		options.Loader.StopAllGRPC()
+	}()
+
+	if options.WatchDog {
+		wd := model.NewWatchDog(
+			options.Loader,
+			options.WatchDogBusyTimeout,
+			options.WatchDogIdleTimeout,
+			options.WatchDogBusy,
+			options.WatchDogIdle)
+		options.Loader.SetWatchDog(wd)
+		go wd.Run()
+		go func() {
+			<-options.Context.Done()
+			log.Debug().Msgf("Context canceled, shutting down")
+			wd.Shutdown()
+		}()
+	}
+
+	return options, cl, nil
 }
 
-func Start(loader *model.ModelLoader, listenAddr string, threads, ctxSize int, f16 bool) error {
+func App(opts ...options.AppOption) (*fiber.App, error) {
+
+	options, cl, err := Startup(opts...)
+	if err != nil {
+		return nil, fmt.Errorf("failed basic startup tasks with error %s", err.Error())
+	}
+
 	// Return errors as JSON responses
 	app := fiber.New(fiber.Config{
+		BodyLimit:             options.UploadLimitMB * 1024 * 1024, // this is the default limit of 4MB
+		DisableStartupMessage: options.DisableMessage,
 		// Override default error handler
 		ErrorHandler: func(ctx *fiber.Ctx, err error) error {
 			// Status code defaults to 500
@@ -298,31 +131,158 @@ func Start(loader *model.ModelLoader, listenAddr string, threads, ctxSize int, f
 			}
 
 			// Send custom error page
-			return ctx.Status(code).JSON(struct {
-				Error string `json:"error"`
-			}{Error: err.Error()})
+			return ctx.Status(code).JSON(
+				schema.ErrorResponse{
+					Error: &schema.APIError{Message: err.Error(), Code: code},
+				},
+			)
 		},
 	})
 
+	if options.Debug {
+		app.Use(logger.New(logger.Config{
+			Format: "[${ip}]:${port} ${status} - ${method} ${path}\n",
+		}))
+	}
+
 	// Default middleware config
 	app.Use(recover.New())
-	app.Use(cors.New())
+	if options.Metrics != nil {
+		app.Use(metrics.APIMiddleware(options.Metrics))
+	}
 
-	// This is still needed, see: https://github.com/ggerganov/llama.cpp/discussions/784
-	mu := map[string]*sync.Mutex{}
-	var mumutex = &sync.Mutex{}
+	// Auth middleware checking if API key is valid. If no API key is set, no auth is required.
+	auth := func(c *fiber.Ctx) error {
+		if len(options.ApiKeys) == 0 {
+			return c.Next()
+		}
+
+		// Check for api_keys.json file
+		fileContent, err := os.ReadFile("api_keys.json")
+		if err == nil {
+			// Parse JSON content from the file
+			var fileKeys []string
+			err := json.Unmarshal(fileContent, &fileKeys)
+			if err != nil {
+				return c.Status(fiber.StatusInternalServerError).JSON(fiber.Map{"message": "Error parsing api_keys.json"})
+			}
+
+			// Add file keys to options.ApiKeys
+			options.ApiKeys = append(options.ApiKeys, fileKeys...)
+		}
+
+		if len(options.ApiKeys) == 0 {
+			return c.Next()
+		}
+
+		authHeader := c.Get("Authorization")
+		if authHeader == "" {
+			return c.Status(fiber.StatusUnauthorized).JSON(fiber.Map{"message": "Authorization header missing"})
+		}
+		authHeaderParts := strings.Split(authHeader, " ")
+		if len(authHeaderParts) != 2 || authHeaderParts[0] != "Bearer" {
+			return c.Status(fiber.StatusUnauthorized).JSON(fiber.Map{"message": "Invalid Authorization header format"})
+		}
+
+		apiKey := authHeaderParts[1]
+		for _, key := range options.ApiKeys {
+			if apiKey == key {
+				return c.Next()
+			}
+		}
+
+		return c.Status(fiber.StatusUnauthorized).JSON(fiber.Map{"message": "Invalid API key"})
+
+	}
+
+	if options.CORS {
+		var c func(ctx *fiber.Ctx) error
+		if options.CORSAllowOrigins == "" {
+			c = cors.New()
+		} else {
+			c = cors.New(cors.Config{AllowOrigins: options.CORSAllowOrigins})
+		}
+
+		app.Use(c)
+	}
+
+	// LocalAI API endpoints
+	galleryService := localai.NewGalleryService(options.Loader.ModelPath)
+	galleryService.Start(options.Context, cl)
+
+	app.Get("/version", auth, func(c *fiber.Ctx) error {
+		return c.JSON(struct {
+			Version string `json:"version"`
+		}{Version: internal.PrintableVersion()})
+	})
+
+	// Make sure directories exists
+	os.MkdirAll(options.ImageDir, 0755)
+	os.MkdirAll(options.AudioDir, 0755)
+	os.MkdirAll(options.Loader.ModelPath, 0755)
+
+	modelGalleryService := localai.CreateModelGalleryService(options.Galleries, options.Loader.ModelPath, galleryService)
+	app.Post("/models/apply", auth, modelGalleryService.ApplyModelGalleryEndpoint())
+	app.Get("/models/available", auth, modelGalleryService.ListModelFromGalleryEndpoint())
+	app.Get("/models/galleries", auth, modelGalleryService.ListModelGalleriesEndpoint())
+	app.Post("/models/galleries", auth, modelGalleryService.AddModelGalleryEndpoint())
+	app.Delete("/models/galleries", auth, modelGalleryService.RemoveModelGalleryEndpoint())
+	app.Get("/models/jobs/:uuid", auth, modelGalleryService.GetOpStatusEndpoint())
+	app.Get("/models/jobs", auth, modelGalleryService.GetAllStatusEndpoint())
 
 	// openAI compatible API endpoint
-	app.Post("/v1/chat/completions", openAIEndpoint(true, loader, threads, ctxSize, f16, mumutex, mu))
-	app.Post("/chat/completions", openAIEndpoint(true, loader, threads, ctxSize, f16, mumutex, mu))
 
-	app.Post("/v1/completions", openAIEndpoint(false, loader, threads, ctxSize, f16, mumutex, mu))
-	app.Post("/completions", openAIEndpoint(false, loader, threads, ctxSize, f16, mumutex, mu))
+	// chat
+	app.Post("/v1/chat/completions", auth, openai.ChatEndpoint(cl, options))
+	app.Post("/chat/completions", auth, openai.ChatEndpoint(cl, options))
 
-	app.Get("/v1/models", listModels(loader))
-	app.Get("/models", listModels(loader))
+	// edit
+	app.Post("/v1/edits", auth, openai.EditEndpoint(cl, options))
+	app.Post("/edits", auth, openai.EditEndpoint(cl, options))
 
-	// Start the server
-	app.Listen(listenAddr)
-	return nil
+	// completion
+	app.Post("/v1/completions", auth, openai.CompletionEndpoint(cl, options))
+	app.Post("/completions", auth, openai.CompletionEndpoint(cl, options))
+	app.Post("/v1/engines/:model/completions", auth, openai.CompletionEndpoint(cl, options))
+
+	// embeddings
+	app.Post("/v1/embeddings", auth, openai.EmbeddingsEndpoint(cl, options))
+	app.Post("/embeddings", auth, openai.EmbeddingsEndpoint(cl, options))
+	app.Post("/v1/engines/:model/embeddings", auth, openai.EmbeddingsEndpoint(cl, options))
+
+	// audio
+	app.Post("/v1/audio/transcriptions", auth, openai.TranscriptEndpoint(cl, options))
+	app.Post("/tts", auth, localai.TTSEndpoint(cl, options))
+
+	// images
+	app.Post("/v1/images/generations", auth, openai.ImageEndpoint(cl, options))
+
+	if options.ImageDir != "" {
+		app.Static("/generated-images", options.ImageDir)
+	}
+
+	if options.AudioDir != "" {
+		app.Static("/generated-audio", options.AudioDir)
+	}
+
+	ok := func(c *fiber.Ctx) error {
+		return c.SendStatus(200)
+	}
+
+	// Kubernetes health checks
+	app.Get("/healthz", ok)
+	app.Get("/readyz", ok)
+
+	// Experimental Backend Statistics Module
+	backendMonitor := localai.NewBackendMonitor(cl, options) // Split out for now
+	app.Get("/backend/monitor", localai.BackendMonitorEndpoint(backendMonitor))
+	app.Post("/backend/shutdown", localai.BackendShutdownEndpoint(backendMonitor))
+
+	// models
+	app.Get("/v1/models", auth, openai.ListModelsEndpoint(options.Loader, cl))
+	app.Get("/models", auth, openai.ListModelsEndpoint(options.Loader, cl))
+
+	app.Get("/metrics", metrics.MetricsHandler())
+
+	return app, nil
 }

api/api_test.go

@@ -0,0 +1,861 @@
+package api_test
+
+import (
+	"bytes"
+	"context"
+	"embed"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"os"
+	"path/filepath"
+	"runtime"
+
+	. "github.com/go-skynet/LocalAI/api"
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/metrics"
+	"github.com/go-skynet/LocalAI/pkg/downloader"
+	"github.com/go-skynet/LocalAI/pkg/gallery"
+	"github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/gofiber/fiber/v2"
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+	"gopkg.in/yaml.v3"
+
+	openaigo "github.com/otiai10/openaigo"
+	"github.com/sashabaranov/go-openai"
+	"github.com/sashabaranov/go-openai/jsonschema"
+)
+
+type modelApplyRequest struct {
+	ID        string                 `json:"id"`
+	URL       string                 `json:"url"`
+	Name      string                 `json:"name"`
+	Overrides map[string]interface{} `json:"overrides"`
+}
+
+func getModelStatus(url string) (response map[string]interface{}) {
+	// Create the HTTP request
+	resp, err := http.Get(url)
+	if err != nil {
+		fmt.Println("Error creating request:", err)
+		return
+	}
+	defer resp.Body.Close()
+
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		fmt.Println("Error reading response body:", err)
+		return
+	}
+
+	// Unmarshal the response into a map[string]interface{}
+	err = json.Unmarshal(body, &response)
+	if err != nil {
+		fmt.Println("Error unmarshaling JSON response:", err)
+		return
+	}
+	return
+}
+
+func getModels(url string) (response []gallery.GalleryModel) {
+	downloader.GetURI(url, func(url string, i []byte) error {
+		// Unmarshal YAML data into a struct
+		return json.Unmarshal(i, &response)
+	})
+	return
+}
+
+func postModelApplyRequest(url string, request modelApplyRequest) (response map[string]interface{}) {
+
+	//url := "http://localhost:AI/models/apply"
+
+	// Create the request payload
+
+	payload, err := json.Marshal(request)
+	if err != nil {
+		fmt.Println("Error marshaling JSON:", err)
+		return
+	}
+
+	// Create the HTTP request
+	req, err := http.NewRequest("POST", url, bytes.NewBuffer(payload))
+	if err != nil {
+		fmt.Println("Error creating request:", err)
+		return
+	}
+	req.Header.Set("Content-Type", "application/json")
+
+	// Make the request
+	client := &http.Client{}
+	resp, err := client.Do(req)
+	if err != nil {
+		fmt.Println("Error making request:", err)
+		return
+	}
+	defer resp.Body.Close()
+
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		fmt.Println("Error reading response body:", err)
+		return
+	}
+
+	// Unmarshal the response into a map[string]interface{}
+	err = json.Unmarshal(body, &response)
+	if err != nil {
+		fmt.Println("Error unmarshaling JSON response:", err)
+		return
+	}
+	return
+}
+
+//go:embed backend-assets/*
+var backendAssets embed.FS
+
+var _ = Describe("API test", func() {
+
+	var app *fiber.App
+	var modelLoader *model.ModelLoader
+	var client *openai.Client
+	var client2 *openaigo.Client
+	var c context.Context
+	var cancel context.CancelFunc
+	var tmpdir string
+
+	commonOpts := []options.AppOption{
+		options.WithDebug(true),
+		options.WithDisableMessage(true),
+	}
+
+	Context("API with ephemeral models", func() {
+		BeforeEach(func() {
+			var err error
+			tmpdir, err = os.MkdirTemp("", "")
+			Expect(err).ToNot(HaveOccurred())
+
+			modelLoader = model.NewModelLoader(tmpdir)
+			c, cancel = context.WithCancel(context.Background())
+
+			g := []gallery.GalleryModel{
+				{
+					Name: "bert",
+					URL:  "https://raw.githubusercontent.com/go-skynet/model-gallery/main/bert-embeddings.yaml",
+				},
+				{
+					Name:            "bert2",
+					URL:             "https://raw.githubusercontent.com/go-skynet/model-gallery/main/bert-embeddings.yaml",
+					Overrides:       map[string]interface{}{"foo": "bar"},
+					AdditionalFiles: []gallery.File{{Filename: "foo.yaml", URI: "https://raw.githubusercontent.com/go-skynet/model-gallery/main/bert-embeddings.yaml"}},
+				},
+			}
+			out, err := yaml.Marshal(g)
+			Expect(err).ToNot(HaveOccurred())
+			err = os.WriteFile(filepath.Join(tmpdir, "gallery_simple.yaml"), out, 0644)
+			Expect(err).ToNot(HaveOccurred())
+
+			galleries := []gallery.Gallery{
+				{
+					Name: "test",
+					URL:  "file://" + filepath.Join(tmpdir, "gallery_simple.yaml"),
+				},
+			}
+
+			metricsService, err := metrics.SetupMetrics()
+			Expect(err).ToNot(HaveOccurred())
+
+			app, err = App(
+				append(commonOpts,
+					options.WithMetrics(metricsService),
+					options.WithContext(c),
+					options.WithGalleries(galleries),
+					options.WithModelLoader(modelLoader), options.WithBackendAssets(backendAssets), options.WithBackendAssetsOutput(tmpdir))...)
+			Expect(err).ToNot(HaveOccurred())
+			go app.Listen("127.0.0.1:9090")
+
+			defaultConfig := openai.DefaultConfig("")
+			defaultConfig.BaseURL = "http://127.0.0.1:9090/v1"
+
+			client2 = openaigo.NewClient("")
+			client2.BaseURL = defaultConfig.BaseURL
+
+			// Wait for API to be ready
+			client = openai.NewClientWithConfig(defaultConfig)
+			Eventually(func() error {
+				_, err := client.ListModels(context.TODO())
+				return err
+			}, "2m").ShouldNot(HaveOccurred())
+		})
+
+		AfterEach(func() {
+			cancel()
+			app.Shutdown()
+			os.RemoveAll(tmpdir)
+		})
+
+		Context("Applying models", func() {
+			It("applies models from a gallery", func() {
+
+				models := getModels("http://127.0.0.1:9090/models/available")
+				Expect(len(models)).To(Equal(2), fmt.Sprint(models))
+				Expect(models[0].Installed).To(BeFalse(), fmt.Sprint(models))
+				Expect(models[1].Installed).To(BeFalse(), fmt.Sprint(models))
+
+				response := postModelApplyRequest("http://127.0.0.1:9090/models/apply", modelApplyRequest{
+					ID: "test@bert2",
+				})
+
+				Expect(response["uuid"]).ToNot(BeEmpty(), fmt.Sprint(response))
+
+				uuid := response["uuid"].(string)
+				resp := map[string]interface{}{}
+				Eventually(func() bool {
+					response := getModelStatus("http://127.0.0.1:9090/models/jobs/" + uuid)
+					fmt.Println(response)
+					resp = response
+					return response["processed"].(bool)
+				}, "360s", "10s").Should(Equal(true))
+				Expect(resp["message"]).ToNot(ContainSubstring("error"))
+
+				dat, err := os.ReadFile(filepath.Join(tmpdir, "bert2.yaml"))
+				Expect(err).ToNot(HaveOccurred())
+
+				_, err = os.ReadFile(filepath.Join(tmpdir, "foo.yaml"))
+				Expect(err).ToNot(HaveOccurred())
+
+				content := map[string]interface{}{}
+				err = yaml.Unmarshal(dat, &content)
+				Expect(err).ToNot(HaveOccurred())
+				Expect(content["backend"]).To(Equal("bert-embeddings"))
+				Expect(content["foo"]).To(Equal("bar"))
+
+				models = getModels("http://127.0.0.1:9090/models/available")
+				Expect(len(models)).To(Equal(2), fmt.Sprint(models))
+				Expect(models[0].Name).To(Or(Equal("bert"), Equal("bert2")))
+				Expect(models[1].Name).To(Or(Equal("bert"), Equal("bert2")))
+				for _, m := range models {
+					if m.Name == "bert2" {
+						Expect(m.Installed).To(BeTrue())
+					} else {
+						Expect(m.Installed).To(BeFalse())
+					}
+				}
+			})
+			It("overrides models", func() {
+				response := postModelApplyRequest("http://127.0.0.1:9090/models/apply", modelApplyRequest{
+					URL:  "https://raw.githubusercontent.com/go-skynet/model-gallery/main/bert-embeddings.yaml",
+					Name: "bert",
+					Overrides: map[string]interface{}{
+						"backend": "llama",
+					},
+				})
+
+				Expect(response["uuid"]).ToNot(BeEmpty(), fmt.Sprint(response))
+
+				uuid := response["uuid"].(string)
+
+				Eventually(func() bool {
+					response := getModelStatus("http://127.0.0.1:9090/models/jobs/" + uuid)
+					return response["processed"].(bool)
+				}, "360s", "10s").Should(Equal(true))
+
+				dat, err := os.ReadFile(filepath.Join(tmpdir, "bert.yaml"))
+				Expect(err).ToNot(HaveOccurred())
+
+				content := map[string]interface{}{}
+				err = yaml.Unmarshal(dat, &content)
+				Expect(err).ToNot(HaveOccurred())
+				Expect(content["backend"]).To(Equal("llama"))
+			})
+			It("apply models without overrides", func() {
+				response := postModelApplyRequest("http://127.0.0.1:9090/models/apply", modelApplyRequest{
+					URL:       "https://raw.githubusercontent.com/go-skynet/model-gallery/main/bert-embeddings.yaml",
+					Name:      "bert",
+					Overrides: map[string]interface{}{},
+				})
+
+				Expect(response["uuid"]).ToNot(BeEmpty(), fmt.Sprint(response))
+
+				uuid := response["uuid"].(string)
+
+				Eventually(func() bool {
+					response := getModelStatus("http://127.0.0.1:9090/models/jobs/" + uuid)
+					return response["processed"].(bool)
+				}, "360s", "10s").Should(Equal(true))
+
+				dat, err := os.ReadFile(filepath.Join(tmpdir, "bert.yaml"))
+				Expect(err).ToNot(HaveOccurred())
+
+				content := map[string]interface{}{}
+				err = yaml.Unmarshal(dat, &content)
+				Expect(err).ToNot(HaveOccurred())
+				Expect(content["backend"]).To(Equal("bert-embeddings"))
+			})
+
+			It("runs openllama(llama-ggml backend)", Label("llama"), func() {
+				if runtime.GOOS != "linux" {
+					Skip("test supported only on linux")
+				}
+				response := postModelApplyRequest("http://127.0.0.1:9090/models/apply", modelApplyRequest{
+					URL:       "github:go-skynet/model-gallery/openllama_3b.yaml",
+					Name:      "openllama_3b",
+					Overrides: map[string]interface{}{"backend": "llama-ggml", "mmap": true, "f16": true, "context_size": 128},
+				})
+
+				Expect(response["uuid"]).ToNot(BeEmpty(), fmt.Sprint(response))
+
+				uuid := response["uuid"].(string)
+
+				Eventually(func() bool {
+					response := getModelStatus("http://127.0.0.1:9090/models/jobs/" + uuid)
+					return response["processed"].(bool)
+				}, "360s", "10s").Should(Equal(true))
+
+				By("testing completion")
+				resp, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "openllama_3b", Prompt: "Count up to five: one, two, three, four, "})
+				Expect(err).ToNot(HaveOccurred())
+				Expect(len(resp.Choices)).To(Equal(1))
+				Expect(resp.Choices[0].Text).To(ContainSubstring("five"))
+
+				By("testing functions")
+				resp2, err := client.CreateChatCompletion(
+					context.TODO(),
+					openai.ChatCompletionRequest{
+						Model: "openllama_3b",
+						Messages: []openai.ChatCompletionMessage{
+							{
+								Role:    "user",
+								Content: "What is the weather like in San Francisco (celsius)?",
+							},
+						},
+						Functions: []openai.FunctionDefinition{
+							openai.FunctionDefinition{
+								Name:        "get_current_weather",
+								Description: "Get the current weather",
+								Parameters: jsonschema.Definition{
+									Type: jsonschema.Object,
+									Properties: map[string]jsonschema.Definition{
+										"location": {
+											Type:        jsonschema.String,
+											Description: "The city and state, e.g. San Francisco, CA",
+										},
+										"unit": {
+											Type: jsonschema.String,
+											Enum: []string{"celcius", "fahrenheit"},
+										},
+									},
+									Required: []string{"location"},
+								},
+							},
+						},
+					})
+				Expect(err).ToNot(HaveOccurred())
+				Expect(len(resp2.Choices)).To(Equal(1))
+				Expect(resp2.Choices[0].Message.FunctionCall).ToNot(BeNil())
+				Expect(resp2.Choices[0].Message.FunctionCall.Name).To(Equal("get_current_weather"), resp2.Choices[0].Message.FunctionCall.Name)
+
+				var res map[string]string
+				err = json.Unmarshal([]byte(resp2.Choices[0].Message.FunctionCall.Arguments), &res)
+				Expect(err).ToNot(HaveOccurred())
+				Expect(res["location"]).To(Equal("San Francisco, California, United States"), fmt.Sprint(res))
+				Expect(res["unit"]).To(Equal("celcius"), fmt.Sprint(res))
+				Expect(string(resp2.Choices[0].FinishReason)).To(Equal("function_call"), fmt.Sprint(resp2.Choices[0].FinishReason))
+
+			})
+
+			It("runs openllama gguf(llama-cpp)", Label("llama-gguf"), func() {
+				if runtime.GOOS != "linux" {
+					Skip("test supported only on linux")
+				}
+				modelName := "codellama"
+				response := postModelApplyRequest("http://127.0.0.1:9090/models/apply", modelApplyRequest{
+					URL:       "github:go-skynet/model-gallery/codellama-7b-instruct.yaml",
+					Name:      modelName,
+					Overrides: map[string]interface{}{"backend": "llama", "mmap": true, "f16": true, "context_size": 128},
+				})
+
+				Expect(response["uuid"]).ToNot(BeEmpty(), fmt.Sprint(response))
+
+				uuid := response["uuid"].(string)
+
+				Eventually(func() bool {
+					response := getModelStatus("http://127.0.0.1:9090/models/jobs/" + uuid)
+					return response["processed"].(bool)
+				}, "360s", "10s").Should(Equal(true))
+
+				By("testing chat")
+				resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: modelName, Messages: []openai.ChatCompletionMessage{
+					{
+						Role:    "user",
+						Content: "How much is 2+2?",
+					},
+				}})
+				Expect(err).ToNot(HaveOccurred())
+				Expect(len(resp.Choices)).To(Equal(1))
+				Expect(resp.Choices[0].Message.Content).To(Or(ContainSubstring("4"), ContainSubstring("four")))
+
+				By("testing functions")
+				resp2, err := client.CreateChatCompletion(
+					context.TODO(),
+					openai.ChatCompletionRequest{
+						Model: modelName,
+						Messages: []openai.ChatCompletionMessage{
+							{
+								Role:    "user",
+								Content: "What is the weather like in San Francisco (celsius)?",
+							},
+						},
+						Functions: []openai.FunctionDefinition{
+							openai.FunctionDefinition{
+								Name:        "get_current_weather",
+								Description: "Get the current weather",
+								Parameters: jsonschema.Definition{
+									Type: jsonschema.Object,
+									Properties: map[string]jsonschema.Definition{
+										"location": {
+											Type:        jsonschema.String,
+											Description: "The city and state, e.g. San Francisco, CA",
+										},
+										"unit": {
+											Type: jsonschema.String,
+											Enum: []string{"celcius", "fahrenheit"},
+										},
+									},
+									Required: []string{"location"},
+								},
+							},
+						},
+					})
+				Expect(err).ToNot(HaveOccurred())
+				Expect(len(resp2.Choices)).To(Equal(1))
+				Expect(resp2.Choices[0].Message.FunctionCall).ToNot(BeNil())
+				Expect(resp2.Choices[0].Message.FunctionCall.Name).To(Equal("get_current_weather"), resp2.Choices[0].Message.FunctionCall.Name)
+
+				var res map[string]string
+				err = json.Unmarshal([]byte(resp2.Choices[0].Message.FunctionCall.Arguments), &res)
+				Expect(err).ToNot(HaveOccurred())
+				Expect(res["location"]).To(Equal("San Francisco"), fmt.Sprint(res))
+				Expect(res["unit"]).To(Equal("celcius"), fmt.Sprint(res))
+				Expect(string(resp2.Choices[0].FinishReason)).To(Equal("function_call"), fmt.Sprint(resp2.Choices[0].FinishReason))
+			})
+
+			It("runs gpt4all", Label("gpt4all"), func() {
+				if runtime.GOOS != "linux" {
+					Skip("test supported only on linux")
+				}
+
+				response := postModelApplyRequest("http://127.0.0.1:9090/models/apply", modelApplyRequest{
+					URL:  "github:go-skynet/model-gallery/gpt4all-j.yaml",
+					Name: "gpt4all-j",
+				})
+
+				Expect(response["uuid"]).ToNot(BeEmpty(), fmt.Sprint(response))
+
+				uuid := response["uuid"].(string)
+
+				Eventually(func() bool {
+					response := getModelStatus("http://127.0.0.1:9090/models/jobs/" + uuid)
+					return response["processed"].(bool)
+				}, "960s", "10s").Should(Equal(true))
+
+				resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: "gpt4all-j", Messages: []openai.ChatCompletionMessage{openai.ChatCompletionMessage{Role: "user", Content: "How are you?"}}})
+				Expect(err).ToNot(HaveOccurred())
+				Expect(len(resp.Choices)).To(Equal(1))
+				Expect(resp.Choices[0].Message.Content).To(ContainSubstring("well"))
+			})
+
+		})
+	})
+
+	Context("Model gallery", func() {
+		BeforeEach(func() {
+			var err error
+			tmpdir, err = os.MkdirTemp("", "")
+			Expect(err).ToNot(HaveOccurred())
+
+			modelLoader = model.NewModelLoader(tmpdir)
+			c, cancel = context.WithCancel(context.Background())
+
+			galleries := []gallery.Gallery{
+				{
+					Name: "model-gallery",
+					URL:  "https://raw.githubusercontent.com/go-skynet/model-gallery/main/index.yaml",
+				},
+			}
+
+			metricsService, err := metrics.SetupMetrics()
+			Expect(err).ToNot(HaveOccurred())
+
+			app, err = App(
+				append(commonOpts,
+					options.WithContext(c),
+					options.WithMetrics(metricsService),
+					options.WithAudioDir(tmpdir),
+					options.WithImageDir(tmpdir),
+					options.WithGalleries(galleries),
+					options.WithModelLoader(modelLoader),
+					options.WithBackendAssets(backendAssets),
+					options.WithBackendAssetsOutput(tmpdir))...,
+			)
+			Expect(err).ToNot(HaveOccurred())
+			go app.Listen("127.0.0.1:9090")
+
+			defaultConfig := openai.DefaultConfig("")
+			defaultConfig.BaseURL = "http://127.0.0.1:9090/v1"
+
+			client2 = openaigo.NewClient("")
+			client2.BaseURL = defaultConfig.BaseURL
+
+			// Wait for API to be ready
+			client = openai.NewClientWithConfig(defaultConfig)
+			Eventually(func() error {
+				_, err := client.ListModels(context.TODO())
+				return err
+			}, "2m").ShouldNot(HaveOccurred())
+		})
+
+		AfterEach(func() {
+			cancel()
+			app.Shutdown()
+			os.RemoveAll(tmpdir)
+		})
+		It("installs and is capable to run tts", Label("tts"), func() {
+			if runtime.GOOS != "linux" {
+				Skip("test supported only on linux")
+			}
+
+			response := postModelApplyRequest("http://127.0.0.1:9090/models/apply", modelApplyRequest{
+				ID: "model-gallery@voice-en-us-kathleen-low",
+			})
+
+			Expect(response["uuid"]).ToNot(BeEmpty(), fmt.Sprint(response))
+
+			uuid := response["uuid"].(string)
+
+			Eventually(func() bool {
+				response := getModelStatus("http://127.0.0.1:9090/models/jobs/" + uuid)
+				fmt.Println(response)
+				return response["processed"].(bool)
+			}, "360s", "10s").Should(Equal(true))
+
+			// An HTTP Post to the /tts endpoint should return a wav audio file
+			resp, err := http.Post("http://127.0.0.1:9090/tts", "application/json", bytes.NewBuffer([]byte(`{"input": "Hello world", "model": "en-us-kathleen-low.onnx"}`)))
+			Expect(err).ToNot(HaveOccurred(), fmt.Sprint(resp))
+			dat, err := io.ReadAll(resp.Body)
+			Expect(err).ToNot(HaveOccurred(), fmt.Sprint(resp))
+
+			Expect(resp.StatusCode).To(Equal(200), fmt.Sprint(string(dat)))
+			Expect(resp.Header.Get("Content-Type")).To(Equal("audio/x-wav"))
+		})
+		It("installs and is capable to generate images", Label("stablediffusion"), func() {
+			if runtime.GOOS != "linux" {
+				Skip("test supported only on linux")
+			}
+
+			response := postModelApplyRequest("http://127.0.0.1:9090/models/apply", modelApplyRequest{
+				ID: "model-gallery@stablediffusion",
+				Overrides: map[string]interface{}{
+					"parameters": map[string]interface{}{"model": "stablediffusion_assets"},
+				},
+			})
+
+			Expect(response["uuid"]).ToNot(BeEmpty(), fmt.Sprint(response))
+
+			uuid := response["uuid"].(string)
+
+			Eventually(func() bool {
+				response := getModelStatus("http://127.0.0.1:9090/models/jobs/" + uuid)
+				fmt.Println(response)
+				return response["processed"].(bool)
+			}, "360s", "10s").Should(Equal(true))
+
+			resp, err := http.Post(
+				"http://127.0.0.1:9090/v1/images/generations",
+				"application/json",
+				bytes.NewBuffer([]byte(`{
+					 			"prompt": "floating hair, portrait, ((loli)), ((one girl)), cute face, hidden hands, asymmetrical bangs, beautiful detailed eyes, eye shadow, hair ornament, ribbons, bowties, buttons, pleated skirt, (((masterpiece))), ((best quality)), colorful|((part of the head)), ((((mutated hands and fingers)))), deformed, blurry, bad anatomy, disfigured, poorly drawn face, mutation, mutated, extra limb, ugly, poorly drawn hands, missing limb, blurry, floating limbs, disconnected limbs, malformed hands, blur, out of focus, long neck, long body, Octane renderer, lowres, bad anatomy, bad hands, text",
+								"mode": 2,  "seed":9000,
+					 			"size": "256x256", "n":2}`)))
+			// The response should contain an URL
+			Expect(err).ToNot(HaveOccurred(), fmt.Sprint(resp))
+			dat, err := io.ReadAll(resp.Body)
+			Expect(err).ToNot(HaveOccurred(), string(dat))
+			Expect(string(dat)).To(ContainSubstring("http://127.0.0.1:9090/"), string(dat))
+			Expect(string(dat)).To(ContainSubstring(".png"), string(dat))
+
+		})
+	})
+
+	Context("API query", func() {
+		BeforeEach(func() {
+			modelLoader = model.NewModelLoader(os.Getenv("MODELS_PATH"))
+			c, cancel = context.WithCancel(context.Background())
+
+			metricsService, err := metrics.SetupMetrics()
+			Expect(err).ToNot(HaveOccurred())
+
+			app, err = App(
+				append(commonOpts,
+					options.WithExternalBackend("huggingface", os.Getenv("HUGGINGFACE_GRPC")),
+					options.WithContext(c),
+					options.WithModelLoader(modelLoader),
+					options.WithMetrics(metricsService),
+				)...)
+			Expect(err).ToNot(HaveOccurred())
+			go app.Listen("127.0.0.1:9090")
+
+			defaultConfig := openai.DefaultConfig("")
+			defaultConfig.BaseURL = "http://127.0.0.1:9090/v1"
+
+			client2 = openaigo.NewClient("")
+			client2.BaseURL = defaultConfig.BaseURL
+
+			// Wait for API to be ready
+			client = openai.NewClientWithConfig(defaultConfig)
+			Eventually(func() error {
+				_, err := client.ListModels(context.TODO())
+				return err
+			}, "2m").ShouldNot(HaveOccurred())
+		})
+		AfterEach(func() {
+			cancel()
+			app.Shutdown()
+		})
+		It("returns the models list", func() {
+			models, err := client.ListModels(context.TODO())
+			Expect(err).ToNot(HaveOccurred())
+			Expect(len(models.Models)).To(Equal(6)) // If "config.yaml" should be included, this should be 8?
+		})
+		It("can generate completions", func() {
+			resp, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "testmodel", Prompt: "abcdedfghikl"})
+			Expect(err).ToNot(HaveOccurred())
+			Expect(len(resp.Choices)).To(Equal(1))
+			Expect(resp.Choices[0].Text).ToNot(BeEmpty())
+		})
+
+		It("can generate chat completions ", func() {
+			resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: "testmodel", Messages: []openai.ChatCompletionMessage{openai.ChatCompletionMessage{Role: "user", Content: "abcdedfghikl"}}})
+			Expect(err).ToNot(HaveOccurred())
+			Expect(len(resp.Choices)).To(Equal(1))
+			Expect(resp.Choices[0].Message.Content).ToNot(BeEmpty())
+		})
+
+		It("can generate completions from model configs", func() {
+			resp, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "gpt4all", Prompt: "abcdedfghikl"})
+			Expect(err).ToNot(HaveOccurred())
+			Expect(len(resp.Choices)).To(Equal(1))
+			Expect(resp.Choices[0].Text).ToNot(BeEmpty())
+		})
+
+		It("can generate chat completions from model configs", func() {
+			resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: "gpt4all-2", Messages: []openai.ChatCompletionMessage{openai.ChatCompletionMessage{Role: "user", Content: "abcdedfghikl"}}})
+			Expect(err).ToNot(HaveOccurred())
+			Expect(len(resp.Choices)).To(Equal(1))
+			Expect(resp.Choices[0].Message.Content).ToNot(BeEmpty())
+		})
+
+		It("returns errors", func() {
+			backends := len(model.AutoLoadBackends) + 1 // +1 for huggingface
+			_, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "foomodel", Prompt: "abcdedfghikl"})
+			Expect(err).To(HaveOccurred())
+			Expect(err.Error()).To(ContainSubstring(fmt.Sprintf("error, status code: 500, message: could not load model - all backends returned error: %d errors occurred:", backends)))
+		})
+		It("transcribes audio", func() {
+			if runtime.GOOS != "linux" {
+				Skip("test supported only on linux")
+			}
+			resp, err := client.CreateTranscription(
+				context.Background(),
+				openai.AudioRequest{
+					Model:    openai.Whisper1,
+					FilePath: filepath.Join(os.Getenv("TEST_DIR"), "audio.wav"),
+				},
+			)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(resp.Text).To(ContainSubstring("This is the Micro Machine Man presenting"))
+		})
+
+		It("calculate embeddings", func() {
+			if runtime.GOOS != "linux" {
+				Skip("test supported only on linux")
+			}
+			resp, err := client.CreateEmbeddings(
+				context.Background(),
+				openai.EmbeddingRequest{
+					Model: openai.AdaEmbeddingV2,
+					Input: []string{"sun", "cat"},
+				},
+			)
+			Expect(err).ToNot(HaveOccurred(), err)
+			Expect(len(resp.Data[0].Embedding)).To(BeNumerically("==", 384))
+			Expect(len(resp.Data[1].Embedding)).To(BeNumerically("==", 384))
+
+			sunEmbedding := resp.Data[0].Embedding
+			resp2, err := client.CreateEmbeddings(
+				context.Background(),
+				openai.EmbeddingRequest{
+					Model: openai.AdaEmbeddingV2,
+					Input: []string{"sun"},
+				},
+			)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(resp2.Data[0].Embedding).To(Equal(sunEmbedding))
+		})
+
+		Context("External gRPC calls", func() {
+			It("calculate embeddings with sentencetransformers", func() {
+				if runtime.GOOS != "linux" {
+					Skip("test supported only on linux")
+				}
+				resp, err := client.CreateEmbeddings(
+					context.Background(),
+					openai.EmbeddingRequest{
+						Model: openai.AdaCodeSearchCode,
+						Input: []string{"sun", "cat"},
+					},
+				)
+				Expect(err).ToNot(HaveOccurred())
+				Expect(len(resp.Data[0].Embedding)).To(BeNumerically("==", 384))
+				Expect(len(resp.Data[1].Embedding)).To(BeNumerically("==", 384))
+
+				sunEmbedding := resp.Data[0].Embedding
+				resp2, err := client.CreateEmbeddings(
+					context.Background(),
+					openai.EmbeddingRequest{
+						Model: openai.AdaCodeSearchCode,
+						Input: []string{"sun"},
+					},
+				)
+				Expect(err).ToNot(HaveOccurred())
+				Expect(resp2.Data[0].Embedding).To(Equal(sunEmbedding))
+				Expect(resp2.Data[0].Embedding).ToNot(Equal(resp.Data[1].Embedding))
+			})
+		})
+
+		Context("backends", func() {
+			It("runs rwkv completion", func() {
+				if runtime.GOOS != "linux" {
+					Skip("test supported only on linux")
+				}
+				resp, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "rwkv_test", Prompt: "Count up to five: one, two, three, four,"})
+				Expect(err).ToNot(HaveOccurred())
+				Expect(len(resp.Choices) > 0).To(BeTrue())
+				Expect(resp.Choices[0].Text).To(ContainSubstring("five"))
+
+				stream, err := client.CreateCompletionStream(context.TODO(), openai.CompletionRequest{
+					Model: "rwkv_test", Prompt: "Count up to five: one, two, three, four,", Stream: true,
+				})
+				Expect(err).ToNot(HaveOccurred())
+				defer stream.Close()
+
+				tokens := 0
+				text := ""
+				for {
+					response, err := stream.Recv()
+					if errors.Is(err, io.EOF) {
+						break
+					}
+
+					Expect(err).ToNot(HaveOccurred())
+					text += response.Choices[0].Text
+					tokens++
+				}
+				Expect(text).ToNot(BeEmpty())
+				Expect(text).To(ContainSubstring("five"))
+				Expect(tokens).ToNot(Or(Equal(1), Equal(0)))
+			})
+			It("runs rwkv chat completion", func() {
+				if runtime.GOOS != "linux" {
+					Skip("test supported only on linux")
+				}
+				resp, err := client.CreateChatCompletion(context.TODO(),
+					openai.ChatCompletionRequest{Model: "rwkv_test", Messages: []openai.ChatCompletionMessage{{Content: "Can you count up to five?", Role: "user"}}})
+				Expect(err).ToNot(HaveOccurred())
+				Expect(len(resp.Choices) > 0).To(BeTrue())
+				Expect(resp.Choices[0].Message.Content).To(Or(ContainSubstring("Sure"), ContainSubstring("five")))
+
+				stream, err := client.CreateChatCompletionStream(context.TODO(), openai.ChatCompletionRequest{Model: "rwkv_test", Messages: []openai.ChatCompletionMessage{{Content: "Can you count up to five?", Role: "user"}}})
+				Expect(err).ToNot(HaveOccurred())
+				defer stream.Close()
+
+				tokens := 0
+				text := ""
+				for {
+					response, err := stream.Recv()
+					if errors.Is(err, io.EOF) {
+						break
+					}
+
+					Expect(err).ToNot(HaveOccurred())
+					text += response.Choices[0].Delta.Content
+					tokens++
+				}
+				Expect(text).ToNot(BeEmpty())
+				Expect(text).To(Or(ContainSubstring("Sure"), ContainSubstring("five")))
+
+				Expect(tokens).ToNot(Or(Equal(1), Equal(0)))
+			})
+		})
+	})
+
+	Context("Config file", func() {
+		BeforeEach(func() {
+			modelLoader = model.NewModelLoader(os.Getenv("MODELS_PATH"))
+			c, cancel = context.WithCancel(context.Background())
+
+			metricsService, err := metrics.SetupMetrics()
+			Expect(err).ToNot(HaveOccurred())
+
+			app, err = App(
+				append(commonOpts,
+					options.WithContext(c),
+					options.WithMetrics(metricsService),
+					options.WithModelLoader(modelLoader),
+					options.WithConfigFile(os.Getenv("CONFIG_FILE")))...,
+			)
+			Expect(err).ToNot(HaveOccurred())
+			go app.Listen("127.0.0.1:9090")
+
+			defaultConfig := openai.DefaultConfig("")
+			defaultConfig.BaseURL = "http://127.0.0.1:9090/v1"
+			client2 = openaigo.NewClient("")
+			client2.BaseURL = defaultConfig.BaseURL
+			// Wait for API to be ready
+			client = openai.NewClientWithConfig(defaultConfig)
+			Eventually(func() error {
+				_, err := client.ListModels(context.TODO())
+				return err
+			}, "2m").ShouldNot(HaveOccurred())
+		})
+		AfterEach(func() {
+			cancel()
+			app.Shutdown()
+		})
+		It("can generate chat completions from config file (list1)", func() {
+			resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: "list1", Messages: []openai.ChatCompletionMessage{{Role: "user", Content: "abcdedfghikl"}}})
+			Expect(err).ToNot(HaveOccurred())
+			Expect(len(resp.Choices)).To(Equal(1))
+			Expect(resp.Choices[0].Message.Content).ToNot(BeEmpty())
+		})
+		It("can generate chat completions from config file (list2)", func() {
+			resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: "list2", Messages: []openai.ChatCompletionMessage{{Role: "user", Content: "abcdedfghikl"}}})
+			Expect(err).ToNot(HaveOccurred())
+			Expect(len(resp.Choices)).To(Equal(1))
+			Expect(resp.Choices[0].Message.Content).ToNot(BeEmpty())
+		})
+		It("can generate edit completions from config file", func() {
+			request := openaigo.EditCreateRequestBody{
+				Model:       "list2",
+				Instruction: "foo",
+				Input:       "bar",
+			}
+			resp, err := client2.CreateEdit(context.Background(), request)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(len(resp.Choices)).To(Equal(1))
+			Expect(resp.Choices[0].Text).ToNot(BeEmpty())
+		})
+
+	})
+})

api/apt_suite_test.go

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

api/backend/embeddings.go

@@ -0,0 +1,92 @@
+package backend
+
+import (
+	"fmt"
+
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/pkg/grpc"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+)
+
+func ModelEmbedding(s string, tokens []int, loader *model.ModelLoader, c config.Config, o *options.Option) (func() ([]float32, error), error) {
+	if !c.Embeddings {
+		return nil, fmt.Errorf("endpoint disabled for this model by API configuration")
+	}
+
+	modelFile := c.Model
+
+	grpcOpts := gRPCModelOpts(c)
+
+	var inferenceModel interface{}
+	var err error
+
+	opts := modelOpts(c, o, []model.Option{
+		model.WithLoadGRPCLoadModelOpts(grpcOpts),
+		model.WithThreads(uint32(c.Threads)),
+		model.WithAssetDir(o.AssetsDestination),
+		model.WithModel(modelFile),
+		model.WithContext(o.Context),
+	})
+
+	if c.Backend == "" {
+		inferenceModel, err = loader.GreedyLoader(opts...)
+	} else {
+		opts = append(opts, model.WithBackendString(c.Backend))
+		inferenceModel, err = loader.BackendLoader(opts...)
+	}
+	if err != nil {
+		return nil, err
+	}
+
+	var fn func() ([]float32, error)
+	switch model := inferenceModel.(type) {
+	case grpc.Backend:
+		fn = func() ([]float32, error) {
+			predictOptions := gRPCPredictOpts(c, loader.ModelPath)
+			if len(tokens) > 0 {
+				embeds := []int32{}
+
+				for _, t := range tokens {
+					embeds = append(embeds, int32(t))
+				}
+				predictOptions.EmbeddingTokens = embeds
+
+				res, err := model.Embeddings(o.Context, predictOptions)
+				if err != nil {
+					return nil, err
+				}
+
+				return res.Embeddings, nil
+			}
+			predictOptions.Embeddings = s
+
+			res, err := model.Embeddings(o.Context, predictOptions)
+			if err != nil {
+				return nil, err
+			}
+
+			return res.Embeddings, nil
+		}
+	default:
+		fn = func() ([]float32, error) {
+			return nil, fmt.Errorf("embeddings not supported by the backend")
+		}
+	}
+
+	return func() ([]float32, error) {
+		embeds, err := fn()
+		if err != nil {
+			return embeds, err
+		}
+		// Remove trailing 0s
+		for i := len(embeds) - 1; i >= 0; i-- {
+			if embeds[i] == 0.0 {
+				embeds = embeds[:i]
+			} else {
+				break
+			}
+		}
+		return embeds, nil
+	}, nil
+}

api/backend/image.go

@@ -0,0 +1,61 @@
+package backend
+
+import (
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/pkg/grpc/proto"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+)
+
+func ImageGeneration(height, width, mode, step, seed int, positive_prompt, negative_prompt, src, dst string, loader *model.ModelLoader, c config.Config, o *options.Option) (func() error, error) {
+
+	opts := modelOpts(c, o, []model.Option{
+		model.WithBackendString(c.Backend),
+		model.WithAssetDir(o.AssetsDestination),
+		model.WithThreads(uint32(c.Threads)),
+		model.WithContext(o.Context),
+		model.WithModel(c.Model),
+		model.WithLoadGRPCLoadModelOpts(&proto.ModelOptions{
+			CUDA:          c.CUDA || c.Diffusers.CUDA,
+			SchedulerType: c.Diffusers.SchedulerType,
+			PipelineType:  c.Diffusers.PipelineType,
+			CFGScale:      c.Diffusers.CFGScale,
+			LoraAdapter:   c.LoraAdapter,
+			LoraScale:     c.LoraScale,
+			LoraBase:      c.LoraBase,
+			IMG2IMG:       c.Diffusers.IMG2IMG,
+			CLIPModel:     c.Diffusers.ClipModel,
+			CLIPSubfolder: c.Diffusers.ClipSubFolder,
+			CLIPSkip:      int32(c.Diffusers.ClipSkip),
+			ControlNet:    c.Diffusers.ControlNet,
+		}),
+	})
+
+	inferenceModel, err := loader.BackendLoader(
+		opts...,
+	)
+	if err != nil {
+		return nil, err
+	}
+
+	fn := func() error {
+		_, err := inferenceModel.GenerateImage(
+			o.Context,
+			&proto.GenerateImageRequest{
+				Height:           int32(height),
+				Width:            int32(width),
+				Mode:             int32(mode),
+				Step:             int32(step),
+				Seed:             int32(seed),
+				CLIPSkip:         int32(c.Diffusers.ClipSkip),
+				PositivePrompt:   positive_prompt,
+				NegativePrompt:   negative_prompt,
+				Dst:              dst,
+				Src:              src,
+				EnableParameters: c.Diffusers.EnableParameters,
+			})
+		return err
+	}
+
+	return fn, nil
+}

api/backend/llm.go

@@ -0,0 +1,167 @@
+package backend
+
+import (
+	"context"
+	"os"
+	"regexp"
+	"strings"
+	"sync"
+	"unicode/utf8"
+
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/pkg/gallery"
+	"github.com/go-skynet/LocalAI/pkg/grpc"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/go-skynet/LocalAI/pkg/utils"
+)
+
+type LLMResponse struct {
+	Response string // should this be []byte?
+	Usage    TokenUsage
+}
+
+type TokenUsage struct {
+	Prompt     int
+	Completion int
+}
+
+func ModelInference(ctx context.Context, s string, images []string, loader *model.ModelLoader, c config.Config, o *options.Option, tokenCallback func(string, TokenUsage) bool) (func() (LLMResponse, error), error) {
+	modelFile := c.Model
+
+	grpcOpts := gRPCModelOpts(c)
+
+	var inferenceModel grpc.Backend
+	var err error
+
+	opts := modelOpts(c, o, []model.Option{
+		model.WithLoadGRPCLoadModelOpts(grpcOpts),
+		model.WithThreads(uint32(c.Threads)), // some models uses this to allocate threads during startup
+		model.WithAssetDir(o.AssetsDestination),
+		model.WithModel(modelFile),
+		model.WithContext(o.Context),
+	})
+
+	if c.Backend != "" {
+		opts = append(opts, model.WithBackendString(c.Backend))
+	}
+
+	// Check if the modelFile exists, if it doesn't try to load it from the gallery
+	if o.AutoloadGalleries { // experimental
+		if _, err := os.Stat(modelFile); os.IsNotExist(err) {
+			utils.ResetDownloadTimers()
+			// if we failed to load the model, we try to download it
+			err := gallery.InstallModelFromGalleryByName(o.Galleries, modelFile, loader.ModelPath, gallery.GalleryModel{}, utils.DisplayDownloadFunction)
+			if err != nil {
+				return nil, err
+			}
+		}
+	}
+
+	if c.Backend == "" {
+		inferenceModel, err = loader.GreedyLoader(opts...)
+	} else {
+		inferenceModel, err = loader.BackendLoader(opts...)
+	}
+
+	if err != nil {
+		return nil, err
+	}
+
+	// in GRPC, the backend is supposed to answer to 1 single token if stream is not supported
+	fn := func() (LLMResponse, error) {
+		opts := gRPCPredictOpts(c, loader.ModelPath)
+		opts.Prompt = s
+		opts.Images = images
+
+		tokenUsage := TokenUsage{}
+
+		// check the per-model feature flag for usage, since tokenCallback may have a cost.
+		// Defaults to off as for now it is still experimental
+		if c.FeatureFlag.Enabled("usage") {
+			userTokenCallback := tokenCallback
+			if userTokenCallback == nil {
+				userTokenCallback = func(token string, usage TokenUsage) bool {
+					return true
+				}
+			}
+
+			promptInfo, pErr := inferenceModel.TokenizeString(ctx, opts)
+			if pErr == nil && promptInfo.Length > 0 {
+				tokenUsage.Prompt = int(promptInfo.Length)
+			}
+
+			tokenCallback = func(token string, usage TokenUsage) bool {
+				tokenUsage.Completion++
+				return userTokenCallback(token, tokenUsage)
+			}
+		}
+
+		if tokenCallback != nil {
+			ss := ""
+
+			var partialRune []byte
+			err := inferenceModel.PredictStream(ctx, opts, func(chars []byte) {
+				partialRune = append(partialRune, chars...)
+
+				for len(partialRune) > 0 {
+					r, size := utf8.DecodeRune(partialRune)
+					if r == utf8.RuneError {
+						// incomplete rune, wait for more bytes
+						break
+					}
+
+					tokenCallback(string(r), tokenUsage)
+					ss += string(r)
+
+					partialRune = partialRune[size:]
+				}
+			})
+			return LLMResponse{
+				Response: ss,
+				Usage:    tokenUsage,
+			}, err
+		} else {
+			// TODO: Is the chicken bit the only way to get here? is that acceptable?
+			reply, err := inferenceModel.Predict(ctx, opts)
+			if err != nil {
+				return LLMResponse{}, err
+			}
+			return LLMResponse{
+				Response: string(reply.Message),
+				Usage:    tokenUsage,
+			}, err
+		}
+	}
+
+	return fn, nil
+}
+
+var cutstrings map[string]*regexp.Regexp = make(map[string]*regexp.Regexp)
+var mu sync.Mutex = sync.Mutex{}
+
+func Finetune(config config.Config, input, prediction string) string {
+	if config.Echo {
+		prediction = input + prediction
+	}
+
+	for _, c := range config.Cutstrings {
+		mu.Lock()
+		reg, ok := cutstrings[c]
+		if !ok {
+			cutstrings[c] = regexp.MustCompile(c)
+			reg = cutstrings[c]
+		}
+		mu.Unlock()
+		prediction = reg.ReplaceAllString(prediction, "")
+	}
+
+	for _, c := range config.TrimSpace {
+		prediction = strings.TrimSpace(strings.TrimPrefix(prediction, c))
+	}
+
+	for _, c := range config.TrimSuffix {
+		prediction = strings.TrimSpace(strings.TrimSuffix(prediction, c))
+	}
+	return prediction
+}

api/backend/options.go

@@ -0,0 +1,129 @@
+package backend
+
+import (
+	"os"
+	"path/filepath"
+
+	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+)
+
+func modelOpts(c config.Config, o *options.Option, opts []model.Option) []model.Option {
+	if o.SingleBackend {
+		opts = append(opts, model.WithSingleActiveBackend())
+	}
+
+	if o.ParallelBackendRequests {
+		opts = append(opts, model.EnableParallelRequests)
+	}
+
+	if c.GRPC.Attempts != 0 {
+		opts = append(opts, model.WithGRPCAttempts(c.GRPC.Attempts))
+	}
+
+	if c.GRPC.AttemptsSleepTime != 0 {
+		opts = append(opts, model.WithGRPCAttemptsDelay(c.GRPC.AttemptsSleepTime))
+	}
+
+	for k, v := range o.ExternalGRPCBackends {
+		opts = append(opts, model.WithExternalBackend(k, v))
+	}
+
+	return opts
+}
+
+func gRPCModelOpts(c config.Config) *pb.ModelOptions {
+	b := 512
+	if c.Batch != 0 {
+		b = c.Batch
+	}
+
+	return &pb.ModelOptions{
+		ContextSize:    int32(c.ContextSize),
+		Seed:           int32(c.Seed),
+		NBatch:         int32(b),
+		NoMulMatQ:      c.NoMulMatQ,
+		CUDA:           c.CUDA, // diffusers, transformers
+		DraftModel:     c.DraftModel,
+		AudioPath:      c.VallE.AudioPath,
+		Quantization:   c.Quantization,
+		MMProj:         c.MMProj,
+		YarnExtFactor:  c.YarnExtFactor,
+		YarnAttnFactor: c.YarnAttnFactor,
+		YarnBetaFast:   c.YarnBetaFast,
+		YarnBetaSlow:   c.YarnBetaSlow,
+		LoraAdapter:    c.LoraAdapter,
+		LoraBase:       c.LoraBase,
+		LoraScale:      c.LoraScale,
+		NGQA:           c.NGQA,
+		RMSNormEps:     c.RMSNormEps,
+		F16Memory:      c.F16,
+		MLock:          c.MMlock,
+		RopeFreqBase:   c.RopeFreqBase,
+		RopeScaling:    c.RopeScaling,
+		Type:           c.ModelType,
+		RopeFreqScale:  c.RopeFreqScale,
+		NUMA:           c.NUMA,
+		Embeddings:     c.Embeddings,
+		LowVRAM:        c.LowVRAM,
+		NGPULayers:     int32(c.NGPULayers),
+		MMap:           c.MMap,
+		MainGPU:        c.MainGPU,
+		Threads:        int32(c.Threads),
+		TensorSplit:    c.TensorSplit,
+		// AutoGPTQ
+		ModelBaseName:    c.AutoGPTQ.ModelBaseName,
+		Device:           c.AutoGPTQ.Device,
+		UseTriton:        c.AutoGPTQ.Triton,
+		UseFastTokenizer: c.AutoGPTQ.UseFastTokenizer,
+		// RWKV
+		Tokenizer: c.Tokenizer,
+	}
+}
+
+func gRPCPredictOpts(c config.Config, modelPath string) *pb.PredictOptions {
+	promptCachePath := ""
+	if c.PromptCachePath != "" {
+		p := filepath.Join(modelPath, c.PromptCachePath)
+		os.MkdirAll(filepath.Dir(p), 0755)
+		promptCachePath = p
+	}
+	return &pb.PredictOptions{
+		Temperature:         float32(c.Temperature),
+		TopP:                float32(c.TopP),
+		NDraft:              c.NDraft,
+		TopK:                int32(c.TopK),
+		Tokens:              int32(c.Maxtokens),
+		Threads:             int32(c.Threads),
+		PromptCacheAll:      c.PromptCacheAll,
+		PromptCacheRO:       c.PromptCacheRO,
+		PromptCachePath:     promptCachePath,
+		F16KV:               c.F16,
+		DebugMode:           c.Debug,
+		Grammar:             c.Grammar,
+		NegativePromptScale: c.NegativePromptScale,
+		RopeFreqBase:        c.RopeFreqBase,
+		RopeFreqScale:       c.RopeFreqScale,
+		NegativePrompt:      c.NegativePrompt,
+		Mirostat:            int32(c.LLMConfig.Mirostat),
+		MirostatETA:         float32(c.LLMConfig.MirostatETA),
+		MirostatTAU:         float32(c.LLMConfig.MirostatTAU),
+		Debug:               c.Debug,
+		StopPrompts:         c.StopWords,
+		Repeat:              int32(c.RepeatPenalty),
+		NKeep:               int32(c.Keep),
+		Batch:               int32(c.Batch),
+		IgnoreEOS:           c.IgnoreEOS,
+		Seed:                int32(c.Seed),
+		FrequencyPenalty:    float32(c.FrequencyPenalty),
+		MLock:               c.MMlock,
+		MMap:                c.MMap,
+		MainGPU:             c.MainGPU,
+		TensorSplit:         c.TensorSplit,
+		TailFreeSamplingZ:   float32(c.TFZ),
+		TypicalP:            float32(c.TypicalP),
+	}
+}

api/backend/transcript.go

@@ -0,0 +1,39 @@
+package backend
+
+import (
+	"context"
+	"fmt"
+
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/schema"
+
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/pkg/grpc/proto"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+)
+
+func ModelTranscription(audio, language string, loader *model.ModelLoader, c config.Config, o *options.Option) (*schema.Result, error) {
+
+	opts := modelOpts(c, o, []model.Option{
+		model.WithBackendString(model.WhisperBackend),
+		model.WithModel(c.Model),
+		model.WithContext(o.Context),
+		model.WithThreads(uint32(c.Threads)),
+		model.WithAssetDir(o.AssetsDestination),
+	})
+
+	whisperModel, err := o.Loader.BackendLoader(opts...)
+	if err != nil {
+		return nil, err
+	}
+
+	if whisperModel == nil {
+		return nil, fmt.Errorf("could not load whisper model")
+	}
+
+	return whisperModel.AudioTranscription(context.Background(), &proto.TranscriptRequest{
+		Dst:      audio,
+		Language: language,
+		Threads:  uint32(c.Threads),
+	})
+}

api/backend/tts.go

@@ -0,0 +1,79 @@
+package backend
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"path/filepath"
+
+	api_config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/pkg/grpc/proto"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/go-skynet/LocalAI/pkg/utils"
+)
+
+func generateUniqueFileName(dir, baseName, ext string) string {
+	counter := 1
+	fileName := baseName + ext
+
+	for {
+		filePath := filepath.Join(dir, fileName)
+		_, err := os.Stat(filePath)
+		if os.IsNotExist(err) {
+			return fileName
+		}
+
+		counter++
+		fileName = fmt.Sprintf("%s_%d%s", baseName, counter, ext)
+	}
+}
+
+func ModelTTS(backend, text, modelFile string, loader *model.ModelLoader, o *options.Option) (string, *proto.Result, error) {
+	bb := backend
+	if bb == "" {
+		bb = model.PiperBackend
+	}
+	opts := modelOpts(api_config.Config{}, o, []model.Option{
+		model.WithBackendString(bb),
+		model.WithModel(modelFile),
+		model.WithContext(o.Context),
+		model.WithAssetDir(o.AssetsDestination),
+	})
+	piperModel, err := o.Loader.BackendLoader(opts...)
+	if err != nil {
+		return "", nil, err
+	}
+
+	if piperModel == nil {
+		return "", nil, fmt.Errorf("could not load piper model")
+	}
+
+	if err := os.MkdirAll(o.AudioDir, 0755); err != nil {
+		return "", nil, fmt.Errorf("failed creating audio directory: %s", err)
+	}
+
+	fileName := generateUniqueFileName(o.AudioDir, "piper", ".wav")
+	filePath := filepath.Join(o.AudioDir, fileName)
+
+	// If the model file is not empty, we pass it joined with the model path
+	modelPath := ""
+	if modelFile != "" {
+		if bb != model.TransformersMusicGen {
+			modelPath = filepath.Join(o.Loader.ModelPath, modelFile)
+			if err := utils.VerifyPath(modelPath, o.Loader.ModelPath); err != nil {
+				return "", nil, err
+			}
+		} else {
+			modelPath = modelFile
+		}
+	}
+
+	res, err := piperModel.TTS(context.Background(), &proto.TTSRequest{
+		Text:  text,
+		Model: modelPath,
+		Dst:   filePath,
+	})
+
+	return filePath, res, err
+}

api/config/config.go

@@ -0,0 +1,374 @@
+package api_config
+
+import (
+	"errors"
+	"fmt"
+	"io/fs"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+
+	"github.com/go-skynet/LocalAI/pkg/downloader"
+	"github.com/go-skynet/LocalAI/pkg/utils"
+	"github.com/rs/zerolog/log"
+	"gopkg.in/yaml.v3"
+)
+
+type Config struct {
+	PredictionOptions `yaml:"parameters"`
+	Name              string `yaml:"name"`
+
+	F16            bool              `yaml:"f16"`
+	Threads        int               `yaml:"threads"`
+	Debug          bool              `yaml:"debug"`
+	Roles          map[string]string `yaml:"roles"`
+	Embeddings     bool              `yaml:"embeddings"`
+	Backend        string            `yaml:"backend"`
+	TemplateConfig TemplateConfig    `yaml:"template"`
+
+	PromptStrings, InputStrings                []string `yaml:"-"`
+	InputToken                                 [][]int  `yaml:"-"`
+	functionCallString, functionCallNameString string   `yaml:"-"`
+
+	FunctionsConfig Functions `yaml:"function"`
+
+	FeatureFlag FeatureFlag `yaml:"feature_flags"` // Feature Flag registry. We move fast, and features may break on a per model/backend basis. Registry for (usually temporary) flags that indicate aborting something early.
+	// LLM configs (GPT4ALL, Llama.cpp, ...)
+	LLMConfig `yaml:",inline"`
+
+	// AutoGPTQ specifics
+	AutoGPTQ AutoGPTQ `yaml:"autogptq"`
+
+	// Diffusers
+	Diffusers Diffusers `yaml:"diffusers"`
+	Step      int       `yaml:"step"`
+
+	// GRPC Options
+	GRPC GRPC `yaml:"grpc"`
+
+	// Vall-e-x
+	VallE VallE `yaml:"vall-e"`
+
+	// CUDA
+	// Explicitly enable CUDA or not (some backends might need it)
+	CUDA bool `yaml:"cuda"`
+
+	DownloadFiles []File `yaml:"download_files"`
+
+	Description string `yaml:"description"`
+	Usage       string `yaml:"usage"`
+}
+
+type File struct {
+	Filename string `yaml:"filename" json:"filename"`
+	SHA256   string `yaml:"sha256" json:"sha256"`
+	URI      string `yaml:"uri" json:"uri"`
+}
+
+type VallE struct {
+	AudioPath string `yaml:"audio_path"`
+}
+
+type FeatureFlag map[string]*bool
+
+func (ff FeatureFlag) Enabled(s string) bool {
+	v, exist := ff[s]
+	return exist && v != nil && *v
+}
+
+type GRPC struct {
+	Attempts          int `yaml:"attempts"`
+	AttemptsSleepTime int `yaml:"attempts_sleep_time"`
+}
+
+type Diffusers struct {
+	CUDA             bool    `yaml:"cuda"`
+	PipelineType     string  `yaml:"pipeline_type"`
+	SchedulerType    string  `yaml:"scheduler_type"`
+	EnableParameters string  `yaml:"enable_parameters"` // A list of comma separated parameters to specify
+	CFGScale         float32 `yaml:"cfg_scale"`         // Classifier-Free Guidance Scale
+	IMG2IMG          bool    `yaml:"img2img"`           // Image to Image Diffuser
+	ClipSkip         int     `yaml:"clip_skip"`         // Skip every N frames
+	ClipModel        string  `yaml:"clip_model"`        // Clip model to use
+	ClipSubFolder    string  `yaml:"clip_subfolder"`    // Subfolder to use for clip model
+	ControlNet       string  `yaml:"control_net"`
+}
+
+type LLMConfig struct {
+	SystemPrompt    string   `yaml:"system_prompt"`
+	TensorSplit     string   `yaml:"tensor_split"`
+	MainGPU         string   `yaml:"main_gpu"`
+	RMSNormEps      float32  `yaml:"rms_norm_eps"`
+	NGQA            int32    `yaml:"ngqa"`
+	PromptCachePath string   `yaml:"prompt_cache_path"`
+	PromptCacheAll  bool     `yaml:"prompt_cache_all"`
+	PromptCacheRO   bool     `yaml:"prompt_cache_ro"`
+	MirostatETA     float64  `yaml:"mirostat_eta"`
+	MirostatTAU     float64  `yaml:"mirostat_tau"`
+	Mirostat        int      `yaml:"mirostat"`
+	NGPULayers      int      `yaml:"gpu_layers"`
+	MMap            bool     `yaml:"mmap"`
+	MMlock          bool     `yaml:"mmlock"`
+	LowVRAM         bool     `yaml:"low_vram"`
+	Grammar         string   `yaml:"grammar"`
+	StopWords       []string `yaml:"stopwords"`
+	Cutstrings      []string `yaml:"cutstrings"`
+	TrimSpace       []string `yaml:"trimspace"`
+	TrimSuffix      []string `yaml:"trimsuffix"`
+
+	ContextSize  int     `yaml:"context_size"`
+	NUMA         bool    `yaml:"numa"`
+	LoraAdapter  string  `yaml:"lora_adapter"`
+	LoraBase     string  `yaml:"lora_base"`
+	LoraScale    float32 `yaml:"lora_scale"`
+	NoMulMatQ    bool    `yaml:"no_mulmatq"`
+	DraftModel   string  `yaml:"draft_model"`
+	NDraft       int32   `yaml:"n_draft"`
+	Quantization string  `yaml:"quantization"`
+	MMProj       string  `yaml:"mmproj"`
+
+	RopeScaling string `yaml:"rope_scaling"`
+	ModelType   string `yaml:"type"`
+
+	YarnExtFactor  float32 `yaml:"yarn_ext_factor"`
+	YarnAttnFactor float32 `yaml:"yarn_attn_factor"`
+	YarnBetaFast   float32 `yaml:"yarn_beta_fast"`
+	YarnBetaSlow   float32 `yaml:"yarn_beta_slow"`
+}
+
+type AutoGPTQ struct {
+	ModelBaseName    string `yaml:"model_base_name"`
+	Device           string `yaml:"device"`
+	Triton           bool   `yaml:"triton"`
+	UseFastTokenizer bool   `yaml:"use_fast_tokenizer"`
+}
+
+type Functions struct {
+	DisableNoAction         bool   `yaml:"disable_no_action"`
+	NoActionFunctionName    string `yaml:"no_action_function_name"`
+	NoActionDescriptionName string `yaml:"no_action_description_name"`
+}
+
+type TemplateConfig struct {
+	Chat        string `yaml:"chat"`
+	ChatMessage string `yaml:"chat_message"`
+	Completion  string `yaml:"completion"`
+	Edit        string `yaml:"edit"`
+	Functions   string `yaml:"function"`
+}
+
+type ConfigLoader struct {
+	configs map[string]Config
+	sync.Mutex
+}
+
+func (c *Config) SetFunctionCallString(s string) {
+	c.functionCallString = s
+}
+
+func (c *Config) SetFunctionCallNameString(s string) {
+	c.functionCallNameString = s
+}
+
+func (c *Config) ShouldUseFunctions() bool {
+	return ((c.functionCallString != "none" || c.functionCallString == "") || c.ShouldCallSpecificFunction())
+}
+
+func (c *Config) ShouldCallSpecificFunction() bool {
+	return len(c.functionCallNameString) > 0
+}
+
+func (c *Config) FunctionToCall() string {
+	return c.functionCallNameString
+}
+
+func defaultPredictOptions(modelFile string) PredictionOptions {
+	return PredictionOptions{
+		TopP:        0.7,
+		TopK:        80,
+		Maxtokens:   512,
+		Temperature: 0.9,
+		Model:       modelFile,
+	}
+}
+
+func DefaultConfig(modelFile string) *Config {
+	return &Config{
+		PredictionOptions: defaultPredictOptions(modelFile),
+	}
+}
+
+func NewConfigLoader() *ConfigLoader {
+	return &ConfigLoader{
+		configs: make(map[string]Config),
+	}
+}
+func ReadConfigFile(file string) ([]*Config, error) {
+	c := &[]*Config{}
+	f, err := os.ReadFile(file)
+	if err != nil {
+		return nil, fmt.Errorf("cannot read config file: %w", err)
+	}
+	if err := yaml.Unmarshal(f, c); err != nil {
+		return nil, fmt.Errorf("cannot unmarshal config file: %w", err)
+	}
+
+	return *c, nil
+}
+
+func ReadConfig(file string) (*Config, error) {
+	c := &Config{}
+	f, err := os.ReadFile(file)
+	if err != nil {
+		return nil, fmt.Errorf("cannot read config file: %w", err)
+	}
+	if err := yaml.Unmarshal(f, c); err != nil {
+		return nil, fmt.Errorf("cannot unmarshal config file: %w", err)
+	}
+
+	return c, nil
+}
+
+func (cm *ConfigLoader) LoadConfigFile(file string) error {
+	cm.Lock()
+	defer cm.Unlock()
+	c, err := ReadConfigFile(file)
+	if err != nil {
+		return fmt.Errorf("cannot load config file: %w", err)
+	}
+
+	for _, cc := range c {
+		cm.configs[cc.Name] = *cc
+	}
+	return nil
+}
+
+func (cm *ConfigLoader) LoadConfig(file string) error {
+	cm.Lock()
+	defer cm.Unlock()
+	c, err := ReadConfig(file)
+	if err != nil {
+		return fmt.Errorf("cannot read config file: %w", err)
+	}
+
+	cm.configs[c.Name] = *c
+	return nil
+}
+
+func (cm *ConfigLoader) GetConfig(m string) (Config, bool) {
+	cm.Lock()
+	defer cm.Unlock()
+	v, exists := cm.configs[m]
+	return v, exists
+}
+
+func (cm *ConfigLoader) GetAllConfigs() []Config {
+	cm.Lock()
+	defer cm.Unlock()
+	var res []Config
+	for _, v := range cm.configs {
+		res = append(res, v)
+	}
+	return res
+}
+
+func (cm *ConfigLoader) ListConfigs() []string {
+	cm.Lock()
+	defer cm.Unlock()
+	var res []string
+	for k := range cm.configs {
+		res = append(res, k)
+	}
+	return res
+}
+
+// Preload prepare models if they are not local but url or huggingface repositories
+func (cm *ConfigLoader) Preload(modelPath string) error {
+	cm.Lock()
+	defer cm.Unlock()
+
+	status := func(fileName, current, total string, percent float64) {
+		utils.DisplayDownloadFunction(fileName, current, total, percent)
+	}
+
+	log.Info().Msgf("Preloading models from %s", modelPath)
+
+	for i, config := range cm.configs {
+
+		// Download files and verify their SHA
+		for _, file := range config.DownloadFiles {
+			log.Debug().Msgf("Checking %q exists and matches SHA", file.Filename)
+
+			if err := utils.VerifyPath(file.Filename, modelPath); err != nil {
+				return err
+			}
+			// Create file path
+			filePath := filepath.Join(modelPath, file.Filename)
+
+			if err := downloader.DownloadFile(file.URI, filePath, file.SHA256, status); err != nil {
+				return err
+			}
+		}
+
+		modelURL := config.PredictionOptions.Model
+		modelURL = downloader.ConvertURL(modelURL)
+
+		if downloader.LooksLikeURL(modelURL) {
+			// md5 of model name
+			md5Name := utils.MD5(modelURL)
+
+			// check if file exists
+			if _, err := os.Stat(filepath.Join(modelPath, md5Name)); errors.Is(err, os.ErrNotExist) {
+				err := downloader.DownloadFile(modelURL, filepath.Join(modelPath, md5Name), "", status)
+				if err != nil {
+					return err
+				}
+			}
+
+			cc := cm.configs[i]
+			c := &cc
+			c.PredictionOptions.Model = md5Name
+			cm.configs[i] = *c
+		}
+		if cm.configs[i].Name != "" {
+			log.Info().Msgf("Model name: %s", cm.configs[i].Name)
+		}
+		if cm.configs[i].Description != "" {
+			log.Info().Msgf("Model description: %s", cm.configs[i].Description)
+		}
+		if cm.configs[i].Usage != "" {
+			log.Info().Msgf("Model usage: \n%s", cm.configs[i].Usage)
+		}
+	}
+	return nil
+}
+
+func (cm *ConfigLoader) LoadConfigs(path string) error {
+	cm.Lock()
+	defer cm.Unlock()
+	entries, err := os.ReadDir(path)
+	if err != nil {
+		return err
+	}
+	files := make([]fs.FileInfo, 0, len(entries))
+	for _, entry := range entries {
+		info, err := entry.Info()
+		if err != nil {
+			return err
+		}
+		files = append(files, info)
+	}
+	for _, file := range files {
+		// Skip templates, YAML and .keep files
+		if !strings.Contains(file.Name(), ".yaml") && !strings.Contains(file.Name(), ".yml") {
+			continue
+		}
+		c, err := ReadConfig(filepath.Join(path, file.Name()))
+		if err == nil {
+			cm.configs[c.Name] = *c
+		}
+	}
+
+	return nil
+}

api/config/config_test.go

@@ -0,0 +1,56 @@
+package api_config_test
+
+import (
+	"os"
+
+	. "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/pkg/model"
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+var _ = Describe("Test cases for config related functions", func() {
+
+	var (
+		configFile string
+	)
+
+	Context("Test Read configuration functions", func() {
+		configFile = os.Getenv("CONFIG_FILE")
+		It("Test ReadConfigFile", func() {
+			config, err := ReadConfigFile(configFile)
+			Expect(err).To(BeNil())
+			Expect(config).ToNot(BeNil())
+			// two configs in config.yaml
+			Expect(config[0].Name).To(Equal("list1"))
+			Expect(config[1].Name).To(Equal("list2"))
+		})
+
+		It("Test LoadConfigs", func() {
+			cm := NewConfigLoader()
+			opts := options.NewOptions()
+			modelLoader := model.NewModelLoader(os.Getenv("MODELS_PATH"))
+			options.WithModelLoader(modelLoader)(opts)
+
+			err := cm.LoadConfigs(opts.Loader.ModelPath)
+			Expect(err).To(BeNil())
+			Expect(cm.ListConfigs()).ToNot(BeNil())
+
+			// config should includes gpt4all models's api.config
+			Expect(cm.ListConfigs()).To(ContainElements("gpt4all"))
+
+			// config should includes gpt2 models's api.config
+			Expect(cm.ListConfigs()).To(ContainElements("gpt4all-2"))
+
+			// config should includes text-embedding-ada-002 models's api.config
+			Expect(cm.ListConfigs()).To(ContainElements("text-embedding-ada-002"))
+
+			// config should includes rwkv_test models's api.config
+			Expect(cm.ListConfigs()).To(ContainElements("rwkv_test"))
+
+			// config should includes whisper-1 models's api.config
+			Expect(cm.ListConfigs()).To(ContainElements("whisper-1"))
+		})
+	})
+})

api/config/prediction.go

@@ -0,0 +1,50 @@
+package api_config
+
+type PredictionOptions struct {
+
+	// Also part of the OpenAI official spec
+	Model string `json:"model" yaml:"model"`
+
+	// Also part of the OpenAI official spec
+	Language string `json:"language"`
+
+	// Also part of the OpenAI official spec. use it for returning multiple results
+	N int `json:"n"`
+
+	// Common options between all the API calls, part of the OpenAI spec
+	TopP        float64 `json:"top_p" yaml:"top_p"`
+	TopK        int     `json:"top_k" yaml:"top_k"`
+	Temperature float64 `json:"temperature" yaml:"temperature"`
+	Maxtokens   int     `json:"max_tokens" yaml:"max_tokens"`
+	Echo        bool    `json:"echo"`
+
+	// Custom parameters - not present in the OpenAI API
+	Batch         int     `json:"batch" yaml:"batch"`
+	F16           bool    `json:"f16" yaml:"f16"`
+	IgnoreEOS     bool    `json:"ignore_eos" yaml:"ignore_eos"`
+	RepeatPenalty float64 `json:"repeat_penalty" yaml:"repeat_penalty"`
+	Keep          int     `json:"n_keep" yaml:"n_keep"`
+
+	MirostatETA float64 `json:"mirostat_eta" yaml:"mirostat_eta"`
+	MirostatTAU float64 `json:"mirostat_tau" yaml:"mirostat_tau"`
+	Mirostat    int     `json:"mirostat" yaml:"mirostat"`
+
+	FrequencyPenalty float64 `json:"frequency_penalty" yaml:"frequency_penalty"`
+	TFZ              float64 `json:"tfz" yaml:"tfz"`
+
+	TypicalP float64 `json:"typical_p" yaml:"typical_p"`
+	Seed     int     `json:"seed" yaml:"seed"`
+
+	NegativePrompt      string  `json:"negative_prompt" yaml:"negative_prompt"`
+	RopeFreqBase        float32 `json:"rope_freq_base" yaml:"rope_freq_base"`
+	RopeFreqScale       float32 `json:"rope_freq_scale" yaml:"rope_freq_scale"`
+	NegativePromptScale float32 `json:"negative_prompt_scale" yaml:"negative_prompt_scale"`
+	// AutoGPTQ
+	UseFastTokenizer bool `json:"use_fast_tokenizer" yaml:"use_fast_tokenizer"`
+
+	// Diffusers
+	ClipSkip int `json:"clip_skip" yaml:"clip_skip"`
+
+	// RWKV (?)
+	Tokenizer string `json:"tokenizer" yaml:"tokenizer"`
+}

api/localai/backend_monitor.go

@@ -0,0 +1,162 @@
+package localai
+
+import (
+	"context"
+	"fmt"
+	"strings"
+
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/pkg/grpc/proto"
+
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/gofiber/fiber/v2"
+	"github.com/rs/zerolog/log"
+
+	gopsutil "github.com/shirou/gopsutil/v3/process"
+)
+
+type BackendMonitorRequest struct {
+	Model string `json:"model" yaml:"model"`
+}
+
+type BackendMonitorResponse struct {
+	MemoryInfo    *gopsutil.MemoryInfoStat
+	MemoryPercent float32
+	CPUPercent    float64
+}
+
+type BackendMonitor struct {
+	configLoader *config.ConfigLoader
+	options      *options.Option // Taking options in case we need to inspect ExternalGRPCBackends, though that's out of scope for now, hence the name.
+}
+
+func NewBackendMonitor(configLoader *config.ConfigLoader, options *options.Option) BackendMonitor {
+	return BackendMonitor{
+		configLoader: configLoader,
+		options:      options,
+	}
+}
+
+func (bm *BackendMonitor) SampleLocalBackendProcess(model string) (*BackendMonitorResponse, error) {
+	config, exists := bm.configLoader.GetConfig(model)
+	var backend string
+	if exists {
+		backend = config.Model
+	} else {
+		// Last ditch effort: use it raw, see if a backend happens to match.
+		backend = model
+	}
+
+	if !strings.HasSuffix(backend, ".bin") {
+		backend = fmt.Sprintf("%s.bin", backend)
+	}
+
+	pid, err := bm.options.Loader.GetGRPCPID(backend)
+
+	if err != nil {
+		log.Error().Msgf("model %s : failed to find pid %+v", model, err)
+		return nil, err
+	}
+
+	// Name is slightly frightening but this does _not_ create a new process, rather it looks up an existing process by PID.
+	backendProcess, err := gopsutil.NewProcess(int32(pid))
+
+	if err != nil {
+		log.Error().Msgf("model %s [PID %d] : error getting process info %+v", model, pid, err)
+		return nil, err
+	}
+
+	memInfo, err := backendProcess.MemoryInfo()
+
+	if err != nil {
+		log.Error().Msgf("model %s [PID %d] : error getting memory info %+v", model, pid, err)
+		return nil, err
+	}
+
+	memPercent, err := backendProcess.MemoryPercent()
+	if err != nil {
+		log.Error().Msgf("model %s [PID %d] : error getting memory percent %+v", model, pid, err)
+		return nil, err
+	}
+
+	cpuPercent, err := backendProcess.CPUPercent()
+	if err != nil {
+		log.Error().Msgf("model %s [PID %d] : error getting cpu percent %+v", model, pid, err)
+		return nil, err
+	}
+
+	return &BackendMonitorResponse{
+		MemoryInfo:    memInfo,
+		MemoryPercent: memPercent,
+		CPUPercent:    cpuPercent,
+	}, nil
+}
+
+func (bm BackendMonitor) getModelLoaderIDFromCtx(c *fiber.Ctx) (string, error) {
+	input := new(BackendMonitorRequest)
+	// Get input data from the request body
+	if err := c.BodyParser(input); err != nil {
+		return "", err
+	}
+
+	config, exists := bm.configLoader.GetConfig(input.Model)
+	var backendId string
+	if exists {
+		backendId = config.Model
+	} else {
+		// Last ditch effort: use it raw, see if a backend happens to match.
+		backendId = input.Model
+	}
+
+	if !strings.HasSuffix(backendId, ".bin") {
+		backendId = fmt.Sprintf("%s.bin", backendId)
+	}
+
+	return backendId, nil
+}
+
+func BackendMonitorEndpoint(bm BackendMonitor) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+
+		backendId, err := bm.getModelLoaderIDFromCtx(c)
+		if err != nil {
+			return err
+		}
+
+		model := bm.options.Loader.CheckIsLoaded(backendId)
+		if model == "" {
+			return fmt.Errorf("backend %s is not currently loaded", backendId)
+		}
+
+		status, rpcErr := model.GRPC(false, nil).Status(context.TODO())
+		if rpcErr != nil {
+			log.Warn().Msgf("backend %s experienced an error retrieving status info: %s", backendId, rpcErr.Error())
+			val, slbErr := bm.SampleLocalBackendProcess(backendId)
+			if slbErr != nil {
+				return fmt.Errorf("backend %s experienced an error retrieving status info via rpc: %s, then failed local node process sample: %s", backendId, rpcErr.Error(), slbErr.Error())
+			}
+			return c.JSON(proto.StatusResponse{
+				State: proto.StatusResponse_ERROR,
+				Memory: &proto.MemoryUsageData{
+					Total: val.MemoryInfo.VMS,
+					Breakdown: map[string]uint64{
+						"gopsutil-RSS": val.MemoryInfo.RSS,
+					},
+				},
+			})
+		}
+
+		return c.JSON(status)
+	}
+}
+
+func BackendShutdownEndpoint(bm BackendMonitor) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		backendId, err := bm.getModelLoaderIDFromCtx(c)
+		if err != nil {
+			return err
+		}
+
+		return bm.options.Loader.ShutdownModel(backendId)
+	}
+}

api/localai/gallery.go

@@ -0,0 +1,326 @@
+package localai
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"slices"
+	"strings"
+	"sync"
+
+	json "github.com/json-iterator/go"
+	"gopkg.in/yaml.v3"
+
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/pkg/gallery"
+	"github.com/go-skynet/LocalAI/pkg/utils"
+
+	"github.com/gofiber/fiber/v2"
+	"github.com/google/uuid"
+	"github.com/rs/zerolog/log"
+)
+
+type galleryOp struct {
+	req         gallery.GalleryModel
+	id          string
+	galleries   []gallery.Gallery
+	galleryName string
+}
+
+type galleryOpStatus struct {
+	FileName           string  `json:"file_name"`
+	Error              error   `json:"error"`
+	Processed          bool    `json:"processed"`
+	Message            string  `json:"message"`
+	Progress           float64 `json:"progress"`
+	TotalFileSize      string  `json:"file_size"`
+	DownloadedFileSize string  `json:"downloaded_size"`
+}
+
+type galleryApplier struct {
+	modelPath string
+	sync.Mutex
+	C        chan galleryOp
+	statuses map[string]*galleryOpStatus
+}
+
+func NewGalleryService(modelPath string) *galleryApplier {
+	return &galleryApplier{
+		modelPath: modelPath,
+		C:         make(chan galleryOp),
+		statuses:  make(map[string]*galleryOpStatus),
+	}
+}
+
+func prepareModel(modelPath string, req gallery.GalleryModel, cm *config.ConfigLoader, downloadStatus func(string, string, string, float64)) error {
+
+	config, err := gallery.GetGalleryConfigFromURL(req.URL)
+	if err != nil {
+		return err
+	}
+
+	config.Files = append(config.Files, req.AdditionalFiles...)
+
+	return gallery.InstallModel(modelPath, req.Name, &config, req.Overrides, downloadStatus)
+}
+
+func (g *galleryApplier) updateStatus(s string, op *galleryOpStatus) {
+	g.Lock()
+	defer g.Unlock()
+	g.statuses[s] = op
+}
+
+func (g *galleryApplier) getStatus(s string) *galleryOpStatus {
+	g.Lock()
+	defer g.Unlock()
+
+	return g.statuses[s]
+}
+
+func (g *galleryApplier) getAllStatus() map[string]*galleryOpStatus {
+	g.Lock()
+	defer g.Unlock()
+
+	return g.statuses
+}
+
+func (g *galleryApplier) Start(c context.Context, cm *config.ConfigLoader) {
+	go func() {
+		for {
+			select {
+			case <-c.Done():
+				return
+			case op := <-g.C:
+				utils.ResetDownloadTimers()
+
+				g.updateStatus(op.id, &galleryOpStatus{Message: "processing", Progress: 0})
+
+				// updates the status with an error
+				updateError := func(e error) {
+					g.updateStatus(op.id, &galleryOpStatus{Error: e, Processed: true, Message: "error: " + e.Error()})
+				}
+
+				// displayDownload displays the download progress
+				progressCallback := func(fileName string, current string, total string, percentage float64) {
+					g.updateStatus(op.id, &galleryOpStatus{Message: "processing", FileName: fileName, Progress: percentage, TotalFileSize: total, DownloadedFileSize: current})
+					utils.DisplayDownloadFunction(fileName, current, total, percentage)
+				}
+
+				var err error
+				// if the request contains a gallery name, we apply the gallery from the gallery list
+				if op.galleryName != "" {
+					if strings.Contains(op.galleryName, "@") {
+						err = gallery.InstallModelFromGallery(op.galleries, op.galleryName, g.modelPath, op.req, progressCallback)
+					} else {
+						err = gallery.InstallModelFromGalleryByName(op.galleries, op.galleryName, g.modelPath, op.req, progressCallback)
+					}
+				} else {
+					err = prepareModel(g.modelPath, op.req, cm, progressCallback)
+				}
+
+				if err != nil {
+					updateError(err)
+					continue
+				}
+
+				// Reload models
+				err = cm.LoadConfigs(g.modelPath)
+				if err != nil {
+					updateError(err)
+					continue
+				}
+
+				err = cm.Preload(g.modelPath)
+				if err != nil {
+					updateError(err)
+					continue
+				}
+
+				g.updateStatus(op.id, &galleryOpStatus{Processed: true, Message: "completed", Progress: 100})
+			}
+		}
+	}()
+}
+
+type galleryModel struct {
+	gallery.GalleryModel `yaml:",inline"` // https://github.com/go-yaml/yaml/issues/63
+	ID                   string           `json:"id"`
+}
+
+func processRequests(modelPath, s string, cm *config.ConfigLoader, galleries []gallery.Gallery, requests []galleryModel) error {
+	var err error
+	for _, r := range requests {
+		utils.ResetDownloadTimers()
+		if r.ID == "" {
+			err = prepareModel(modelPath, r.GalleryModel, cm, utils.DisplayDownloadFunction)
+		} else {
+			if strings.Contains(r.ID, "@") {
+				err = gallery.InstallModelFromGallery(
+					galleries, r.ID, modelPath, r.GalleryModel, utils.DisplayDownloadFunction)
+			} else {
+				err = gallery.InstallModelFromGalleryByName(
+					galleries, r.ID, modelPath, r.GalleryModel, utils.DisplayDownloadFunction)
+			}
+		}
+	}
+	return err
+}
+
+func ApplyGalleryFromFile(modelPath, s string, cm *config.ConfigLoader, galleries []gallery.Gallery) error {
+	dat, err := os.ReadFile(s)
+	if err != nil {
+		return err
+	}
+	var requests []galleryModel
+
+	if err := yaml.Unmarshal(dat, &requests); err != nil {
+		return err
+	}
+
+	return processRequests(modelPath, s, cm, galleries, requests)
+}
+
+func ApplyGalleryFromString(modelPath, s string, cm *config.ConfigLoader, galleries []gallery.Gallery) error {
+	var requests []galleryModel
+	err := json.Unmarshal([]byte(s), &requests)
+	if err != nil {
+		return err
+	}
+
+	return processRequests(modelPath, s, cm, galleries, requests)
+}
+
+/// Endpoint Service
+
+type ModelGalleryService struct {
+	galleries      []gallery.Gallery
+	modelPath      string
+	galleryApplier *galleryApplier
+}
+
+type GalleryModel struct {
+	ID string `json:"id"`
+	gallery.GalleryModel
+}
+
+func CreateModelGalleryService(galleries []gallery.Gallery, modelPath string, galleryApplier *galleryApplier) ModelGalleryService {
+	return ModelGalleryService{
+		galleries:      galleries,
+		modelPath:      modelPath,
+		galleryApplier: galleryApplier,
+	}
+}
+
+func (mgs *ModelGalleryService) GetOpStatusEndpoint() func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		status := mgs.galleryApplier.getStatus(c.Params("uuid"))
+		if status == nil {
+			return fmt.Errorf("could not find any status for ID")
+		}
+		return c.JSON(status)
+	}
+}
+
+func (mgs *ModelGalleryService) GetAllStatusEndpoint() func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		return c.JSON(mgs.galleryApplier.getAllStatus())
+	}
+}
+
+func (mgs *ModelGalleryService) ApplyModelGalleryEndpoint() func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		input := new(GalleryModel)
+		// Get input data from the request body
+		if err := c.BodyParser(input); err != nil {
+			return err
+		}
+
+		uuid, err := uuid.NewUUID()
+		if err != nil {
+			return err
+		}
+		mgs.galleryApplier.C <- galleryOp{
+			req:         input.GalleryModel,
+			id:          uuid.String(),
+			galleryName: input.ID,
+			galleries:   mgs.galleries,
+		}
+		return c.JSON(struct {
+			ID        string `json:"uuid"`
+			StatusURL string `json:"status"`
+		}{ID: uuid.String(), StatusURL: c.BaseURL() + "/models/jobs/" + uuid.String()})
+	}
+}
+
+func (mgs *ModelGalleryService) ListModelFromGalleryEndpoint() func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		log.Debug().Msgf("Listing models from galleries: %+v", mgs.galleries)
+
+		models, err := gallery.AvailableGalleryModels(mgs.galleries, mgs.modelPath)
+		if err != nil {
+			return err
+		}
+		log.Debug().Msgf("Models found from galleries: %+v", models)
+		for _, m := range models {
+			log.Debug().Msgf("Model found from galleries: %+v", m)
+		}
+		dat, err := json.Marshal(models)
+		if err != nil {
+			return err
+		}
+		return c.Send(dat)
+	}
+}
+
+// NOTE: This is different (and much simpler!) than above! This JUST lists the model galleries that have been loaded, not their contents!
+func (mgs *ModelGalleryService) ListModelGalleriesEndpoint() func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		log.Debug().Msgf("Listing model galleries %+v", mgs.galleries)
+		dat, err := json.Marshal(mgs.galleries)
+		if err != nil {
+			return err
+		}
+		return c.Send(dat)
+	}
+}
+
+func (mgs *ModelGalleryService) AddModelGalleryEndpoint() func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		input := new(gallery.Gallery)
+		// Get input data from the request body
+		if err := c.BodyParser(input); err != nil {
+			return err
+		}
+		if slices.ContainsFunc(mgs.galleries, func(gallery gallery.Gallery) bool {
+			return gallery.Name == input.Name
+		}) {
+			return fmt.Errorf("%s already exists", input.Name)
+		}
+		dat, err := json.Marshal(mgs.galleries)
+		if err != nil {
+			return err
+		}
+		log.Debug().Msgf("Adding %+v to gallery list", *input)
+		mgs.galleries = append(mgs.galleries, *input)
+		return c.Send(dat)
+	}
+}
+
+func (mgs *ModelGalleryService) RemoveModelGalleryEndpoint() func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		input := new(gallery.Gallery)
+		// Get input data from the request body
+		if err := c.BodyParser(input); err != nil {
+			return err
+		}
+		if !slices.ContainsFunc(mgs.galleries, func(gallery gallery.Gallery) bool {
+			return gallery.Name == input.Name
+		}) {
+			return fmt.Errorf("%s is not currently registered", input.Name)
+		}
+		mgs.galleries = slices.DeleteFunc(mgs.galleries, func(gallery gallery.Gallery) bool {
+			return gallery.Name == input.Name
+		})
+		return c.Send(nil)
+	}
+}

api/localai/localai.go

@@ -0,0 +1,32 @@
+package localai
+
+import (
+	"github.com/go-skynet/LocalAI/api/backend"
+	config "github.com/go-skynet/LocalAI/api/config"
+
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/gofiber/fiber/v2"
+)
+
+type TTSRequest struct {
+	Model   string `json:"model" yaml:"model"`
+	Input   string `json:"input" yaml:"input"`
+	Backend string `json:"backend" yaml:"backend"`
+}
+
+func TTSEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+
+		input := new(TTSRequest)
+		// Get input data from the request body
+		if err := c.BodyParser(input); err != nil {
+			return err
+		}
+
+		filePath, _, err := backend.ModelTTS(input.Backend, input.Input, input.Model, o.Loader, o)
+		if err != nil {
+			return err
+		}
+		return c.Download(filePath)
+	}
+}

api/openai/chat.go

@@ -0,0 +1,399 @@
+package openai
+
+import (
+	"bufio"
+	"bytes"
+	"encoding/json"
+	"fmt"
+	"strings"
+	"time"
+
+	"github.com/go-skynet/LocalAI/api/backend"
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/api/schema"
+	"github.com/go-skynet/LocalAI/pkg/grammar"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/go-skynet/LocalAI/pkg/utils"
+	"github.com/gofiber/fiber/v2"
+	"github.com/google/uuid"
+	"github.com/rs/zerolog/log"
+	"github.com/valyala/fasthttp"
+)
+
+func ChatEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
+	emptyMessage := ""
+	id := uuid.New().String()
+	created := int(time.Now().Unix())
+
+	process := func(s string, req *schema.OpenAIRequest, config *config.Config, loader *model.ModelLoader, responses chan schema.OpenAIResponse) {
+		initialMessage := schema.OpenAIResponse{
+			ID:      id,
+			Created: created,
+			Model:   req.Model, // we have to return what the user sent here, due to OpenAI spec.
+			Choices: []schema.Choice{{Delta: &schema.Message{Role: "assistant", Content: &emptyMessage}}},
+			Object:  "chat.completion.chunk",
+		}
+		responses <- initialMessage
+
+		ComputeChoices(req, s, config, o, loader, func(s string, c *[]schema.Choice) {}, func(s string, usage backend.TokenUsage) bool {
+			resp := schema.OpenAIResponse{
+				ID:      id,
+				Created: created,
+				Model:   req.Model, // we have to return what the user sent here, due to OpenAI spec.
+				Choices: []schema.Choice{{Delta: &schema.Message{Content: &s}, Index: 0}},
+				Object:  "chat.completion.chunk",
+				Usage: schema.OpenAIUsage{
+					PromptTokens:     usage.Prompt,
+					CompletionTokens: usage.Completion,
+					TotalTokens:      usage.Prompt + usage.Completion,
+				},
+			}
+
+			responses <- resp
+			return true
+		})
+		close(responses)
+	}
+	return func(c *fiber.Ctx) error {
+		processFunctions := false
+		funcs := grammar.Functions{}
+		modelFile, input, err := readInput(c, o, true)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		config, input, err := readConfig(modelFile, input, cm, o.Loader, o.Debug, o.Threads, o.ContextSize, o.F16)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+		log.Debug().Msgf("Configuration read: %+v", config)
+
+		// Allow the user to set custom actions via config file
+		// to be "embedded" in each model
+		noActionName := "answer"
+		noActionDescription := "use this action to answer without performing any action"
+
+		if config.FunctionsConfig.NoActionFunctionName != "" {
+			noActionName = config.FunctionsConfig.NoActionFunctionName
+		}
+		if config.FunctionsConfig.NoActionDescriptionName != "" {
+			noActionDescription = config.FunctionsConfig.NoActionDescriptionName
+		}
+
+		if input.ResponseFormat.Type == "json_object" {
+			input.Grammar = grammar.JSONBNF
+		}
+
+		// process functions if we have any defined or if we have a function call string
+		if len(input.Functions) > 0 && config.ShouldUseFunctions() {
+			log.Debug().Msgf("Response needs to process functions")
+
+			processFunctions = true
+
+			noActionGrammar := grammar.Function{
+				Name:        noActionName,
+				Description: noActionDescription,
+				Parameters: map[string]interface{}{
+					"properties": map[string]interface{}{
+						"message": map[string]interface{}{
+							"type":        "string",
+							"description": "The message to reply the user with",
+						}},
+				},
+			}
+
+			// Append the no action function
+			funcs = append(funcs, input.Functions...)
+			if !config.FunctionsConfig.DisableNoAction {
+				funcs = append(funcs, noActionGrammar)
+			}
+
+			// Force picking one of the functions by the request
+			if config.FunctionToCall() != "" {
+				funcs = funcs.Select(config.FunctionToCall())
+			}
+
+			// Update input grammar
+			jsStruct := funcs.ToJSONStructure()
+			config.Grammar = jsStruct.Grammar("")
+		} else if input.JSONFunctionGrammarObject != nil {
+			config.Grammar = input.JSONFunctionGrammarObject.Grammar("")
+		}
+
+		// functions are not supported in stream mode (yet?)
+		toStream := input.Stream && !processFunctions
+
+		log.Debug().Msgf("Parameters: %+v", config)
+
+		var predInput string
+
+		suppressConfigSystemPrompt := false
+		mess := []string{}
+		for messageIndex, i := range input.Messages {
+			var content string
+			role := i.Role
+
+			// if function call, we might want to customize the role so we can display better that the "assistant called a json action"
+			// if an "assistant_function_call" role is defined, we use it, otherwise we use the role that is passed by in the request
+			if i.FunctionCall != nil && i.Role == "assistant" {
+				roleFn := "assistant_function_call"
+				r := config.Roles[roleFn]
+				if r != "" {
+					role = roleFn
+				}
+			}
+			r := config.Roles[role]
+			contentExists := i.Content != nil && i.StringContent != ""
+			// First attempt to populate content via a chat message specific template
+			if config.TemplateConfig.ChatMessage != "" {
+				chatMessageData := model.ChatMessageTemplateData{
+					SystemPrompt: config.SystemPrompt,
+					Role:         r,
+					RoleName:     role,
+					Content:      i.StringContent,
+					MessageIndex: messageIndex,
+				}
+				templatedChatMessage, err := o.Loader.EvaluateTemplateForChatMessage(config.TemplateConfig.ChatMessage, chatMessageData)
+				if err != nil {
+					log.Error().Msgf("error processing message %+v using template \"%s\": %v. Skipping!", chatMessageData, config.TemplateConfig.ChatMessage, err)
+				} else {
+					if templatedChatMessage == "" {
+						log.Warn().Msgf("template \"%s\" produced blank output for %+v. Skipping!", config.TemplateConfig.ChatMessage, chatMessageData)
+						continue // TODO: This continue is here intentionally to skip over the line `mess = append(mess, content)` below, and to prevent the sprintf
+					}
+					log.Debug().Msgf("templated message for chat: %s", templatedChatMessage)
+					content = templatedChatMessage
+				}
+			}
+			// If this model doesn't have such a template, or if that template fails to return a value, template at the message level.
+			if content == "" {
+				if r != "" {
+					if contentExists {
+						content = fmt.Sprint(r, i.StringContent)
+					}
+					if i.FunctionCall != nil {
+						j, err := json.Marshal(i.FunctionCall)
+						if err == nil {
+							if contentExists {
+								content += "\n" + fmt.Sprint(r, " ", string(j))
+							} else {
+								content = fmt.Sprint(r, " ", string(j))
+							}
+						}
+					}
+				} else {
+					if contentExists {
+						content = fmt.Sprint(i.StringContent)
+					}
+					if i.FunctionCall != nil {
+						j, err := json.Marshal(i.FunctionCall)
+						if err == nil {
+							if contentExists {
+								content += "\n" + string(j)
+							} else {
+								content = string(j)
+							}
+						}
+					}
+				}
+				// Special Handling: System. We care if it was printed at all, not the r branch, so check seperately
+				if contentExists && role == "system" {
+					suppressConfigSystemPrompt = true
+				}
+			}
+
+			mess = append(mess, content)
+		}
+
+		predInput = strings.Join(mess, "\n")
+		log.Debug().Msgf("Prompt (before templating): %s", predInput)
+
+		if toStream {
+			log.Debug().Msgf("Stream request received")
+			c.Context().SetContentType("text/event-stream")
+			//c.Response().Header.SetContentType(fiber.MIMETextHTMLCharsetUTF8)
+			//	c.Set("Content-Type", "text/event-stream")
+			c.Set("Cache-Control", "no-cache")
+			c.Set("Connection", "keep-alive")
+			c.Set("Transfer-Encoding", "chunked")
+		}
+
+		templateFile := ""
+
+		// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
+		if o.Loader.ExistsInModelPath(fmt.Sprintf("%s.tmpl", config.Model)) {
+			templateFile = config.Model
+		}
+
+		if config.TemplateConfig.Chat != "" && !processFunctions {
+			templateFile = config.TemplateConfig.Chat
+		}
+
+		if config.TemplateConfig.Functions != "" && processFunctions {
+			templateFile = config.TemplateConfig.Functions
+		}
+
+		if templateFile != "" {
+			templatedInput, err := o.Loader.EvaluateTemplateForPrompt(model.ChatPromptTemplate, templateFile, model.PromptTemplateData{
+				SystemPrompt:         config.SystemPrompt,
+				SuppressSystemPrompt: suppressConfigSystemPrompt,
+				Input:                predInput,
+				Functions:            funcs,
+			})
+			if err == nil {
+				predInput = templatedInput
+				log.Debug().Msgf("Template found, input modified to: %s", predInput)
+			} else {
+				log.Debug().Msgf("Template failed loading: %s", err.Error())
+			}
+		}
+
+		log.Debug().Msgf("Prompt (after templating): %s", predInput)
+		if processFunctions {
+			log.Debug().Msgf("Grammar: %+v", config.Grammar)
+		}
+
+		if toStream {
+			responses := make(chan schema.OpenAIResponse)
+
+			go process(predInput, input, config, o.Loader, responses)
+
+			c.Context().SetBodyStreamWriter(fasthttp.StreamWriter(func(w *bufio.Writer) {
+
+				usage := &schema.OpenAIUsage{}
+
+				for ev := range responses {
+					usage = &ev.Usage // Copy a pointer to the latest usage chunk so that the stop message can reference it
+					var buf bytes.Buffer
+					enc := json.NewEncoder(&buf)
+					enc.Encode(ev)
+					log.Debug().Msgf("Sending chunk: %s", buf.String())
+					_, err := fmt.Fprintf(w, "data: %v\n", buf.String())
+					if err != nil {
+						log.Debug().Msgf("Sending chunk failed: %v", err)
+						input.Cancel()
+						break
+					}
+					w.Flush()
+				}
+
+				resp := &schema.OpenAIResponse{
+					ID:      id,
+					Created: created,
+					Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
+					Choices: []schema.Choice{
+						{
+							FinishReason: "stop",
+							Index:        0,
+							Delta:        &schema.Message{Content: &emptyMessage},
+						}},
+					Object: "chat.completion.chunk",
+					Usage:  *usage,
+				}
+				respData, _ := json.Marshal(resp)
+
+				w.WriteString(fmt.Sprintf("data: %s\n\n", respData))
+				w.WriteString("data: [DONE]\n\n")
+				w.Flush()
+			}))
+			return nil
+		}
+
+		result, tokenUsage, err := ComputeChoices(input, predInput, config, o, o.Loader, func(s string, c *[]schema.Choice) {
+			if processFunctions {
+				// As we have to change the result before processing, we can't stream the answer (yet?)
+				ss := map[string]interface{}{}
+				// This prevent newlines to break JSON parsing for clients
+				s = utils.EscapeNewLines(s)
+				json.Unmarshal([]byte(s), &ss)
+				log.Debug().Msgf("Function return: %s %+v", s, ss)
+
+				// The grammar defines the function name as "function", while OpenAI returns "name"
+				func_name := ss["function"]
+				// Similarly, while here arguments is a map[string]interface{}, OpenAI actually want a stringified object
+				args := ss["arguments"] // arguments needs to be a string, but we return an object from the grammar result (TODO: fix)
+				d, _ := json.Marshal(args)
+
+				ss["arguments"] = string(d)
+				ss["name"] = func_name
+
+				// if do nothing, reply with a message
+				if func_name == noActionName {
+					log.Debug().Msgf("nothing to do, computing a reply")
+
+					// If there is a message that the LLM already sends as part of the JSON reply, use it
+					arguments := map[string]interface{}{}
+					json.Unmarshal([]byte(d), &arguments)
+					m, exists := arguments["message"]
+					if exists {
+						switch message := m.(type) {
+						case string:
+							if message != "" {
+								log.Debug().Msgf("Reply received from LLM: %s", message)
+								message = backend.Finetune(*config, predInput, message)
+								log.Debug().Msgf("Reply received from LLM(finetuned): %s", message)
+
+								*c = append(*c, schema.Choice{Message: &schema.Message{Role: "assistant", Content: &message}})
+								return
+							}
+						}
+					}
+
+					log.Debug().Msgf("No action received from LLM, without a message, computing a reply")
+					// Otherwise ask the LLM to understand the JSON output and the context, and return a message
+					// Note: This costs (in term of CPU) another computation
+					config.Grammar = ""
+					images := []string{}
+					for _, m := range input.Messages {
+						images = append(images, m.StringImages...)
+					}
+					predFunc, err := backend.ModelInference(input.Context, predInput, images, o.Loader, *config, o, nil)
+					if err != nil {
+						log.Error().Msgf("inference error: %s", err.Error())
+						return
+					}
+
+					prediction, err := predFunc()
+					if err != nil {
+						log.Error().Msgf("inference error: %s", err.Error())
+						return
+					}
+
+					fineTunedResponse := backend.Finetune(*config, predInput, prediction.Response)
+					*c = append(*c, schema.Choice{Message: &schema.Message{Role: "assistant", Content: &fineTunedResponse}})
+				} else {
+					// otherwise reply with the function call
+					*c = append(*c, schema.Choice{
+						FinishReason: "function_call",
+						Message:      &schema.Message{Role: "assistant", FunctionCall: ss},
+					})
+				}
+
+				return
+			}
+			*c = append(*c, schema.Choice{FinishReason: "stop", Index: 0, Message: &schema.Message{Role: "assistant", Content: &s}})
+		}, nil)
+		if err != nil {
+			return err
+		}
+
+		resp := &schema.OpenAIResponse{
+			ID:      id,
+			Created: created,
+			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
+			Choices: result,
+			Object:  "chat.completion",
+			Usage: schema.OpenAIUsage{
+				PromptTokens:     tokenUsage.Prompt,
+				CompletionTokens: tokenUsage.Completion,
+				TotalTokens:      tokenUsage.Prompt + tokenUsage.Completion,
+			},
+		}
+		respData, _ := json.Marshal(resp)
+		log.Debug().Msgf("Response: %s", respData)
+
+		// Return the prediction in the response body
+		return c.JSON(resp)
+	}
+}

api/openai/completion.go

@@ -0,0 +1,199 @@
+package openai
+
+import (
+	"bufio"
+	"bytes"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"time"
+
+	"github.com/go-skynet/LocalAI/api/backend"
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/api/schema"
+	"github.com/go-skynet/LocalAI/pkg/grammar"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/gofiber/fiber/v2"
+	"github.com/google/uuid"
+	"github.com/rs/zerolog/log"
+	"github.com/valyala/fasthttp"
+)
+
+// https://platform.openai.com/docs/api-reference/completions
+func CompletionEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
+	id := uuid.New().String()
+	created := int(time.Now().Unix())
+
+	process := func(s string, req *schema.OpenAIRequest, config *config.Config, loader *model.ModelLoader, responses chan schema.OpenAIResponse) {
+		ComputeChoices(req, s, config, o, loader, func(s string, c *[]schema.Choice) {}, func(s string, usage backend.TokenUsage) bool {
+			resp := schema.OpenAIResponse{
+				ID:      id,
+				Created: created,
+				Model:   req.Model, // we have to return what the user sent here, due to OpenAI spec.
+				Choices: []schema.Choice{
+					{
+						Index: 0,
+						Text:  s,
+					},
+				},
+				Object: "text_completion",
+				Usage: schema.OpenAIUsage{
+					PromptTokens:     usage.Prompt,
+					CompletionTokens: usage.Completion,
+					TotalTokens:      usage.Prompt + usage.Completion,
+				},
+			}
+			log.Debug().Msgf("Sending goroutine: %s", s)
+
+			responses <- resp
+			return true
+		})
+		close(responses)
+	}
+
+	return func(c *fiber.Ctx) error {
+		modelFile, input, err := readInput(c, o, true)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		log.Debug().Msgf("`input`: %+v", input)
+
+		config, input, err := readConfig(modelFile, input, cm, o.Loader, o.Debug, o.Threads, o.ContextSize, o.F16)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		if input.ResponseFormat.Type == "json_object" {
+			input.Grammar = grammar.JSONBNF
+		}
+
+		log.Debug().Msgf("Parameter Config: %+v", config)
+
+		if input.Stream {
+			log.Debug().Msgf("Stream request received")
+			c.Context().SetContentType("text/event-stream")
+			//c.Response().Header.SetContentType(fiber.MIMETextHTMLCharsetUTF8)
+			//c.Set("Content-Type", "text/event-stream")
+			c.Set("Cache-Control", "no-cache")
+			c.Set("Connection", "keep-alive")
+			c.Set("Transfer-Encoding", "chunked")
+		}
+
+		templateFile := ""
+
+		// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
+		if o.Loader.ExistsInModelPath(fmt.Sprintf("%s.tmpl", config.Model)) {
+			templateFile = config.Model
+		}
+
+		if config.TemplateConfig.Completion != "" {
+			templateFile = config.TemplateConfig.Completion
+		}
+
+		if input.Stream {
+			if len(config.PromptStrings) > 1 {
+				return errors.New("cannot handle more than 1 `PromptStrings` when Streaming")
+			}
+
+			predInput := config.PromptStrings[0]
+
+			if templateFile != "" {
+				templatedInput, err := o.Loader.EvaluateTemplateForPrompt(model.CompletionPromptTemplate, templateFile, model.PromptTemplateData{
+					Input: predInput,
+				})
+				if err == nil {
+					predInput = templatedInput
+					log.Debug().Msgf("Template found, input modified to: %s", predInput)
+				}
+			}
+
+			responses := make(chan schema.OpenAIResponse)
+
+			go process(predInput, input, config, o.Loader, responses)
+
+			c.Context().SetBodyStreamWriter(fasthttp.StreamWriter(func(w *bufio.Writer) {
+
+				for ev := range responses {
+					var buf bytes.Buffer
+					enc := json.NewEncoder(&buf)
+					enc.Encode(ev)
+
+					log.Debug().Msgf("Sending chunk: %s", buf.String())
+					fmt.Fprintf(w, "data: %v\n", buf.String())
+					w.Flush()
+				}
+
+				resp := &schema.OpenAIResponse{
+					ID:      id,
+					Created: created,
+					Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
+					Choices: []schema.Choice{
+						{
+							Index:        0,
+							FinishReason: "stop",
+						},
+					},
+					Object: "text_completion",
+				}
+				respData, _ := json.Marshal(resp)
+
+				w.WriteString(fmt.Sprintf("data: %s\n\n", respData))
+				w.WriteString("data: [DONE]\n\n")
+				w.Flush()
+			}))
+			return nil
+		}
+
+		var result []schema.Choice
+
+		totalTokenUsage := backend.TokenUsage{}
+
+		for k, i := range config.PromptStrings {
+			if templateFile != "" {
+				// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
+				templatedInput, err := o.Loader.EvaluateTemplateForPrompt(model.CompletionPromptTemplate, templateFile, model.PromptTemplateData{
+					SystemPrompt: config.SystemPrompt,
+					Input:        i,
+				})
+				if err == nil {
+					i = templatedInput
+					log.Debug().Msgf("Template found, input modified to: %s", i)
+				}
+			}
+
+			r, tokenUsage, err := ComputeChoices(
+				input, i, config, o, o.Loader, func(s string, c *[]schema.Choice) {
+					*c = append(*c, schema.Choice{Text: s, FinishReason: "stop", Index: k})
+				}, nil)
+			if err != nil {
+				return err
+			}
+
+			totalTokenUsage.Prompt += tokenUsage.Prompt
+			totalTokenUsage.Completion += tokenUsage.Completion
+
+			result = append(result, r...)
+		}
+
+		resp := &schema.OpenAIResponse{
+			ID:      id,
+			Created: created,
+			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
+			Choices: result,
+			Object:  "text_completion",
+			Usage: schema.OpenAIUsage{
+				PromptTokens:     totalTokenUsage.Prompt,
+				CompletionTokens: totalTokenUsage.Completion,
+				TotalTokens:      totalTokenUsage.Prompt + totalTokenUsage.Completion,
+			},
+		}
+
+		jsonResult, _ := json.Marshal(resp)
+		log.Debug().Msgf("Response: %s", jsonResult)
+
+		// Return the prediction in the response body
+		return c.JSON(resp)
+	}
+}

api/openai/edit.go

@@ -0,0 +1,94 @@
+package openai
+
+import (
+	"encoding/json"
+	"fmt"
+	"time"
+
+	"github.com/go-skynet/LocalAI/api/backend"
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/api/schema"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/gofiber/fiber/v2"
+	"github.com/google/uuid"
+
+	"github.com/rs/zerolog/log"
+)
+
+func EditEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		modelFile, input, err := readInput(c, o, true)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		config, input, err := readConfig(modelFile, input, cm, o.Loader, o.Debug, o.Threads, o.ContextSize, o.F16)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		log.Debug().Msgf("Parameter Config: %+v", config)
+
+		templateFile := ""
+
+		// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
+		if o.Loader.ExistsInModelPath(fmt.Sprintf("%s.tmpl", config.Model)) {
+			templateFile = config.Model
+		}
+
+		if config.TemplateConfig.Edit != "" {
+			templateFile = config.TemplateConfig.Edit
+		}
+
+		var result []schema.Choice
+		totalTokenUsage := backend.TokenUsage{}
+
+		for _, i := range config.InputStrings {
+			if templateFile != "" {
+				templatedInput, err := o.Loader.EvaluateTemplateForPrompt(model.EditPromptTemplate, templateFile, model.PromptTemplateData{
+					Input:        i,
+					Instruction:  input.Instruction,
+					SystemPrompt: config.SystemPrompt,
+				})
+				if err == nil {
+					i = templatedInput
+					log.Debug().Msgf("Template found, input modified to: %s", i)
+				}
+			}
+
+			r, tokenUsage, err := ComputeChoices(input, i, config, o, o.Loader, func(s string, c *[]schema.Choice) {
+				*c = append(*c, schema.Choice{Text: s})
+			}, nil)
+			if err != nil {
+				return err
+			}
+
+			totalTokenUsage.Prompt += tokenUsage.Prompt
+			totalTokenUsage.Completion += tokenUsage.Completion
+
+			result = append(result, r...)
+		}
+
+		id := uuid.New().String()
+		created := int(time.Now().Unix())
+		resp := &schema.OpenAIResponse{
+			ID:      id,
+			Created: created,
+			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
+			Choices: result,
+			Object:  "edit",
+			Usage: schema.OpenAIUsage{
+				PromptTokens:     totalTokenUsage.Prompt,
+				CompletionTokens: totalTokenUsage.Completion,
+				TotalTokens:      totalTokenUsage.Prompt + totalTokenUsage.Completion,
+			},
+		}
+
+		jsonResult, _ := json.Marshal(resp)
+		log.Debug().Msgf("Response: %s", jsonResult)
+
+		// Return the prediction in the response body
+		return c.JSON(resp)
+	}
+}

api/openai/embeddings.go

@@ -0,0 +1,78 @@
+package openai
+
+import (
+	"encoding/json"
+	"fmt"
+	"time"
+
+	"github.com/go-skynet/LocalAI/api/backend"
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/schema"
+	"github.com/google/uuid"
+
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/gofiber/fiber/v2"
+	"github.com/rs/zerolog/log"
+)
+
+// https://platform.openai.com/docs/api-reference/embeddings
+func EmbeddingsEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		model, input, err := readInput(c, o, true)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		config, input, err := readConfig(model, input, cm, o.Loader, o.Debug, o.Threads, o.ContextSize, o.F16)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		log.Debug().Msgf("Parameter Config: %+v", config)
+		items := []schema.Item{}
+
+		for i, s := range config.InputToken {
+			// get the model function to call for the result
+			embedFn, err := backend.ModelEmbedding("", s, o.Loader, *config, o)
+			if err != nil {
+				return err
+			}
+
+			embeddings, err := embedFn()
+			if err != nil {
+				return err
+			}
+			items = append(items, schema.Item{Embedding: embeddings, Index: i, Object: "embedding"})
+		}
+
+		for i, s := range config.InputStrings {
+			// get the model function to call for the result
+			embedFn, err := backend.ModelEmbedding(s, []int{}, o.Loader, *config, o)
+			if err != nil {
+				return err
+			}
+
+			embeddings, err := embedFn()
+			if err != nil {
+				return err
+			}
+			items = append(items, schema.Item{Embedding: embeddings, Index: i, Object: "embedding"})
+		}
+
+		id := uuid.New().String()
+		created := int(time.Now().Unix())
+		resp := &schema.OpenAIResponse{
+			ID:      id,
+			Created: created,
+			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
+			Data:    items,
+			Object:  "list",
+		}
+
+		jsonResult, _ := json.Marshal(resp)
+		log.Debug().Msgf("Response: %s", jsonResult)
+
+		// Return the prediction in the response body
+		return c.JSON(resp)
+	}
+}

api/openai/image.go

@@ -0,0 +1,239 @@
+package openai
+
+import (
+	"bufio"
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"os"
+	"path/filepath"
+	"strconv"
+	"strings"
+	"time"
+
+	"github.com/go-skynet/LocalAI/api/schema"
+	"github.com/google/uuid"
+
+	"github.com/go-skynet/LocalAI/api/backend"
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/gofiber/fiber/v2"
+	"github.com/rs/zerolog/log"
+)
+
+func downloadFile(url string) (string, error) {
+	// Get the data
+	resp, err := http.Get(url)
+	if err != nil {
+		return "", err
+	}
+	defer resp.Body.Close()
+
+	// Create the file
+	out, err := os.CreateTemp("", "image")
+	if err != nil {
+		return "", err
+	}
+	defer out.Close()
+
+	// Write the body to file
+	_, err = io.Copy(out, resp.Body)
+	return out.Name(), err
+}
+
+// https://platform.openai.com/docs/api-reference/images/create
+
+/*
+*
+
+	curl http://localhost:8080/v1/images/generations \
+	  -H "Content-Type: application/json" \
+	  -d '{
+	    "prompt": "A cute baby sea otter",
+	    "n": 1,
+	    "size": "512x512"
+	  }'
+
+*
+*/
+func ImageEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		m, input, err := readInput(c, o, false)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		if m == "" {
+			m = model.StableDiffusionBackend
+		}
+		log.Debug().Msgf("Loading model: %+v", m)
+
+		config, input, err := readConfig(m, input, cm, o.Loader, o.Debug, 0, 0, false)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		src := ""
+		if input.File != "" {
+
+			fileData := []byte{}
+			// check if input.File is an URL, if so download it and save it
+			// to a temporary file
+			if strings.HasPrefix(input.File, "http://") || strings.HasPrefix(input.File, "https://") {
+				out, err := downloadFile(input.File)
+				if err != nil {
+					return fmt.Errorf("failed downloading file:%w", err)
+				}
+				defer os.RemoveAll(out)
+
+				fileData, err = os.ReadFile(out)
+				if err != nil {
+					return fmt.Errorf("failed reading file:%w", err)
+				}
+
+			} else {
+				// base 64 decode the file and write it somewhere
+				// that we will cleanup
+				fileData, err = base64.StdEncoding.DecodeString(input.File)
+				if err != nil {
+					return err
+				}
+			}
+
+			// Create a temporary file
+			outputFile, err := os.CreateTemp(o.ImageDir, "b64")
+			if err != nil {
+				return err
+			}
+			// write the base64 result
+			writer := bufio.NewWriter(outputFile)
+			_, err = writer.Write(fileData)
+			if err != nil {
+				outputFile.Close()
+				return err
+			}
+			outputFile.Close()
+			src = outputFile.Name()
+			defer os.RemoveAll(src)
+		}
+
+		log.Debug().Msgf("Parameter Config: %+v", config)
+
+		switch config.Backend {
+		case "stablediffusion":
+			config.Backend = model.StableDiffusionBackend
+		case "tinydream":
+			config.Backend = model.TinyDreamBackend
+		case "":
+			config.Backend = model.StableDiffusionBackend
+		}
+
+		sizeParts := strings.Split(input.Size, "x")
+		if len(sizeParts) != 2 {
+			return fmt.Errorf("Invalid value for 'size'")
+		}
+		width, err := strconv.Atoi(sizeParts[0])
+		if err != nil {
+			return fmt.Errorf("Invalid value for 'size'")
+		}
+		height, err := strconv.Atoi(sizeParts[1])
+		if err != nil {
+			return fmt.Errorf("Invalid value for 'size'")
+		}
+
+		b64JSON := false
+		if input.ResponseFormat.Type == "b64_json" {
+			b64JSON = true
+		}
+		// src and clip_skip
+		var result []schema.Item
+		for _, i := range config.PromptStrings {
+			n := input.N
+			if input.N == 0 {
+				n = 1
+			}
+			for j := 0; j < n; j++ {
+				prompts := strings.Split(i, "|")
+				positive_prompt := prompts[0]
+				negative_prompt := ""
+				if len(prompts) > 1 {
+					negative_prompt = prompts[1]
+				}
+
+				mode := 0
+				step := config.Step
+				if step == 0 {
+					step = 15
+				}
+
+				if input.Mode != 0 {
+					mode = input.Mode
+				}
+
+				if input.Step != 0 {
+					step = input.Step
+				}
+
+				tempDir := ""
+				if !b64JSON {
+					tempDir = o.ImageDir
+				}
+				// Create a temporary file
+				outputFile, err := os.CreateTemp(tempDir, "b64")
+				if err != nil {
+					return err
+				}
+				outputFile.Close()
+				output := outputFile.Name() + ".png"
+				// Rename the temporary file
+				err = os.Rename(outputFile.Name(), output)
+				if err != nil {
+					return err
+				}
+
+				baseURL := c.BaseURL()
+
+				fn, err := backend.ImageGeneration(height, width, mode, step, input.Seed, positive_prompt, negative_prompt, src, output, o.Loader, *config, o)
+				if err != nil {
+					return err
+				}
+				if err := fn(); err != nil {
+					return err
+				}
+
+				item := &schema.Item{}
+
+				if b64JSON {
+					defer os.RemoveAll(output)
+					data, err := os.ReadFile(output)
+					if err != nil {
+						return err
+					}
+					item.B64JSON = base64.StdEncoding.EncodeToString(data)
+				} else {
+					base := filepath.Base(output)
+					item.URL = baseURL + "/generated-images/" + base
+				}
+
+				result = append(result, *item)
+			}
+		}
+
+		id := uuid.New().String()
+		created := int(time.Now().Unix())
+		resp := &schema.OpenAIResponse{
+			ID:      id,
+			Created: created,
+			Data:    result,
+		}
+
+		jsonResult, _ := json.Marshal(resp)
+		log.Debug().Msgf("Response: %s", jsonResult)
+
+		// Return the prediction in the response body
+		return c.JSON(resp)
+	}
+}

api/openai/inference.go

@@ -0,0 +1,55 @@
+package openai
+
+import (
+	"github.com/go-skynet/LocalAI/api/backend"
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/api/schema"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+)
+
+func ComputeChoices(
+	req *schema.OpenAIRequest,
+	predInput string,
+	config *config.Config,
+	o *options.Option,
+	loader *model.ModelLoader,
+	cb func(string, *[]schema.Choice),
+	tokenCallback func(string, backend.TokenUsage) bool) ([]schema.Choice, backend.TokenUsage, error) {
+	n := req.N // number of completions to return
+	result := []schema.Choice{}
+
+	if n == 0 {
+		n = 1
+	}
+
+	images := []string{}
+	for _, m := range req.Messages {
+		images = append(images, m.StringImages...)
+	}
+
+	// get the model function to call for the result
+	predFunc, err := backend.ModelInference(req.Context, predInput, images, loader, *config, o, tokenCallback)
+	if err != nil {
+		return result, backend.TokenUsage{}, err
+	}
+
+	tokenUsage := backend.TokenUsage{}
+
+	for i := 0; i < n; i++ {
+		prediction, err := predFunc()
+		if err != nil {
+			return result, backend.TokenUsage{}, err
+		}
+
+		tokenUsage.Prompt += prediction.Usage.Prompt
+		tokenUsage.Completion += prediction.Usage.Completion
+
+		finetunedResponse := backend.Finetune(*config, predInput, prediction.Response)
+		cb(finetunedResponse, &result)
+
+		//result = append(result, Choice{Text: prediction})
+
+	}
+	return result, tokenUsage, err
+}

api/openai/list.go

@@ -0,0 +1,69 @@
+package openai
+
+import (
+	"regexp"
+
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/schema"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/gofiber/fiber/v2"
+)
+
+func ListModelsEndpoint(loader *model.ModelLoader, cm *config.ConfigLoader) func(ctx *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		models, err := loader.ListModels()
+		if err != nil {
+			return err
+		}
+		var mm map[string]interface{} = map[string]interface{}{}
+
+		dataModels := []schema.OpenAIModel{}
+
+		var filterFn func(name string) bool
+		filter := c.Query("filter")
+
+		// If filter is not specified, do not filter the list by model name
+		if filter == "" {
+			filterFn = func(_ string) bool { return true }
+		} else {
+			// If filter _IS_ specified, we compile it to a regex which is used to create the filterFn
+			rxp, err := regexp.Compile(filter)
+			if err != nil {
+				return err
+			}
+			filterFn = func(name string) bool {
+				return rxp.MatchString(name)
+			}
+		}
+
+		// By default, exclude any loose files that are already referenced by a configuration file.
+		excludeConfigured := c.QueryBool("excludeConfigured", true)
+
+		// Start with the known configurations
+		for _, c := range cm.GetAllConfigs() {
+			if excludeConfigured {
+				mm[c.Model] = nil
+			}
+
+			if filterFn(c.Name) {
+				dataModels = append(dataModels, schema.OpenAIModel{ID: c.Name, Object: "model"})
+			}
+		}
+
+		// Then iterate through the loose files:
+		for _, m := range models {
+			// And only adds them if they shouldn't be skipped.
+			if _, exists := mm[m]; !exists && filterFn(m) {
+				dataModels = append(dataModels, schema.OpenAIModel{ID: m, Object: "model"})
+			}
+		}
+
+		return c.JSON(struct {
+			Object string               `json:"object"`
+			Data   []schema.OpenAIModel `json:"data"`
+		}{
+			Object: "list",
+			Data:   dataModels,
+		})
+	}
+}

api/openai/request.go

@@ -0,0 +1,336 @@
+package openai
+
+import (
+	"context"
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"io/ioutil"
+	"net/http"
+	"os"
+	"path/filepath"
+	"strings"
+
+	config "github.com/go-skynet/LocalAI/api/config"
+	options "github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/api/schema"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/gofiber/fiber/v2"
+	"github.com/rs/zerolog/log"
+)
+
+func readInput(c *fiber.Ctx, o *options.Option, randomModel bool) (string, *schema.OpenAIRequest, error) {
+	loader := o.Loader
+	input := new(schema.OpenAIRequest)
+	ctx, cancel := context.WithCancel(o.Context)
+	input.Context = ctx
+	input.Cancel = cancel
+	// Get input data from the request body
+	if err := c.BodyParser(input); err != nil {
+		return "", nil, fmt.Errorf("failed parsing request body: %w", err)
+	}
+
+	modelFile := input.Model
+
+	if c.Params("model") != "" {
+		modelFile = c.Params("model")
+	}
+
+	received, _ := json.Marshal(input)
+
+	log.Debug().Msgf("Request received: %s", string(received))
+
+	// Set model from bearer token, if available
+	bearer := strings.TrimLeft(c.Get("authorization"), "Bearer ")
+	bearerExists := bearer != "" && loader.ExistsInModelPath(bearer)
+
+	// If no model was specified, take the first available
+	if modelFile == "" && !bearerExists && randomModel {
+		models, _ := loader.ListModels()
+		if len(models) > 0 {
+			modelFile = models[0]
+			log.Debug().Msgf("No model specified, using: %s", modelFile)
+		} else {
+			log.Debug().Msgf("No model specified, returning error")
+			return "", nil, fmt.Errorf("no model specified")
+		}
+	}
+
+	// If a model is found in bearer token takes precedence
+	if bearerExists {
+		log.Debug().Msgf("Using model from bearer token: %s", bearer)
+		modelFile = bearer
+	}
+	return modelFile, input, nil
+}
+
+// this function check if the string is an URL, if it's an URL downloads the image in memory
+// encodes it in base64 and returns the base64 string
+func getBase64Image(s string) (string, error) {
+	if strings.HasPrefix(s, "http") {
+		// download the image
+		resp, err := http.Get(s)
+		if err != nil {
+			return "", err
+		}
+		defer resp.Body.Close()
+
+		// read the image data into memory
+		data, err := ioutil.ReadAll(resp.Body)
+		if err != nil {
+			return "", err
+		}
+
+		// encode the image data in base64
+		encoded := base64.StdEncoding.EncodeToString(data)
+
+		// return the base64 string
+		return encoded, nil
+	}
+
+	// if the string instead is prefixed with "data:image/jpeg;base64,", drop it
+	if strings.HasPrefix(s, "data:image/jpeg;base64,") {
+		return strings.ReplaceAll(s, "data:image/jpeg;base64,", ""), nil
+	}
+	return "", fmt.Errorf("not valid string")
+}
+
+func updateConfig(config *config.Config, input *schema.OpenAIRequest) {
+	if input.Echo {
+		config.Echo = input.Echo
+	}
+	if input.TopK != 0 {
+		config.TopK = input.TopK
+	}
+	if input.TopP != 0 {
+		config.TopP = input.TopP
+	}
+
+	if input.Backend != "" {
+		config.Backend = input.Backend
+	}
+
+	if input.ClipSkip != 0 {
+		config.Diffusers.ClipSkip = input.ClipSkip
+	}
+
+	if input.ModelBaseName != "" {
+		config.AutoGPTQ.ModelBaseName = input.ModelBaseName
+	}
+
+	if input.NegativePromptScale != 0 {
+		config.NegativePromptScale = input.NegativePromptScale
+	}
+
+	if input.UseFastTokenizer {
+		config.UseFastTokenizer = input.UseFastTokenizer
+	}
+
+	if input.NegativePrompt != "" {
+		config.NegativePrompt = input.NegativePrompt
+	}
+
+	if input.RopeFreqBase != 0 {
+		config.RopeFreqBase = input.RopeFreqBase
+	}
+
+	if input.RopeFreqScale != 0 {
+		config.RopeFreqScale = input.RopeFreqScale
+	}
+
+	if input.Grammar != "" {
+		config.Grammar = input.Grammar
+	}
+
+	if input.Temperature != 0 {
+		config.Temperature = input.Temperature
+	}
+
+	if input.Maxtokens != 0 {
+		config.Maxtokens = input.Maxtokens
+	}
+
+	switch stop := input.Stop.(type) {
+	case string:
+		if stop != "" {
+			config.StopWords = append(config.StopWords, stop)
+		}
+	case []interface{}:
+		for _, pp := range stop {
+			if s, ok := pp.(string); ok {
+				config.StopWords = append(config.StopWords, s)
+			}
+		}
+	}
+
+	// Decode each request's message content
+	index := 0
+	for i, m := range input.Messages {
+		switch content := m.Content.(type) {
+		case string:
+			input.Messages[i].StringContent = content
+		case []interface{}:
+			dat, _ := json.Marshal(content)
+			c := []schema.Content{}
+			json.Unmarshal(dat, &c)
+			for _, pp := range c {
+				if pp.Type == "text" {
+					input.Messages[i].StringContent = pp.Text
+				} else if pp.Type == "image_url" {
+					// Detect if pp.ImageURL is an URL, if it is download the image and encode it in base64:
+					base64, err := getBase64Image(pp.ImageURL.URL)
+					if err == nil {
+						input.Messages[i].StringImages = append(input.Messages[i].StringImages, base64) // TODO: make sure that we only return base64 stuff
+						// set a placeholder for each image
+						input.Messages[i].StringContent = fmt.Sprintf("[img-%d]", index) + input.Messages[i].StringContent
+						index++
+					} else {
+						fmt.Print("Failed encoding image", err)
+					}
+				}
+			}
+		}
+	}
+
+	if input.RepeatPenalty != 0 {
+		config.RepeatPenalty = input.RepeatPenalty
+	}
+
+	if input.Keep != 0 {
+		config.Keep = input.Keep
+	}
+
+	if input.Batch != 0 {
+		config.Batch = input.Batch
+	}
+
+	if input.F16 {
+		config.F16 = input.F16
+	}
+
+	if input.IgnoreEOS {
+		config.IgnoreEOS = input.IgnoreEOS
+	}
+
+	if input.Seed != 0 {
+		config.Seed = input.Seed
+	}
+
+	if input.Mirostat != 0 {
+		config.LLMConfig.Mirostat = input.Mirostat
+	}
+
+	if input.MirostatETA != 0 {
+		config.LLMConfig.MirostatETA = input.MirostatETA
+	}
+
+	if input.MirostatTAU != 0 {
+		config.LLMConfig.MirostatTAU = input.MirostatTAU
+	}
+
+	if input.TypicalP != 0 {
+		config.TypicalP = input.TypicalP
+	}
+
+	switch inputs := input.Input.(type) {
+	case string:
+		if inputs != "" {
+			config.InputStrings = append(config.InputStrings, inputs)
+		}
+	case []interface{}:
+		for _, pp := range inputs {
+			switch i := pp.(type) {
+			case string:
+				config.InputStrings = append(config.InputStrings, i)
+			case []interface{}:
+				tokens := []int{}
+				for _, ii := range i {
+					tokens = append(tokens, int(ii.(float64)))
+				}
+				config.InputToken = append(config.InputToken, tokens)
+			}
+		}
+	}
+
+	// Can be either a string or an object
+	switch fnc := input.FunctionCall.(type) {
+	case string:
+		if fnc != "" {
+			config.SetFunctionCallString(fnc)
+		}
+	case map[string]interface{}:
+		var name string
+		n, exists := fnc["name"]
+		if exists {
+			nn, e := n.(string)
+			if e {
+				name = nn
+			}
+		}
+		config.SetFunctionCallNameString(name)
+	}
+
+	switch p := input.Prompt.(type) {
+	case string:
+		config.PromptStrings = append(config.PromptStrings, p)
+	case []interface{}:
+		for _, pp := range p {
+			if s, ok := pp.(string); ok {
+				config.PromptStrings = append(config.PromptStrings, s)
+			}
+		}
+	}
+}
+
+func readConfig(modelFile string, input *schema.OpenAIRequest, cm *config.ConfigLoader, loader *model.ModelLoader, debug bool, threads, ctx int, f16 bool) (*config.Config, *schema.OpenAIRequest, error) {
+	// Load a config file if present after the model name
+	modelConfig := filepath.Join(loader.ModelPath, modelFile+".yaml")
+
+	var cfg *config.Config
+
+	defaults := func() {
+		cfg = config.DefaultConfig(modelFile)
+		cfg.ContextSize = ctx
+		cfg.Threads = threads
+		cfg.F16 = f16
+		cfg.Debug = debug
+	}
+
+	cfgExisting, exists := cm.GetConfig(modelFile)
+	if !exists {
+		if _, err := os.Stat(modelConfig); err == nil {
+			if err := cm.LoadConfig(modelConfig); err != nil {
+				return nil, nil, fmt.Errorf("failed loading model config (%s) %s", modelConfig, err.Error())
+			}
+			cfgExisting, exists = cm.GetConfig(modelFile)
+			if exists {
+				cfg = &cfgExisting
+			} else {
+				defaults()
+			}
+		} else {
+			defaults()
+		}
+	} else {
+		cfg = &cfgExisting
+	}
+
+	// Set the parameters for the language model prediction
+	updateConfig(cfg, input)
+
+	// Don't allow 0 as setting
+	if cfg.Threads == 0 {
+		if threads != 0 {
+			cfg.Threads = threads
+		} else {
+			cfg.Threads = 4
+		}
+	}
+
+	// Enforce debug flag if passed from CLI
+	if debug {
+		cfg.Debug = true
+	}
+
+	return cfg, input, nil
+}

api/openai/transcription.go

@@ -0,0 +1,71 @@
+package openai
+
+import (
+	"fmt"
+	"io"
+	"net/http"
+	"os"
+	"path"
+	"path/filepath"
+
+	"github.com/go-skynet/LocalAI/api/backend"
+	config "github.com/go-skynet/LocalAI/api/config"
+	"github.com/go-skynet/LocalAI/api/options"
+
+	"github.com/gofiber/fiber/v2"
+	"github.com/rs/zerolog/log"
+)
+
+// https://platform.openai.com/docs/api-reference/audio/create
+func TranscriptEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		m, input, err := readInput(c, o, false)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		config, input, err := readConfig(m, input, cm, o.Loader, o.Debug, o.Threads, o.ContextSize, o.F16)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+		// retrieve the file data from the request
+		file, err := c.FormFile("file")
+		if err != nil {
+			return err
+		}
+		f, err := file.Open()
+		if err != nil {
+			return err
+		}
+		defer f.Close()
+
+		dir, err := os.MkdirTemp("", "whisper")
+
+		if err != nil {
+			return err
+		}
+		defer os.RemoveAll(dir)
+
+		dst := filepath.Join(dir, path.Base(file.Filename))
+		dstFile, err := os.Create(dst)
+		if err != nil {
+			return err
+		}
+
+		if _, err := io.Copy(dstFile, f); err != nil {
+			log.Debug().Msgf("Audio file copying error %+v - %+v - err %+v", file.Filename, dst, err)
+			return err
+		}
+
+		log.Debug().Msgf("Audio file copied to: %+v", dst)
+
+		tr, err := backend.ModelTranscription(dst, input.Language, o.Loader, *config, o)
+		if err != nil {
+			return err
+		}
+
+		log.Debug().Msgf("Trascribed: %+v", tr)
+		// TODO: handle different outputs here
+		return c.Status(http.StatusOK).JSON(tr)
+	}
+}

api/options/options.go

@@ -0,0 +1,262 @@
+package options
+
+import (
+	"context"
+	"embed"
+	"encoding/json"
+	"time"
+
+	"github.com/go-skynet/LocalAI/metrics"
+	"github.com/go-skynet/LocalAI/pkg/gallery"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/rs/zerolog/log"
+)
+
+type Option struct {
+	Context                             context.Context
+	ConfigFile                          string
+	Loader                              *model.ModelLoader
+	UploadLimitMB, Threads, ContextSize int
+	F16                                 bool
+	Debug, DisableMessage               bool
+	ImageDir                            string
+	AudioDir                            string
+	CORS                                bool
+	PreloadJSONModels                   string
+	PreloadModelsFromPath               string
+	CORSAllowOrigins                    string
+	ApiKeys                             []string
+	Metrics                             *metrics.Metrics
+
+	ModelLibraryURL string
+
+	Galleries []gallery.Gallery
+
+	BackendAssets     embed.FS
+	AssetsDestination string
+
+	ExternalGRPCBackends map[string]string
+
+	AutoloadGalleries bool
+
+	SingleBackend           bool
+	ParallelBackendRequests bool
+
+	WatchDogIdle bool
+	WatchDogBusy bool
+	WatchDog     bool
+
+	ModelsURL []string
+
+	WatchDogBusyTimeout, WatchDogIdleTimeout time.Duration
+}
+
+type AppOption func(*Option)
+
+func NewOptions(o ...AppOption) *Option {
+	opt := &Option{
+		Context:        context.Background(),
+		UploadLimitMB:  15,
+		Threads:        1,
+		ContextSize:    512,
+		Debug:          true,
+		DisableMessage: true,
+	}
+	for _, oo := range o {
+		oo(opt)
+	}
+	return opt
+}
+
+func WithModelsURL(urls ...string) AppOption {
+	return func(o *Option) {
+		o.ModelsURL = urls
+	}
+}
+
+func WithCors(b bool) AppOption {
+	return func(o *Option) {
+		o.CORS = b
+	}
+}
+
+func WithModelLibraryURL(url string) AppOption {
+	return func(o *Option) {
+		o.ModelLibraryURL = url
+	}
+}
+
+var EnableWatchDog = func(o *Option) {
+	o.WatchDog = true
+}
+
+var EnableWatchDogIdleCheck = func(o *Option) {
+	o.WatchDog = true
+	o.WatchDogIdle = true
+}
+
+var EnableWatchDogBusyCheck = func(o *Option) {
+	o.WatchDog = true
+	o.WatchDogBusy = true
+}
+
+func SetWatchDogBusyTimeout(t time.Duration) AppOption {
+	return func(o *Option) {
+		o.WatchDogBusyTimeout = t
+	}
+}
+
+func SetWatchDogIdleTimeout(t time.Duration) AppOption {
+	return func(o *Option) {
+		o.WatchDogIdleTimeout = t
+	}
+}
+
+var EnableSingleBackend = func(o *Option) {
+	o.SingleBackend = true
+}
+
+var EnableParallelBackendRequests = func(o *Option) {
+	o.ParallelBackendRequests = true
+}
+
+var EnableGalleriesAutoload = func(o *Option) {
+	o.AutoloadGalleries = true
+}
+
+func WithExternalBackend(name string, uri string) AppOption {
+	return func(o *Option) {
+		if o.ExternalGRPCBackends == nil {
+			o.ExternalGRPCBackends = make(map[string]string)
+		}
+		o.ExternalGRPCBackends[name] = uri
+	}
+}
+
+func WithCorsAllowOrigins(b string) AppOption {
+	return func(o *Option) {
+		o.CORSAllowOrigins = b
+	}
+}
+
+func WithBackendAssetsOutput(out string) AppOption {
+	return func(o *Option) {
+		o.AssetsDestination = out
+	}
+}
+
+func WithBackendAssets(f embed.FS) AppOption {
+	return func(o *Option) {
+		o.BackendAssets = f
+	}
+}
+
+func WithStringGalleries(galls string) AppOption {
+	return func(o *Option) {
+		if galls == "" {
+			log.Debug().Msgf("no galleries to load")
+			o.Galleries = []gallery.Gallery{}
+			return
+		}
+		var galleries []gallery.Gallery
+		if err := json.Unmarshal([]byte(galls), &galleries); err != nil {
+			log.Error().Msgf("failed loading galleries: %s", err.Error())
+		}
+		o.Galleries = append(o.Galleries, galleries...)
+	}
+}
+
+func WithGalleries(galleries []gallery.Gallery) AppOption {
+	return func(o *Option) {
+		o.Galleries = append(o.Galleries, galleries...)
+	}
+}
+
+func WithContext(ctx context.Context) AppOption {
+	return func(o *Option) {
+		o.Context = ctx
+	}
+}
+
+func WithYAMLConfigPreload(configFile string) AppOption {
+	return func(o *Option) {
+		o.PreloadModelsFromPath = configFile
+	}
+}
+
+func WithJSONStringPreload(configFile string) AppOption {
+	return func(o *Option) {
+		o.PreloadJSONModels = configFile
+	}
+}
+func WithConfigFile(configFile string) AppOption {
+	return func(o *Option) {
+		o.ConfigFile = configFile
+	}
+}
+
+func WithModelLoader(loader *model.ModelLoader) AppOption {
+	return func(o *Option) {
+		o.Loader = loader
+	}
+}
+
+func WithUploadLimitMB(limit int) AppOption {
+	return func(o *Option) {
+		o.UploadLimitMB = limit
+	}
+}
+
+func WithThreads(threads int) AppOption {
+	return func(o *Option) {
+		o.Threads = threads
+	}
+}
+
+func WithContextSize(ctxSize int) AppOption {
+	return func(o *Option) {
+		o.ContextSize = ctxSize
+	}
+}
+
+func WithF16(f16 bool) AppOption {
+	return func(o *Option) {
+		o.F16 = f16
+	}
+}
+
+func WithDebug(debug bool) AppOption {
+	return func(o *Option) {
+		o.Debug = debug
+	}
+}
+
+func WithDisableMessage(disableMessage bool) AppOption {
+	return func(o *Option) {
+		o.DisableMessage = disableMessage
+	}
+}
+
+func WithAudioDir(audioDir string) AppOption {
+	return func(o *Option) {
+		o.AudioDir = audioDir
+	}
+}
+
+func WithImageDir(imageDir string) AppOption {
+	return func(o *Option) {
+		o.ImageDir = imageDir
+	}
+}
+
+func WithApiKeys(apiKeys []string) AppOption {
+	return func(o *Option) {
+		o.ApiKeys = apiKeys
+	}
+}
+
+func WithMetrics(meter *metrics.Metrics) AppOption {
+	return func(o *Option) {
+		o.Metrics = meter
+	}
+}

api/schema/openai.go

@@ -0,0 +1,135 @@
+package schema
+
+import (
+	"context"
+
+	config "github.com/go-skynet/LocalAI/api/config"
+
+	"github.com/go-skynet/LocalAI/pkg/grammar"
+)
+
+// APIError provides error information returned by the OpenAI API.
+type APIError struct {
+	Code    any     `json:"code,omitempty"`
+	Message string  `json:"message"`
+	Param   *string `json:"param,omitempty"`
+	Type    string  `json:"type"`
+}
+
+type ErrorResponse struct {
+	Error *APIError `json:"error,omitempty"`
+}
+
+type OpenAIUsage struct {
+	PromptTokens     int `json:"prompt_tokens"`
+	CompletionTokens int `json:"completion_tokens"`
+	TotalTokens      int `json:"total_tokens"`
+}
+
+type Item struct {
+	Embedding []float32 `json:"embedding"`
+	Index     int       `json:"index"`
+	Object    string    `json:"object,omitempty"`
+
+	// Images
+	URL     string `json:"url,omitempty"`
+	B64JSON string `json:"b64_json,omitempty"`
+}
+
+type OpenAIResponse struct {
+	Created int      `json:"created,omitempty"`
+	Object  string   `json:"object,omitempty"`
+	ID      string   `json:"id,omitempty"`
+	Model   string   `json:"model,omitempty"`
+	Choices []Choice `json:"choices,omitempty"`
+	Data    []Item   `json:"data,omitempty"`
+
+	Usage OpenAIUsage `json:"usage"`
+}
+
+type Choice struct {
+	Index        int      `json:"index"`
+	FinishReason string   `json:"finish_reason,omitempty"`
+	Message      *Message `json:"message,omitempty"`
+	Delta        *Message `json:"delta,omitempty"`
+	Text         string   `json:"text,omitempty"`
+}
+
+type Content struct {
+	Type     string     `json:"type" yaml:"type"`
+	Text     string     `json:"text" yaml:"text"`
+	ImageURL ContentURL `json:"image_url" yaml:"image_url"`
+}
+
+type ContentURL struct {
+	URL string `json:"url" yaml:"url"`
+}
+
+type Message struct {
+	// The message role
+	Role string `json:"role,omitempty" yaml:"role"`
+	// The message content
+	Content interface{} `json:"content" yaml:"content"`
+
+	StringContent string   `json:"string_content,omitempty" yaml:"string_content,omitempty"`
+	StringImages  []string `json:"string_images,omitempty" yaml:"string_images,omitempty"`
+
+	// A result of a function call
+	FunctionCall interface{} `json:"function_call,omitempty" yaml:"function_call,omitempty"`
+}
+
+type OpenAIModel struct {
+	ID     string `json:"id"`
+	Object string `json:"object"`
+}
+
+type ChatCompletionResponseFormatType string
+
+type ChatCompletionResponseFormat struct {
+	Type ChatCompletionResponseFormatType `json:"type,omitempty"`
+}
+
+type OpenAIRequest struct {
+	config.PredictionOptions
+
+	Context context.Context
+	Cancel  context.CancelFunc
+
+	// whisper
+	File string `json:"file" validate:"required"`
+	//whisper/image
+	ResponseFormat ChatCompletionResponseFormat `json:"response_format"`
+	// image
+	Size string `json:"size"`
+	// Prompt is read only by completion/image API calls
+	Prompt interface{} `json:"prompt" yaml:"prompt"`
+
+	// Edit endpoint
+	Instruction string      `json:"instruction" yaml:"instruction"`
+	Input       interface{} `json:"input" yaml:"input"`
+
+	Stop interface{} `json:"stop" yaml:"stop"`
+
+	// Messages is read only by chat/completion API calls
+	Messages []Message `json:"messages" yaml:"messages"`
+
+	// A list of available functions to call
+	Functions    []grammar.Function `json:"functions" yaml:"functions"`
+	FunctionCall interface{}        `json:"function_call" yaml:"function_call"` // might be a string or an object
+
+	Stream bool `json:"stream"`
+
+	// Image (not supported by OpenAI)
+	Mode int `json:"mode"`
+	Step int `json:"step"`
+
+	// A grammar to constrain the LLM output
+	Grammar string `json:"grammar" yaml:"grammar"`
+
+	JSONFunctionGrammarObject *grammar.JSONFunctionStructure `json:"grammar_json_functions" yaml:"grammar_json_functions"`
+
+	Backend string `json:"backend" yaml:"backend"`
+
+	// AutoGPTQ
+	ModelBaseName string `json:"model_base_name" yaml:"model_base_name"`
+}

api/schema/whisper.go

@@ -0,0 +1,16 @@
+package schema
+
+import "time"
+
+type Segment struct {
+	Id     int           `json:"id"`
+	Start  time.Duration `json:"start"`
+	End    time.Duration `json:"end"`
+	Text   string        `json:"text"`
+	Tokens []int         `json:"tokens"`
+}
+
+type Result struct {
+	Segments []Segment `json:"segments"`
+	Text     string    `json:"text"`
+}

assets.go

@@ -0,0 +1,6 @@
+package main
+
+import "embed"
+
+//go:embed backend-assets/*
+var backendAssets embed.FS

backend/backend.proto

@@ -0,0 +1,210 @@
+syntax = "proto3";
+
+option go_package = "github.com/go-skynet/LocalAI/pkg/grpc/proto";
+option java_multiple_files = true;
+option java_package = "io.skynet.localai.backend";
+option java_outer_classname = "LocalAIBackend";
+
+package backend;
+
+service Backend {
+  rpc Health(HealthMessage) returns (Reply) {}
+  rpc Predict(PredictOptions) returns (Reply) {}
+  rpc LoadModel(ModelOptions) returns (Result) {}
+  rpc PredictStream(PredictOptions) returns (stream Reply) {}
+  rpc Embedding(PredictOptions) returns (EmbeddingResult) {}
+  rpc GenerateImage(GenerateImageRequest) returns (Result) {}
+  rpc AudioTranscription(TranscriptRequest) returns (TranscriptResult) {}
+  rpc TTS(TTSRequest) returns (Result) {}
+  rpc TokenizeString(PredictOptions) returns (TokenizationResponse) {}
+  rpc Status(HealthMessage) returns (StatusResponse) {}
+}
+
+message HealthMessage {}
+
+// The request message containing the user's name.
+message PredictOptions {
+  string Prompt = 1;
+  int32 Seed = 2;
+  int32 Threads = 3;
+  int32 Tokens = 4;
+  int32 TopK = 5;
+  int32 Repeat = 6;
+  int32 Batch = 7;
+  int32 NKeep = 8;
+  float Temperature = 9;
+  float Penalty = 10;
+  bool F16KV = 11;
+  bool DebugMode = 12;
+  repeated string StopPrompts = 13;
+  bool IgnoreEOS = 14;
+  float TailFreeSamplingZ = 15;
+  float TypicalP = 16;
+  float FrequencyPenalty = 17;
+  float PresencePenalty = 18;
+  int32 Mirostat = 19;
+  float MirostatETA = 20;
+  float MirostatTAU = 21;
+  bool PenalizeNL = 22;
+  string LogitBias = 23;
+  bool MLock = 25;
+  bool MMap = 26;
+  bool PromptCacheAll = 27;
+  bool PromptCacheRO = 28;
+  string Grammar = 29;
+  string MainGPU = 30;
+  string TensorSplit = 31;
+  float TopP = 32;
+  string PromptCachePath = 33;
+  bool Debug = 34;
+  repeated int32 EmbeddingTokens = 35;
+  string Embeddings = 36;
+  float RopeFreqBase = 37;
+  float RopeFreqScale = 38;
+  float NegativePromptScale = 39;
+  string NegativePrompt = 40;
+  int32 NDraft = 41;
+  repeated string Images = 42;
+}
+
+// The response message containing the result
+message Reply {
+  bytes message = 1;
+}
+
+message ModelOptions {
+  string Model = 1;
+  int32 ContextSize = 2;
+  int32 Seed = 3;
+  int32 NBatch = 4;
+  bool F16Memory = 5;
+  bool MLock = 6;
+  bool MMap = 7;
+  bool VocabOnly = 8;
+  bool LowVRAM = 9;
+  bool Embeddings = 10;
+  bool NUMA = 11;
+  int32 NGPULayers = 12;
+  string MainGPU = 13;
+  string TensorSplit = 14;
+  int32 Threads = 15;
+  string LibrarySearchPath = 16;
+  float RopeFreqBase = 17;
+  float RopeFreqScale = 18;
+  float RMSNormEps = 19;
+  int32 NGQA = 20;
+  string ModelFile = 21;
+
+  // AutoGPTQ
+  string Device = 22;
+  bool UseTriton = 23;
+  string ModelBaseName = 24;
+  bool UseFastTokenizer = 25;
+
+  // Diffusers
+  string PipelineType = 26;
+  string SchedulerType = 27;
+  bool CUDA = 28;
+  float CFGScale = 29;
+  bool IMG2IMG = 30;
+  string CLIPModel = 31;
+  string CLIPSubfolder = 32;
+  int32 CLIPSkip = 33;
+  string ControlNet = 48;
+
+  string Tokenizer = 34;
+
+  // LLM (llama.cpp)
+  string LoraBase = 35;
+  string LoraAdapter = 36;
+  float LoraScale = 42;
+
+  bool NoMulMatQ = 37;
+  string DraftModel = 39;
+  
+  string AudioPath = 38;
+
+  // vllm
+  string Quantization = 40;
+
+  string MMProj = 41;
+
+  string RopeScaling = 43;
+  float YarnExtFactor = 44;
+  float YarnAttnFactor = 45;
+  float YarnBetaFast = 46;
+  float YarnBetaSlow = 47;
+
+  string Type = 49;
+}
+
+message Result {
+  string message = 1;
+  bool success = 2;
+}
+
+message EmbeddingResult {
+  repeated float embeddings = 1;
+}
+
+message TranscriptRequest {
+  string dst = 2;
+  string language = 3;
+  uint32 threads = 4;
+}
+
+message TranscriptResult {
+  repeated TranscriptSegment segments = 1;
+  string text = 2;
+}
+
+message TranscriptSegment {
+  int32 id = 1;
+  int64 start = 2;
+  int64 end = 3;
+  string text = 4;
+  repeated int32 tokens = 5;
+}
+
+message GenerateImageRequest {
+  int32 height = 1;
+  int32 width = 2;
+  int32 mode = 3;
+  int32 step = 4;
+  int32 seed = 5;
+  string positive_prompt = 6;
+  string negative_prompt = 7;
+  string dst = 8;
+  string src = 9;
+
+  // Diffusers
+  string EnableParameters = 10;
+  int32 CLIPSkip = 11;
+}
+
+message TTSRequest {
+  string text = 1;
+  string model = 2;
+  string dst = 3;
+}
+
+message TokenizationResponse {
+  int32 length = 1;
+  repeated int32 tokens = 2;
+}
+
+message MemoryUsageData {
+  uint64 total = 1;
+  map<string, uint64> breakdown = 2;
+}
+
+message StatusResponse {
+  enum State {
+    UNINITIALIZED = 0;
+    BUSY = 1;
+    READY = 2;
+    ERROR = -1;
+  }
+  State state = 1;
+  MemoryUsageData memory = 2;
+}

backend/backend_grpc.pb.go

@@ -0,0 +1,457 @@
+// Code generated by protoc-gen-go-grpc. DO NOT EDIT.
+// versions:
+// - protoc-gen-go-grpc v1.2.0
+// - protoc             v4.23.4
+// source: backend/backend.proto
+
+package proto
+
+import (
+	context "context"
+	grpc "google.golang.org/grpc"
+	codes "google.golang.org/grpc/codes"
+	status "google.golang.org/grpc/status"
+)
+
+// This is a compile-time assertion to ensure that this generated file
+// is compatible with the grpc package it is being compiled against.
+// Requires gRPC-Go v1.32.0 or later.
+const _ = grpc.SupportPackageIsVersion7
+
+// BackendClient is the client API for Backend service.
+//
+// For semantics around ctx use and closing/ending streaming RPCs, please refer to https://pkg.go.dev/google.golang.org/grpc/?tab=doc#ClientConn.NewStream.
+type BackendClient interface {
+	Health(ctx context.Context, in *HealthMessage, opts ...grpc.CallOption) (*Reply, error)
+	Predict(ctx context.Context, in *PredictOptions, opts ...grpc.CallOption) (*Reply, error)
+	LoadModel(ctx context.Context, in *ModelOptions, opts ...grpc.CallOption) (*Result, error)
+	PredictStream(ctx context.Context, in *PredictOptions, opts ...grpc.CallOption) (Backend_PredictStreamClient, error)
+	Embedding(ctx context.Context, in *PredictOptions, opts ...grpc.CallOption) (*EmbeddingResult, error)
+	GenerateImage(ctx context.Context, in *GenerateImageRequest, opts ...grpc.CallOption) (*Result, error)
+	AudioTranscription(ctx context.Context, in *TranscriptRequest, opts ...grpc.CallOption) (*TranscriptResult, error)
+	TTS(ctx context.Context, in *TTSRequest, opts ...grpc.CallOption) (*Result, error)
+	TokenizeString(ctx context.Context, in *PredictOptions, opts ...grpc.CallOption) (*TokenizationResponse, error)
+	Status(ctx context.Context, in *HealthMessage, opts ...grpc.CallOption) (*StatusResponse, error)
+}
+
+type backendClient struct {
+	cc grpc.ClientConnInterface
+}
+
+func NewBackendClient(cc grpc.ClientConnInterface) BackendClient {
+	return &backendClient{cc}
+}
+
+func (c *backendClient) Health(ctx context.Context, in *HealthMessage, opts ...grpc.CallOption) (*Reply, error) {
+	out := new(Reply)
+	err := c.cc.Invoke(ctx, "/backend.Backend/Health", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *backendClient) Predict(ctx context.Context, in *PredictOptions, opts ...grpc.CallOption) (*Reply, error) {
+	out := new(Reply)
+	err := c.cc.Invoke(ctx, "/backend.Backend/Predict", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *backendClient) LoadModel(ctx context.Context, in *ModelOptions, opts ...grpc.CallOption) (*Result, error) {
+	out := new(Result)
+	err := c.cc.Invoke(ctx, "/backend.Backend/LoadModel", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *backendClient) PredictStream(ctx context.Context, in *PredictOptions, opts ...grpc.CallOption) (Backend_PredictStreamClient, error) {
+	stream, err := c.cc.NewStream(ctx, &Backend_ServiceDesc.Streams[0], "/backend.Backend/PredictStream", opts...)
+	if err != nil {
+		return nil, err
+	}
+	x := &backendPredictStreamClient{stream}
+	if err := x.ClientStream.SendMsg(in); err != nil {
+		return nil, err
+	}
+	if err := x.ClientStream.CloseSend(); err != nil {
+		return nil, err
+	}
+	return x, nil
+}
+
+type Backend_PredictStreamClient interface {
+	Recv() (*Reply, error)
+	grpc.ClientStream
+}
+
+type backendPredictStreamClient struct {
+	grpc.ClientStream
+}
+
+func (x *backendPredictStreamClient) Recv() (*Reply, error) {
+	m := new(Reply)
+	if err := x.ClientStream.RecvMsg(m); err != nil {
+		return nil, err
+	}
+	return m, nil
+}
+
+func (c *backendClient) Embedding(ctx context.Context, in *PredictOptions, opts ...grpc.CallOption) (*EmbeddingResult, error) {
+	out := new(EmbeddingResult)
+	err := c.cc.Invoke(ctx, "/backend.Backend/Embedding", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *backendClient) GenerateImage(ctx context.Context, in *GenerateImageRequest, opts ...grpc.CallOption) (*Result, error) {
+	out := new(Result)
+	err := c.cc.Invoke(ctx, "/backend.Backend/GenerateImage", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *backendClient) AudioTranscription(ctx context.Context, in *TranscriptRequest, opts ...grpc.CallOption) (*TranscriptResult, error) {
+	out := new(TranscriptResult)
+	err := c.cc.Invoke(ctx, "/backend.Backend/AudioTranscription", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *backendClient) TTS(ctx context.Context, in *TTSRequest, opts ...grpc.CallOption) (*Result, error) {
+	out := new(Result)
+	err := c.cc.Invoke(ctx, "/backend.Backend/TTS", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *backendClient) TokenizeString(ctx context.Context, in *PredictOptions, opts ...grpc.CallOption) (*TokenizationResponse, error) {
+	out := new(TokenizationResponse)
+	err := c.cc.Invoke(ctx, "/backend.Backend/TokenizeString", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *backendClient) Status(ctx context.Context, in *HealthMessage, opts ...grpc.CallOption) (*StatusResponse, error) {
+	out := new(StatusResponse)
+	err := c.cc.Invoke(ctx, "/backend.Backend/Status", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+// BackendServer is the server API for Backend service.
+// All implementations must embed UnimplementedBackendServer
+// for forward compatibility
+type BackendServer interface {
+	Health(context.Context, *HealthMessage) (*Reply, error)
+	Predict(context.Context, *PredictOptions) (*Reply, error)
+	LoadModel(context.Context, *ModelOptions) (*Result, error)
+	PredictStream(*PredictOptions, Backend_PredictStreamServer) error
+	Embedding(context.Context, *PredictOptions) (*EmbeddingResult, error)
+	GenerateImage(context.Context, *GenerateImageRequest) (*Result, error)
+	AudioTranscription(context.Context, *TranscriptRequest) (*TranscriptResult, error)
+	TTS(context.Context, *TTSRequest) (*Result, error)
+	TokenizeString(context.Context, *PredictOptions) (*TokenizationResponse, error)
+	Status(context.Context, *HealthMessage) (*StatusResponse, error)
+	mustEmbedUnimplementedBackendServer()
+}
+
+// UnimplementedBackendServer must be embedded to have forward compatible implementations.
+type UnimplementedBackendServer struct {
+}
+
+func (UnimplementedBackendServer) Health(context.Context, *HealthMessage) (*Reply, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method Health not implemented")
+}
+func (UnimplementedBackendServer) Predict(context.Context, *PredictOptions) (*Reply, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method Predict not implemented")
+}
+func (UnimplementedBackendServer) LoadModel(context.Context, *ModelOptions) (*Result, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method LoadModel not implemented")
+}
+func (UnimplementedBackendServer) PredictStream(*PredictOptions, Backend_PredictStreamServer) error {
+	return status.Errorf(codes.Unimplemented, "method PredictStream not implemented")
+}
+func (UnimplementedBackendServer) Embedding(context.Context, *PredictOptions) (*EmbeddingResult, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method Embedding not implemented")
+}
+func (UnimplementedBackendServer) GenerateImage(context.Context, *GenerateImageRequest) (*Result, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method GenerateImage not implemented")
+}
+func (UnimplementedBackendServer) AudioTranscription(context.Context, *TranscriptRequest) (*TranscriptResult, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method AudioTranscription not implemented")
+}
+func (UnimplementedBackendServer) TTS(context.Context, *TTSRequest) (*Result, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method TTS not implemented")
+}
+func (UnimplementedBackendServer) TokenizeString(context.Context, *PredictOptions) (*TokenizationResponse, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method TokenizeString not implemented")
+}
+func (UnimplementedBackendServer) Status(context.Context, *HealthMessage) (*StatusResponse, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method Status not implemented")
+}
+func (UnimplementedBackendServer) mustEmbedUnimplementedBackendServer() {}
+
+// UnsafeBackendServer may be embedded to opt out of forward compatibility for this service.
+// Use of this interface is not recommended, as added methods to BackendServer will
+// result in compilation errors.
+type UnsafeBackendServer interface {
+	mustEmbedUnimplementedBackendServer()
+}
+
+func RegisterBackendServer(s grpc.ServiceRegistrar, srv BackendServer) {
+	s.RegisterService(&Backend_ServiceDesc, srv)
+}
+
+func _Backend_Health_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(HealthMessage)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).Health(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/Health",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).Health(ctx, req.(*HealthMessage))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Backend_Predict_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(PredictOptions)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).Predict(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/Predict",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).Predict(ctx, req.(*PredictOptions))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Backend_LoadModel_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(ModelOptions)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).LoadModel(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/LoadModel",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).LoadModel(ctx, req.(*ModelOptions))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Backend_PredictStream_Handler(srv interface{}, stream grpc.ServerStream) error {
+	m := new(PredictOptions)
+	if err := stream.RecvMsg(m); err != nil {
+		return err
+	}
+	return srv.(BackendServer).PredictStream(m, &backendPredictStreamServer{stream})
+}
+
+type Backend_PredictStreamServer interface {
+	Send(*Reply) error
+	grpc.ServerStream
+}
+
+type backendPredictStreamServer struct {
+	grpc.ServerStream
+}
+
+func (x *backendPredictStreamServer) Send(m *Reply) error {
+	return x.ServerStream.SendMsg(m)
+}
+
+func _Backend_Embedding_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(PredictOptions)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).Embedding(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/Embedding",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).Embedding(ctx, req.(*PredictOptions))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Backend_GenerateImage_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(GenerateImageRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).GenerateImage(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/GenerateImage",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).GenerateImage(ctx, req.(*GenerateImageRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Backend_AudioTranscription_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(TranscriptRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).AudioTranscription(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/AudioTranscription",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).AudioTranscription(ctx, req.(*TranscriptRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Backend_TTS_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(TTSRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).TTS(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/TTS",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).TTS(ctx, req.(*TTSRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Backend_TokenizeString_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(PredictOptions)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).TokenizeString(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/TokenizeString",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).TokenizeString(ctx, req.(*PredictOptions))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Backend_Status_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(HealthMessage)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).Status(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/Status",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).Status(ctx, req.(*HealthMessage))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+// Backend_ServiceDesc is the grpc.ServiceDesc for Backend service.
+// It's only intended for direct use with grpc.RegisterService,
+// and not to be introspected or modified (even as a copy)
+var Backend_ServiceDesc = grpc.ServiceDesc{
+	ServiceName: "backend.Backend",
+	HandlerType: (*BackendServer)(nil),
+	Methods: []grpc.MethodDesc{
+		{
+			MethodName: "Health",
+			Handler:    _Backend_Health_Handler,
+		},
+		{
+			MethodName: "Predict",
+			Handler:    _Backend_Predict_Handler,
+		},
+		{
+			MethodName: "LoadModel",
+			Handler:    _Backend_LoadModel_Handler,
+		},
+		{
+			MethodName: "Embedding",
+			Handler:    _Backend_Embedding_Handler,
+		},
+		{
+			MethodName: "GenerateImage",
+			Handler:    _Backend_GenerateImage_Handler,
+		},
+		{
+			MethodName: "AudioTranscription",
+			Handler:    _Backend_AudioTranscription_Handler,
+		},
+		{
+			MethodName: "TTS",
+			Handler:    _Backend_TTS_Handler,
+		},
+		{
+			MethodName: "TokenizeString",
+			Handler:    _Backend_TokenizeString_Handler,
+		},
+		{
+			MethodName: "Status",
+			Handler:    _Backend_Status_Handler,
+		},
+	},
+	Streams: []grpc.StreamDesc{
+		{
+			StreamName:    "PredictStream",
+			Handler:       _Backend_PredictStream_Handler,
+			ServerStreams: true,
+		},
+	},
+	Metadata: "backend/backend.proto",
+}

backend/cpp/grpc/.gitignore

@@ -0,0 +1,3 @@
+installed_packages/
+grpc_build/
+grpc_repo/

backend/cpp/grpc/Makefile

@@ -0,0 +1,66 @@
+# Basic platform detection
+HOST_SYSTEM = $(shell uname | cut -f 1 -d_)
+SYSTEM ?= $(HOST_SYSTEM)
+
+TAG_LIB_GRPC?=v1.59.0
+GIT_REPO_LIB_GRPC?=https://github.com/grpc/grpc.git
+GIT_CLONE_DEPTH?=1
+NUM_BUILD_THREADS?=$(shell nproc --ignore=1)
+
+INSTALLED_PACKAGES=installed_packages
+GRPC_REPO=grpc_repo
+GRPC_BUILD=grpc_build
+
+export CMAKE_ARGS?=
+CMAKE_ARGS+=-DCMAKE_BUILD_TYPE=Release
+CMAKE_ARGS+=-DgRPC_INSTALL=ON
+CMAKE_ARGS+=-DEXECUTABLE_OUTPUT_PATH=../$(INSTALLED_PACKAGES)/grpc/bin
+CMAKE_ARGS+=-DLIBRARY_OUTPUT_PATH=../$(INSTALLED_PACKAGES)/grpc/lib
+CMAKE_ARGS+=-DgRPC_BUILD_TESTS=OFF
+CMAKE_ARGS+=-DgRPC_BUILD_CSHARP_EXT=OFF
+CMAKE_ARGS+=-DgRPC_BUILD_GRPC_CPP_PLUGIN=ON
+CMAKE_ARGS+=-DgRPC_BUILD_GRPC_CSHARP_PLUGIN=OFF
+CMAKE_ARGS+=-DgRPC_BUILD_GRPC_NODE_PLUGIN=OFF
+CMAKE_ARGS+=-DgRPC_BUILD_GRPC_OBJECTIVE_C_PLUGIN=OFF
+CMAKE_ARGS+=-DgRPC_BUILD_GRPC_PHP_PLUGIN=OFF
+CMAKE_ARGS+=-DgRPC_BUILD_GRPC_PYTHON_PLUGIN=ON
+CMAKE_ARGS+=-DgRPC_BUILD_GRPC_RUBY_PLUGIN=OFF
+CMAKE_ARGS+=-Dprotobuf_WITH_ZLIB=ON
+CMAKE_ARGS+=-DRE2_BUILD_TESTING=OFF
+CMAKE_ARGS+=-DCMAKE_INSTALL_PREFIX=../$(INSTALLED_PACKAGES)
+
+# windows need to set OPENSSL_NO_ASM. Results in slower crypto performance but doesn't build otherwise.
+# May be resolvable, but for now its set. More info: https://stackoverflow.com/a/75240504/480673
+ifeq ($(SYSTEM),MSYS)
+CMAKE_ARGS+=-DOPENSSL_NO_ASM=ON
+endif
+ifeq ($(SYSTEM),MINGW64)
+CMAKE_ARGS+=-DOPENSSL_NO_ASM=ON
+endif
+ifeq ($(SYSTEM),MINGW32)
+CMAKE_ARGS+=-DOPENSSL_NO_ASM=ON
+endif
+ifeq ($(SYSTEM),CYGWIN)
+CMAKE_ARGS+=-DOPENSSL_NO_ASM=ON
+endif
+
+$(INSTALLED_PACKAGES): grpc_build
+
+$(GRPC_REPO):
+	git clone --depth $(GIT_CLONE_DEPTH) -b $(TAG_LIB_GRPC) $(GIT_REPO_LIB_GRPC) $(GRPC_REPO)/grpc
+	cd $(GRPC_REPO)/grpc && git submodule update --init --recursive --depth $(GIT_CLONE_DEPTH)
+
+$(GRPC_BUILD): $(GRPC_REPO)
+	mkdir -p $(GRPC_BUILD)
+	cd $(GRPC_BUILD) && cmake $(CMAKE_ARGS) ../$(GRPC_REPO)/grpc && cmake --build . -- -j ${NUM_BUILD_THREADS} && cmake --build . --target install -- -j ${NUM_BUILD_THREADS}
+
+build: $(INSTALLED_PACKAGES)
+
+rebuild:
+	rm -rf grpc_build
+	$(MAKE) grpc_build
+
+clean:
+	rm -rf grpc_build
+	rm -rf grpc_repo
+	rm -rf installed_packages

backend/cpp/llama/CMakeLists.txt

@@ -0,0 +1,82 @@
+
+## XXX: In some versions of CMake clip wasn't being built before llama.
+## This is an hack for now, but it should be fixed in the future.
+set(TARGET myclip)
+add_library(${TARGET} clip.cpp clip.h)
+install(TARGETS ${TARGET} LIBRARY)
+target_link_libraries(${TARGET} PRIVATE common ggml ${CMAKE_THREAD_LIBS_INIT})
+target_compile_features(${TARGET} PRIVATE cxx_std_11)
+if (NOT MSVC)
+    target_compile_options(${TARGET} PRIVATE -Wno-cast-qual) # stb_image.h
+endif()
+
+set(TARGET grpc-server)
+# END CLIP hack
+set(CMAKE_CXX_STANDARD 17)
+cmake_minimum_required(VERSION 3.15)
+set(TARGET grpc-server)
+set(_PROTOBUF_LIBPROTOBUF libprotobuf)
+set(_REFLECTION grpc++_reflection)
+
+if (${CMAKE_SYSTEM_NAME} MATCHES "Darwin")
+    # Set correct Homebrew install folder for Apple Silicon and Intel Macs
+    if (CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "arm64")
+        set(HOMEBREW_DEFAULT_PREFIX "/opt/homebrew")
+    else()
+        set(HOMEBREW_DEFAULT_PREFIX "/usr/local")
+    endif()
+
+    link_directories("${HOMEBREW_DEFAULT_PREFIX}/lib")
+    include_directories("${HOMEBREW_DEFAULT_PREFIX}/include")
+endif()
+
+find_package(absl CONFIG REQUIRED)
+find_package(Protobuf CONFIG REQUIRED)
+find_package(gRPC CONFIG REQUIRED)
+
+find_program(_PROTOBUF_PROTOC protoc)
+set(_GRPC_GRPCPP grpc++)
+find_program(_GRPC_CPP_PLUGIN_EXECUTABLE grpc_cpp_plugin)
+
+include_directories(${CMAKE_CURRENT_BINARY_DIR})
+include_directories(${Protobuf_INCLUDE_DIRS})
+
+message(STATUS "Using protobuf version ${Protobuf_VERSION} | Protobuf_INCLUDE_DIRS: ${Protobuf_INCLUDE_DIRS} | CMAKE_CURRENT_BINARY_DIR: ${CMAKE_CURRENT_BINARY_DIR}")
+
+# Proto file
+get_filename_component(hw_proto "../../../../../../backend/backend.proto" ABSOLUTE)
+get_filename_component(hw_proto_path "${hw_proto}" PATH)
+
+# Generated sources
+set(hw_proto_srcs "${CMAKE_CURRENT_BINARY_DIR}/backend.pb.cc")
+set(hw_proto_hdrs "${CMAKE_CURRENT_BINARY_DIR}/backend.pb.h")
+set(hw_grpc_srcs "${CMAKE_CURRENT_BINARY_DIR}/backend.grpc.pb.cc")
+set(hw_grpc_hdrs "${CMAKE_CURRENT_BINARY_DIR}/backend.grpc.pb.h")
+
+add_custom_command(
+      OUTPUT "${hw_proto_srcs}" "${hw_proto_hdrs}" "${hw_grpc_srcs}" "${hw_grpc_hdrs}"
+      COMMAND ${_PROTOBUF_PROTOC}
+      ARGS --grpc_out "${CMAKE_CURRENT_BINARY_DIR}"
+        --cpp_out "${CMAKE_CURRENT_BINARY_DIR}"
+        -I "${hw_proto_path}"
+        --plugin=protoc-gen-grpc="${_GRPC_CPP_PLUGIN_EXECUTABLE}"
+        "${hw_proto}"
+      DEPENDS "${hw_proto}")
+
+# hw_grpc_proto
+add_library(hw_grpc_proto
+  ${hw_grpc_srcs}
+  ${hw_grpc_hdrs}
+  ${hw_proto_srcs}
+  ${hw_proto_hdrs} )
+
+add_executable(${TARGET} grpc-server.cpp json.hpp )
+target_link_libraries(${TARGET} PRIVATE common llama myclip ${CMAKE_THREAD_LIBS_INIT} absl::flags hw_grpc_proto
+  absl::flags_parse
+  gRPC::${_REFLECTION}
+  gRPC::${_GRPC_GRPCPP}
+  protobuf::${_PROTOBUF_LIBPROTOBUF})
+target_compile_features(${TARGET} PRIVATE cxx_std_11)
+if(TARGET BUILD_INFO)
+  add_dependencies(${TARGET} BUILD_INFO)
+endif()

backend/cpp/llama/Makefile

@@ -0,0 +1,53 @@
+
+LLAMA_VERSION?=
+
+CMAKE_ARGS?=
+BUILD_TYPE?=
+
+# If build type is cublas, then we set -DLLAMA_CUBLAS=ON to CMAKE_ARGS automatically
+ifeq ($(BUILD_TYPE),cublas)
+	CMAKE_ARGS+=-DLLAMA_CUBLAS=ON
+# If build type is openblas then we set -DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS
+# to CMAKE_ARGS automatically
+else ifeq ($(BUILD_TYPE),openblas)
+	CMAKE_ARGS+=-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS
+# If build type is clblast (openCL) we set -DLLAMA_CLBLAST=ON -DCLBlast_DIR=/some/path
+else ifeq ($(BUILD_TYPE),clblast)
+	CMAKE_ARGS+=-DLLAMA_CLBLAST=ON -DCLBlast_DIR=/some/path
+# If it's hipblas we do have also to set CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++ 
+else ifeq ($(BUILD_TYPE),hipblas)
+	CMAKE_ARGS+=-DLLAMA_HIPBLAS=ON
+endif
+
+llama.cpp:
+	git clone --recurse-submodules https://github.com/ggerganov/llama.cpp llama.cpp
+	if [ -z "$(LLAMA_VERSION)" ]; then \
+		exit 1; \
+	fi
+	cd llama.cpp && git checkout -b build $(LLAMA_VERSION) && git submodule update --init --recursive --depth 1
+
+llama.cpp/examples/grpc-server:
+	mkdir -p llama.cpp/examples/grpc-server
+	cp -r $(abspath ./)/CMakeLists.txt llama.cpp/examples/grpc-server/
+	cp -r $(abspath ./)/grpc-server.cpp llama.cpp/examples/grpc-server/
+	cp -rfv $(abspath ./)/json.hpp llama.cpp/examples/grpc-server/
+	echo "add_subdirectory(grpc-server)" >> llama.cpp/examples/CMakeLists.txt
+## XXX: In some versions of CMake clip wasn't being built before llama.
+## This is an hack for now, but it should be fixed in the future.
+	cp -rfv llama.cpp/examples/llava/clip.h llama.cpp/examples/grpc-server/clip.h
+	cp -rfv llama.cpp/examples/llava/clip.cpp llama.cpp/examples/grpc-server/clip.cpp
+
+rebuild:
+	cp -rfv $(abspath ./)/CMakeLists.txt llama.cpp/examples/grpc-server/
+	cp -rfv $(abspath ./)/grpc-server.cpp llama.cpp/examples/grpc-server/
+	cp -rfv $(abspath ./)/json.hpp llama.cpp/examples/grpc-server/
+	rm -rf grpc-server
+	$(MAKE) grpc-server
+
+clean:
+	rm -rf llama.cpp
+	rm -rf grpc-server
+
+grpc-server: llama.cpp llama.cpp/examples/grpc-server
+	cd llama.cpp && mkdir -p build && cd build && cmake .. $(CMAKE_ARGS) && cmake --build . --config Release
+	cp llama.cpp/build/bin/grpc-server .

backend/go/image/stablediffusion/main.go

@@ -0,0 +1,21 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &Image{}); err != nil {
+		panic(err)
+	}
+}

backend/go/image/stablediffusion/stablediffusion.go

@@ -0,0 +1,33 @@
+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/go-skynet/LocalAI/pkg/grpc/base"
+	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
+	"github.com/go-skynet/LocalAI/pkg/stablediffusion"
+)
+
+type Image struct {
+	base.SingleThread
+	stablediffusion *stablediffusion.StableDiffusion
+}
+
+func (image *Image) Load(opts *pb.ModelOptions) error {
+	var err error
+	// Note: the Model here is a path to a directory containing the model files
+	image.stablediffusion, err = stablediffusion.New(opts.ModelFile)
+	return err
+}
+
+func (image *Image) GenerateImage(opts *pb.GenerateImageRequest) error {
+	return image.stablediffusion.GenerateImage(
+		int(opts.Height),
+		int(opts.Width),
+		int(opts.Mode),
+		int(opts.Step),
+		int(opts.Seed),
+		opts.PositivePrompt,
+		opts.NegativePrompt,
+		opts.Dst)
+}

backend/go/image/tinydream/main.go

@@ -0,0 +1,21 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &Image{}); err != nil {
+		panic(err)
+	}
+}

backend/go/image/tinydream/tinydream.go

@@ -0,0 +1,32 @@
+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/go-skynet/LocalAI/pkg/grpc/base"
+	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
+	"github.com/go-skynet/LocalAI/pkg/tinydream"
+)
+
+type Image struct {
+	base.SingleThread
+	tinydream *tinydream.TinyDream
+}
+
+func (image *Image) Load(opts *pb.ModelOptions) error {
+	var err error
+	// Note: the Model here is a path to a directory containing the model files
+	image.tinydream, err = tinydream.New(opts.ModelFile)
+	return err
+}
+
+func (image *Image) GenerateImage(opts *pb.GenerateImageRequest) error {
+	return image.tinydream.GenerateImage(
+		int(opts.Height),
+		int(opts.Width),
+		int(opts.Step),
+		int(opts.Seed),
+		opts.PositivePrompt,
+		opts.NegativePrompt,
+		opts.Dst)
+}

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/go-skynet/LocalAI/pkg/grpc/base"
+	pb "github.com/go-skynet/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

@@ -0,0 +1,21 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &Embeddings{}); err != nil {
+		panic(err)
+	}
+}

backend/go/llm/dolly/main.go

@@ -0,0 +1,23 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	transformers "github.com/go-skynet/LocalAI/backend/go/llm/transformers"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &transformers.Dolly{}); err != nil {
+		panic(err)
+	}
+}

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

@@ -0,0 +1,23 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	transformers "github.com/go-skynet/LocalAI/backend/go/llm/transformers"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &transformers.Falcon{}); err != nil {
+		panic(err)
+	}
+}

backend/go/llm/gpt2/main.go

@@ -0,0 +1,23 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	transformers "github.com/go-skynet/LocalAI/backend/go/llm/transformers"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &transformers.GPT2{}); 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/go-skynet/LocalAI/pkg/grpc/base"
+	pb "github.com/go-skynet/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

@@ -0,0 +1,21 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &LLM{}); err != nil {
+		panic(err)
+	}
+}

backend/go/llm/gptj/main.go

@@ -0,0 +1,23 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	transformers "github.com/go-skynet/LocalAI/backend/go/llm/transformers"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &transformers.GPTJ{}); err != nil {
+		panic(err)
+	}
+}

backend/go/llm/gptneox/main.go

@@ -0,0 +1,23 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	transformers "github.com/go-skynet/LocalAI/backend/go/llm/transformers"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &transformers.GPTNeoX{}); err != nil {
+		panic(err)
+	}
+}

backend/go/llm/langchain/langchain.go

@@ -0,0 +1,58 @@
+package main
+
+// This is a wrapper to statisfy the GRPC service interface
+// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
+import (
+	"fmt"
+
+	"github.com/go-skynet/LocalAI/pkg/grpc/base"
+	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
+	"github.com/go-skynet/LocalAI/pkg/langchain"
+)
+
+type LLM struct {
+	base.Base
+
+	langchain *langchain.HuggingFace
+	model     string
+}
+
+func (llm *LLM) Load(opts *pb.ModelOptions) error {
+	llm.langchain, _ = langchain.NewHuggingFace(opts.Model)
+	llm.model = opts.Model
+	return nil
+}
+
+func (llm *LLM) Predict(opts *pb.PredictOptions) (string, error) {
+	o := []langchain.PredictOption{
+		langchain.SetModel(llm.model),
+		langchain.SetMaxTokens(int(opts.Tokens)),
+		langchain.SetTemperature(float64(opts.Temperature)),
+		langchain.SetStopWords(opts.StopPrompts),
+	}
+	pred, err := llm.langchain.PredictHuggingFace(opts.Prompt, o...)
+	if err != nil {
+		return "", err
+	}
+	return pred.Completion, nil
+}
+
+func (llm *LLM) PredictStream(opts *pb.PredictOptions, results chan string) error {
+	o := []langchain.PredictOption{
+		langchain.SetModel(llm.model),
+		langchain.SetMaxTokens(int(opts.Tokens)),
+		langchain.SetTemperature(float64(opts.Temperature)),
+		langchain.SetStopWords(opts.StopPrompts),
+	}
+	go func() {
+		res, err := llm.langchain.PredictHuggingFace(opts.Prompt, o...)
+
+		if err != nil {
+			fmt.Println("err: ", err)
+		}
+		results <- res.Completion
+		close(results)
+	}()
+
+	return nil
+}

backend/go/llm/langchain/main.go

@@ -0,0 +1,21 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &LLM{}); err != nil {
+		panic(err)
+	}
+}

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

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

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

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

backend/go/llm/llama/llama.go

@@ -0,0 +1,257 @@
+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/go-skynet/LocalAI/pkg/grpc/base"
+	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
+	"github.com/go-skynet/go-llama.cpp"
+)
+
+type LLM struct {
+	base.SingleThread
+
+	llama      *llama.LLama
+	draftModel *llama.LLama
+}
+
+func (llm *LLM) Load(opts *pb.ModelOptions) error {
+	ropeFreqBase := float32(10000)
+	ropeFreqScale := float32(1)
+
+	if opts.RopeFreqBase != 0 {
+		ropeFreqBase = opts.RopeFreqBase
+	}
+	if opts.RopeFreqScale != 0 {
+		ropeFreqScale = opts.RopeFreqScale
+	}
+
+	llamaOpts := []llama.ModelOption{
+		llama.WithRopeFreqBase(ropeFreqBase),
+		llama.WithRopeFreqScale(ropeFreqScale),
+	}
+
+	if opts.NoMulMatQ {
+		llamaOpts = append(llamaOpts, llama.SetMulMatQ(false))
+	}
+
+	// Get base path of opts.ModelFile and use the same for lora (assume the same path)
+	basePath := filepath.Dir(opts.ModelFile)
+
+	if opts.LoraAdapter != "" {
+		llamaOpts = append(llamaOpts, llama.SetLoraAdapter(filepath.Join(basePath, opts.LoraAdapter)))
+	}
+
+	if opts.LoraBase != "" {
+		llamaOpts = append(llamaOpts, llama.SetLoraBase(filepath.Join(basePath, opts.LoraBase)))
+	}
+
+	if opts.ContextSize != 0 {
+		llamaOpts = append(llamaOpts, llama.SetContext(int(opts.ContextSize)))
+	}
+	if opts.F16Memory {
+		llamaOpts = append(llamaOpts, llama.EnableF16Memory)
+	}
+	if opts.Embeddings {
+		llamaOpts = append(llamaOpts, llama.EnableEmbeddings)
+	}
+	if opts.NGPULayers != 0 {
+		llamaOpts = append(llamaOpts, llama.SetGPULayers(int(opts.NGPULayers)))
+	}
+
+	llamaOpts = append(llamaOpts, llama.SetMMap(opts.MMap))
+	llamaOpts = append(llamaOpts, llama.SetMainGPU(opts.MainGPU))
+	llamaOpts = append(llamaOpts, llama.SetTensorSplit(opts.TensorSplit))
+	if opts.NBatch != 0 {
+		llamaOpts = append(llamaOpts, llama.SetNBatch(int(opts.NBatch)))
+	} else {
+		llamaOpts = append(llamaOpts, llama.SetNBatch(512))
+	}
+
+	if opts.NUMA {
+		llamaOpts = append(llamaOpts, llama.EnableNUMA)
+	}
+
+	if opts.LowVRAM {
+		llamaOpts = append(llamaOpts, llama.EnabelLowVRAM)
+	}
+
+	if opts.DraftModel != "" {
+		// https://github.com/ggerganov/llama.cpp/blob/71ca2fad7d6c0ef95ef9944fb3a1a843e481f314/examples/speculative/speculative.cpp#L40
+		llamaOpts = append(llamaOpts, llama.SetPerplexity(true))
+	}
+
+	model, err := llama.New(opts.ModelFile, llamaOpts...)
+
+	if opts.DraftModel != "" {
+		// opts.DraftModel is relative to opts.ModelFile, so we need to get the basepath of opts.ModelFile
+		if !filepath.IsAbs(opts.DraftModel) {
+			dir := filepath.Dir(opts.ModelFile)
+			opts.DraftModel = filepath.Join(dir, opts.DraftModel)
+		}
+
+		draftModel, err := llama.New(opts.DraftModel, llamaOpts...)
+		if err != nil {
+			return err
+		}
+		llm.draftModel = draftModel
+	}
+
+	llm.llama = model
+
+	return err
+}
+
+func buildPredictOptions(opts *pb.PredictOptions) []llama.PredictOption {
+	ropeFreqBase := float32(10000)
+	ropeFreqScale := float32(1)
+
+	if opts.RopeFreqBase != 0 {
+		ropeFreqBase = opts.RopeFreqBase
+	}
+	if opts.RopeFreqScale != 0 {
+		ropeFreqScale = opts.RopeFreqScale
+	}
+	predictOptions := []llama.PredictOption{
+		llama.SetTemperature(opts.Temperature),
+		llama.SetTopP(opts.TopP),
+		llama.SetTopK(int(opts.TopK)),
+		llama.SetTokens(int(opts.Tokens)),
+		llama.SetThreads(int(opts.Threads)),
+		llama.WithGrammar(opts.Grammar),
+		llama.SetRopeFreqBase(ropeFreqBase),
+		llama.SetRopeFreqScale(ropeFreqScale),
+		llama.SetNegativePromptScale(opts.NegativePromptScale),
+		llama.SetNegativePrompt(opts.NegativePrompt),
+	}
+
+	if opts.PromptCacheAll {
+		predictOptions = append(predictOptions, llama.EnablePromptCacheAll)
+	}
+
+	if opts.PromptCacheRO {
+		predictOptions = append(predictOptions, llama.EnablePromptCacheRO)
+	}
+
+	// Expected absolute path
+	if opts.PromptCachePath != "" {
+		predictOptions = append(predictOptions, llama.SetPathPromptCache(opts.PromptCachePath))
+	}
+
+	if opts.Mirostat != 0 {
+		predictOptions = append(predictOptions, llama.SetMirostat(int(opts.Mirostat)))
+	}
+
+	if opts.MirostatETA != 0 {
+		predictOptions = append(predictOptions, llama.SetMirostatETA(opts.MirostatETA))
+	}
+
+	if opts.MirostatTAU != 0 {
+		predictOptions = append(predictOptions, llama.SetMirostatTAU(opts.MirostatTAU))
+	}
+
+	if opts.Debug {
+		predictOptions = append(predictOptions, llama.Debug)
+	}
+
+	predictOptions = append(predictOptions, llama.SetStopWords(opts.StopPrompts...))
+
+	if opts.PresencePenalty != 0 {
+		predictOptions = append(predictOptions, llama.SetPenalty(opts.PresencePenalty))
+	}
+
+	if opts.NKeep != 0 {
+		predictOptions = append(predictOptions, llama.SetNKeep(int(opts.NKeep)))
+	}
+
+	if opts.Batch != 0 {
+		predictOptions = append(predictOptions, llama.SetBatch(int(opts.Batch)))
+	}
+
+	if opts.F16KV {
+		predictOptions = append(predictOptions, llama.EnableF16KV)
+	}
+
+	if opts.IgnoreEOS {
+		predictOptions = append(predictOptions, llama.IgnoreEOS)
+	}
+
+	if opts.Seed != 0 {
+		predictOptions = append(predictOptions, llama.SetSeed(int(opts.Seed)))
+	}
+
+	if opts.NDraft != 0 {
+		predictOptions = append(predictOptions, llama.SetNDraft(int(opts.NDraft)))
+	}
+	//predictOptions = append(predictOptions, llama.SetLogitBias(c.Seed))
+
+	predictOptions = append(predictOptions, llama.SetFrequencyPenalty(opts.FrequencyPenalty))
+	predictOptions = append(predictOptions, llama.SetMlock(opts.MLock))
+	predictOptions = append(predictOptions, llama.SetMemoryMap(opts.MMap))
+	predictOptions = append(predictOptions, llama.SetPredictionMainGPU(opts.MainGPU))
+	predictOptions = append(predictOptions, llama.SetPredictionTensorSplit(opts.TensorSplit))
+	predictOptions = append(predictOptions, llama.SetTailFreeSamplingZ(opts.TailFreeSamplingZ))
+	predictOptions = append(predictOptions, llama.SetTypicalP(opts.TypicalP))
+	return predictOptions
+}
+
+func (llm *LLM) Predict(opts *pb.PredictOptions) (string, error) {
+	if llm.draftModel != nil {
+		return llm.llama.SpeculativeSampling(llm.draftModel, opts.Prompt, buildPredictOptions(opts)...)
+	}
+	return llm.llama.Predict(opts.Prompt, buildPredictOptions(opts)...)
+}
+
+func (llm *LLM) PredictStream(opts *pb.PredictOptions, results chan string) error {
+	predictOptions := buildPredictOptions(opts)
+
+	predictOptions = append(predictOptions, llama.SetTokenCallback(func(token string) bool {
+		results <- token
+		return true
+	}))
+
+	go func() {
+		var err error
+		if llm.draftModel != nil {
+			_, err = llm.llama.SpeculativeSampling(llm.draftModel, opts.Prompt, buildPredictOptions(opts)...)
+		} else {
+			_, err = llm.llama.Predict(opts.Prompt, predictOptions...)
+		}
+
+		if err != nil {
+			fmt.Println("err: ", err)
+		}
+		close(results)
+	}()
+
+	return nil
+}
+
+func (llm *LLM) Embeddings(opts *pb.PredictOptions) ([]float32, error) {
+	predictOptions := buildPredictOptions(opts)
+
+	if len(opts.EmbeddingTokens) > 0 {
+		tokens := []int{}
+		for _, t := range opts.EmbeddingTokens {
+			tokens = append(tokens, int(t))
+		}
+		return llm.llama.TokenEmbeddings(tokens, predictOptions...)
+	}
+
+	return llm.llama.Embeddings(opts.Embeddings, predictOptions...)
+}
+
+func (llm *LLM) TokenizeString(opts *pb.PredictOptions) (pb.TokenizationResponse, error) {
+	predictOptions := buildPredictOptions(opts)
+	l, tokens, err := llm.llama.TokenizeString(opts.Prompt, predictOptions...)
+	if err != nil {
+		return pb.TokenizationResponse{}, err
+	}
+	return pb.TokenizationResponse{
+		Length: l,
+		Tokens: tokens,
+	}, nil
+}

backend/go/llm/llama/main.go

@@ -0,0 +1,23 @@
+package main
+
+// GRPC Falcon server
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &LLM{}); err != nil {
+		panic(err)
+	}
+}

backend/go/llm/mpt/main.go

@@ -0,0 +1,23 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	transformers "github.com/go-skynet/LocalAI/backend/go/llm/transformers"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &transformers.MPT{}); err != nil {
+		panic(err)
+	}
+}

backend/go/llm/replit/main.go

@@ -0,0 +1,23 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	transformers "github.com/go-skynet/LocalAI/backend/go/llm/transformers"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &transformers.Replit{}); err != nil {
+		panic(err)
+	}
+}

backend/go/llm/rwkv/main.go

@@ -0,0 +1,21 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &LLM{}); err != nil {
+		panic(err)
+	}
+}

backend/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/go-skynet/LocalAI/pkg/grpc/base"
+	pb "github.com/go-skynet/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("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/llm/starcoder/main.go

@@ -0,0 +1,23 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	transformers "github.com/go-skynet/LocalAI/backend/go/llm/transformers"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &transformers.Starcoder{}); err != nil {
+		panic(err)
+	}
+}

backend/go/llm/transformers/dolly.go

@@ -0,0 +1,44 @@
+package transformers
+
+// This is a wrapper to statisfy the GRPC service interface
+// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
+import (
+	"fmt"
+
+	"github.com/go-skynet/LocalAI/pkg/grpc/base"
+	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
+
+	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
+)
+
+type Dolly struct {
+	base.SingleThread
+
+	dolly *transformers.Dolly
+}
+
+func (llm *Dolly) Load(opts *pb.ModelOptions) error {
+	model, err := transformers.NewDolly(opts.ModelFile)
+	llm.dolly = model
+	return err
+}
+
+func (llm *Dolly) Predict(opts *pb.PredictOptions) (string, error) {
+	return llm.dolly.Predict(opts.Prompt, buildPredictOptions(opts)...)
+}
+
+// fallback to Predict
+func (llm *Dolly) PredictStream(opts *pb.PredictOptions, results chan string) error {
+
+	go func() {
+		res, err := llm.dolly.Predict(opts.Prompt, buildPredictOptions(opts)...)
+
+		if err != nil {
+			fmt.Println("err: ", err)
+		}
+		results <- res
+		close(results)
+	}()
+
+	return nil
+}

backend/go/llm/transformers/falcon.go

@@ -0,0 +1,43 @@
+package transformers
+
+// This is a wrapper to statisfy the GRPC service interface
+// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
+import (
+	"fmt"
+
+	"github.com/go-skynet/LocalAI/pkg/grpc/base"
+	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
+
+	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
+)
+
+type Falcon struct {
+	base.SingleThread
+
+	falcon *transformers.Falcon
+}
+
+func (llm *Falcon) Load(opts *pb.ModelOptions) error {
+	model, err := transformers.NewFalcon(opts.ModelFile)
+	llm.falcon = model
+	return err
+}
+
+func (llm *Falcon) Predict(opts *pb.PredictOptions) (string, error) {
+	return llm.falcon.Predict(opts.Prompt, buildPredictOptions(opts)...)
+}
+
+// fallback to Predict
+func (llm *Falcon) PredictStream(opts *pb.PredictOptions, results chan string) error {
+	go func() {
+		res, err := llm.falcon.Predict(opts.Prompt, buildPredictOptions(opts)...)
+
+		if err != nil {
+			fmt.Println("err: ", err)
+		}
+		results <- res
+		close(results)
+	}()
+
+	return nil
+}

backend/go/llm/transformers/gpt2.go

@@ -0,0 +1,42 @@
+package transformers
+
+// This is a wrapper to statisfy the GRPC service interface
+// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
+import (
+	"fmt"
+
+	"github.com/go-skynet/LocalAI/pkg/grpc/base"
+	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
+
+	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
+)
+
+type GPT2 struct {
+	base.SingleThread
+
+	gpt2 *transformers.GPT2
+}
+
+func (llm *GPT2) Load(opts *pb.ModelOptions) error {
+	model, err := transformers.New(opts.ModelFile)
+	llm.gpt2 = model
+	return err
+}
+
+func (llm *GPT2) Predict(opts *pb.PredictOptions) (string, error) {
+	return llm.gpt2.Predict(opts.Prompt, buildPredictOptions(opts)...)
+}
+
+// fallback to Predict
+func (llm *GPT2) PredictStream(opts *pb.PredictOptions, results chan string) error {
+	go func() {
+		res, err := llm.gpt2.Predict(opts.Prompt, buildPredictOptions(opts)...)
+
+		if err != nil {
+			fmt.Println("err: ", err)
+		}
+		results <- res
+		close(results)
+	}()
+	return nil
+}

backend/go/llm/transformers/gptj.go

@@ -0,0 +1,42 @@
+package transformers
+
+// This is a wrapper to statisfy the GRPC service interface
+// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
+import (
+	"fmt"
+
+	"github.com/go-skynet/LocalAI/pkg/grpc/base"
+	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
+
+	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
+)
+
+type GPTJ struct {
+	base.SingleThread
+
+	gptj *transformers.GPTJ
+}
+
+func (llm *GPTJ) Load(opts *pb.ModelOptions) error {
+	model, err := transformers.NewGPTJ(opts.ModelFile)
+	llm.gptj = model
+	return err
+}
+
+func (llm *GPTJ) Predict(opts *pb.PredictOptions) (string, error) {
+	return llm.gptj.Predict(opts.Prompt, buildPredictOptions(opts)...)
+}
+
+// fallback to Predict
+func (llm *GPTJ) PredictStream(opts *pb.PredictOptions, results chan string) error {
+	go func() {
+		res, err := llm.gptj.Predict(opts.Prompt, buildPredictOptions(opts)...)
+
+		if err != nil {
+			fmt.Println("err: ", err)
+		}
+		results <- res
+		close(results)
+	}()
+	return nil
+}

backend/go/llm/transformers/gptneox.go

@@ -0,0 +1,42 @@
+package transformers
+
+// This is a wrapper to statisfy the GRPC service interface
+// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
+import (
+	"fmt"
+
+	"github.com/go-skynet/LocalAI/pkg/grpc/base"
+	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
+
+	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
+)
+
+type GPTNeoX struct {
+	base.SingleThread
+
+	gptneox *transformers.GPTNeoX
+}
+
+func (llm *GPTNeoX) Load(opts *pb.ModelOptions) error {
+	model, err := transformers.NewGPTNeoX(opts.ModelFile)
+	llm.gptneox = model
+	return err
+}
+
+func (llm *GPTNeoX) Predict(opts *pb.PredictOptions) (string, error) {
+	return llm.gptneox.Predict(opts.Prompt, buildPredictOptions(opts)...)
+}
+
+// fallback to Predict
+func (llm *GPTNeoX) PredictStream(opts *pb.PredictOptions, results chan string) error {
+	go func() {
+		res, err := llm.gptneox.Predict(opts.Prompt, buildPredictOptions(opts)...)
+
+		if err != nil {
+			fmt.Println("err: ", err)
+		}
+		results <- res
+		close(results)
+	}()
+	return nil
+}

backend/go/llm/transformers/mpt.go

@@ -0,0 +1,42 @@
+package transformers
+
+// This is a wrapper to statisfy the GRPC service interface
+// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
+import (
+	"fmt"
+
+	"github.com/go-skynet/LocalAI/pkg/grpc/base"
+	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
+
+	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
+)
+
+type MPT struct {
+	base.SingleThread
+
+	mpt *transformers.MPT
+}
+
+func (llm *MPT) Load(opts *pb.ModelOptions) error {
+	model, err := transformers.NewMPT(opts.ModelFile)
+	llm.mpt = model
+	return err
+}
+
+func (llm *MPT) Predict(opts *pb.PredictOptions) (string, error) {
+	return llm.mpt.Predict(opts.Prompt, buildPredictOptions(opts)...)
+}
+
+// fallback to Predict
+func (llm *MPT) PredictStream(opts *pb.PredictOptions, results chan string) error {
+	go func() {
+		res, err := llm.mpt.Predict(opts.Prompt, buildPredictOptions(opts)...)
+
+		if err != nil {
+			fmt.Println("err: ", err)
+		}
+		results <- res
+		close(results)
+	}()
+	return nil
+}

backend/go/llm/transformers/predict.go

@@ -0,0 +1,26 @@
+package transformers
+
+import (
+	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
+	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
+)
+
+func buildPredictOptions(opts *pb.PredictOptions) []transformers.PredictOption {
+	predictOptions := []transformers.PredictOption{
+		transformers.SetTemperature(float64(opts.Temperature)),
+		transformers.SetTopP(float64(opts.TopP)),
+		transformers.SetTopK(int(opts.TopK)),
+		transformers.SetTokens(int(opts.Tokens)),
+		transformers.SetThreads(int(opts.Threads)),
+	}
+
+	if opts.Batch != 0 {
+		predictOptions = append(predictOptions, transformers.SetBatch(int(opts.Batch)))
+	}
+
+	if opts.Seed != 0 {
+		predictOptions = append(predictOptions, transformers.SetSeed(int(opts.Seed)))
+	}
+
+	return predictOptions
+}

backend/go/llm/transformers/replit.go

@@ -0,0 +1,42 @@
+package transformers
+
+// This is a wrapper to statisfy the GRPC service interface
+// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
+import (
+	"fmt"
+
+	"github.com/go-skynet/LocalAI/pkg/grpc/base"
+	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
+
+	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
+)
+
+type Replit struct {
+	base.SingleThread
+
+	replit *transformers.Replit
+}
+
+func (llm *Replit) Load(opts *pb.ModelOptions) error {
+	model, err := transformers.NewReplit(opts.ModelFile)
+	llm.replit = model
+	return err
+}
+
+func (llm *Replit) Predict(opts *pb.PredictOptions) (string, error) {
+	return llm.replit.Predict(opts.Prompt, buildPredictOptions(opts)...)
+}
+
+// fallback to Predict
+func (llm *Replit) PredictStream(opts *pb.PredictOptions, results chan string) error {
+	go func() {
+		res, err := llm.replit.Predict(opts.Prompt, buildPredictOptions(opts)...)
+
+		if err != nil {
+			fmt.Println("err: ", err)
+		}
+		results <- res
+		close(results)
+	}()
+	return nil
+}