Mise à jour de 'README.md'

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

.devcontainer/Dockerfile

@@ -1,3 +0,0 @@
-ARG GO_VERSION=1.20
-FROM mcr.microsoft.com/devcontainers/go:0-$GO_VERSION-bullseye
-RUN apt-get update && apt-get install -y cmake

.devcontainer/devcontainer.json

@@ -1,46 +0,0 @@
-// For format details, see https://aka.ms/devcontainer.json. For config options, see the
-// README at: https://github.com/devcontainers/templates/tree/main/src/docker-existing-docker-compose
-{
-	"name": "Existing Docker Compose (Extend)",
-
-	// Update the 'dockerComposeFile' list if you have more compose files or use different names.
-	// The .devcontainer/docker-compose.yml file contains any overrides you need/want to make.
-	"dockerComposeFile": [
-		"../docker-compose.yaml",
-		"docker-compose.yml"
-	],
-
-	// The 'service' property is the name of the service for the container that VS Code should
-	// use. Update this value and .devcontainer/docker-compose.yml to the real service name.
-	"service": "api",
-
-	// The optional 'workspaceFolder' property is the path VS Code should open by default when
-	// connected. This is typically a file mount in .devcontainer/docker-compose.yml
-	"workspaceFolder": "/workspace",
-
-	"features": {
-		"ghcr.io/devcontainers/features/go:1": {},
-		"ghcr.io/azutake/devcontainer-features/go-packages-install:0": {}
-	},
-
-	// Features to add to the dev container. More info: https://containers.dev/features.
-	// "features": {},
-
-	// Use 'forwardPorts' to make a list of ports inside the container available locally.
-	// "forwardPorts": [],
-
-	// Uncomment the next line if you want start specific services in your Docker Compose config.
-	// "runServices": [],
-
-	// Uncomment the next line if you want to keep your containers running after VS Code shuts down.
-	// "shutdownAction": "none",
-
-	// Uncomment the next line to run commands after the container is created.
-	"postCreateCommand": "make prepare"
-
-	// Configure tool-specific properties.
-	// "customizations": {},
-
-	// Uncomment to connect as an existing user other than the container default. More info: https://aka.ms/dev-containers-non-root.
-	// "remoteUser": "devcontainer"
-}

.devcontainer/docker-compose.yml

@@ -1,26 +0,0 @@
-version: '3.6'
-services:
-  # Update this to the name of the service you want to work with in your docker-compose.yml file
-  api:
-    # Uncomment if you want to override the service's Dockerfile to one in the .devcontainer 
-    # folder. Note that the path of the Dockerfile and context is relative to the *primary* 
-    # docker-compose.yml file (the first in the devcontainer.json "dockerComposeFile"
-    # array). The sample below assumes your primary file is in the root of your project.
-    #
-    build:
-      context: .
-      dockerfile: .devcontainer/Dockerfile
-
-    volumes:
-      # Update this to wherever you want VS Code to mount the folder of your project
-      - .:/workspace:cached
-
-    # Uncomment the next four lines if you will use a ptrace-based debugger like C++, Go, and Rust.
-    # cap_add:
-    #   - SYS_PTRACE
-    # security_opt:
-    #   - seccomp:unconfined
-
-    # Overrides default command so things don't shut down after the process ends.
-    command: /bin/sh -c "while sleep 1000; do :; done"
- 

.dockerignore

@@ -1,3 +1,4 @@
+.idea
 models
 examples/chatbot-ui/models
 examples/rwkv/models

.env

@@ -7,20 +7,33 @@
 
 ## 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
 
 ## Enable debug mode
 # DEBUG=true
 
-## Specify a build type. Available: cublas, openblas.
+## Specify a build type. Available: cublas, openblas, clblas.
 # BUILD_TYPE=openblas
 
-## Uncomment and set to false to disable rebuilding from source
-# REBUILD=false
+## Uncomment and set to true to enable rebuilding from source
+# REBUILD=true
 
-## Enable image generation with stablediffusion (requires 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

.github/FUNDING.yml

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

.github/workflows/bump_deps.yaml

@@ -12,6 +12,9 @@ jobs:
           - repository: "go-skynet/go-llama.cpp"
             variable: "GOLLAMA_VERSION"
             branch: "master"
+          - repository: "go-skynet/go-llama.cpp"
+            variable: "GOLLAMA_GRAMMAR_VERSION"
+            branch: "master"
           - repository: "go-skynet/go-ggml-transformers.cpp"
             variable: "GOGGMLTRANSFORMERS_VERSION"
             branch: "master"
@@ -30,6 +33,15 @@ jobs:
           - 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@v3

.github/workflows/image.yml

@@ -15,34 +15,97 @@ concurrency:
 
 jobs:
   docker:
+    strategy:
+      matrix:
+        include:
+          - build-type: ''
+            platforms: 'linux/amd64,linux/arm64'
+            tag-latest: 'auto'
+            tag-suffix: ''
+            ffmpeg: ''
+          - build-type: 'cublas'
+            cuda-major-version: 11
+            cuda-minor-version: 7
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda11'
+            ffmpeg: ''
+          - build-type: 'cublas'
+            cuda-major-version: 12
+            cuda-minor-version: 1
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda12'
+            ffmpeg: ''
+          - build-type: ''
+            platforms: 'linux/amd64,linux/arm64'
+            tag-latest: 'false'
+            tag-suffix: '-ffmpeg'
+            ffmpeg: 'true'
+          - build-type: 'cublas'
+            cuda-major-version: 11
+            cuda-minor-version: 7
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda11-ffmpeg'
+            ffmpeg: 'true'
+          - build-type: 'cublas'
+            cuda-major-version: 12
+            cuda-minor-version: 1
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda12-ffmpeg'
+            ffmpeg: 'true'
+
     runs-on: ubuntu-latest
     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: 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: Docker meta
+        id: meta
+        uses: docker/metadata-action@v4
+        with:
+          images: quay.io/go-skynet/local-ai
+          tags: |
+            type=ref,event=branch
+            type=semver,pattern={{raw}}
+            type=sha
+          flavor: |
+            latest=${{ matrix.tag-latest }}
+            suffix=${{ matrix.tag-suffix }}
 
       - name: Set up QEMU
         uses: docker/setup-qemu-action@master
@@ -60,23 +123,19 @@ jobs:
           registry: quay.io
           username: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
           password: ${{ secrets.LOCALAI_REGISTRY_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'
+
+      - name: Build and push
         uses: docker/build-push-action@v4
         with:
           builder: ${{ steps.buildx.outputs.name }}
+          build-args: |
+            BUILD_TYPE=${{ matrix.build-type }}
+            CUDA_MAJOR_VERSION=${{ matrix.cuda-major-version }}
+            CUDA_MINOR_VERSION=${{ matrix.cuda-minor-version }}
+            FFMPEG=${{ matrix.ffmpeg }}
           context: .
           file: ./Dockerfile
-          platforms: linux/amd64
-          push: false
-          tags: ${{ steps.prep.outputs.tags }}
+          platforms: ${{ matrix.platforms }}
+          push: ${{ github.event_name != 'pull_request' }}
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}

.github/workflows/release.yaml

@@ -29,10 +29,10 @@ jobs:
       - name: Build
         id: build
         env:
-          CMAKE_ARGS: "${{ matrix.define }}"
+          CMAKE_ARGS: "${{ matrix.defines }}"
           BUILD_ID: "${{ matrix.build }}"
         run: |
-          make dist
+          STATIC=true make dist
       - uses: actions/upload-artifact@v3
         with:
           name: ${{ matrix.build }}
@@ -60,15 +60,10 @@ jobs:
         uses: actions/checkout@v3
         with:
           submodules: true
-
-      - name: Dependencies
-        run: |
-          brew update
-          brew install sdl2 ffmpeg
       - name: Build
         id: build
         env:
-          CMAKE_ARGS: "${{ matrix.define }}"
+          CMAKE_ARGS: "${{ matrix.defines }}"
           BUILD_ID: "${{ matrix.build }}"
         run: |
           make dist

.github/workflows/test.yml

@@ -26,9 +26,29 @@ jobs:
         run: |
           sudo apt-get update
           sudo apt-get install build-essential ffmpeg
+          
+          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 pip install -r extra/requirements.txt
+
+          sudo mkdir /build && sudo chmod -R 777 /build && cd /build && \
+          curl -L "https://github.com/gabime/spdlog/archive/refs/tags/v1.11.0.tar.gz" | \
+          tar -xzvf - && \
+          mkdir -p "spdlog-1.11.0/build" && \
+          cd "spdlog-1.11.0/build" && \
+          cmake ..  && \
+          make -j8 && \
+          sudo cmake --install . --prefix /usr && mkdir -p "lib/Linux-$(uname -m)" && \
+          cd /build && \
+          mkdir -p "lib/Linux-$(uname -m)/piper_phonemize" && \
+          curl -L "https://github.com/rhasspy/piper-phonemize/releases/download/v1.0.0/libpiper_phonemize-amd64.tar.gz" | \
+          tar -C "lib/Linux-$(uname -m)/piper_phonemize" -xzvf - && ls -liah /build/lib/Linux-$(uname -m)/piper_phonemize/ && \
+          sudo cp -rfv /build/lib/Linux-$(uname -m)/piper_phonemize/lib/. /lib64/ && \
+          sudo cp -rfv /build/lib/Linux-$(uname -m)/piper_phonemize/lib/. /usr/lib/ && \
+          sudo cp -rfv /build/lib/Linux-$(uname -m)/piper_phonemize/include/. /usr/include/
       - name: Test
         run: |
-          make test
+          ESPEAK_DATA="/build/lib/Linux-$(uname -m)/piper_phonemize/lib/espeak-ng-data" GO_TAGS="tts stablediffusion" make test
 
   macOS-latest:
     runs-on: macOS-latest
@@ -39,10 +59,6 @@ jobs:
         with: 
           submodules: true
 
-      - name: Dependencies
-        run: |
-          brew update
-          brew install sdl2 ffmpeg
       - name: Test
         run: |
-          make test
+          CMAKE_ARGS="-DLLAMA_F16C=OFF -DLLAMA_AVX512=OFF -DLLAMA_AVX2=OFF -DLLAMA_FMA=OFF" make test

.gitignore

@@ -1,10 +1,21 @@
 # go-llama build artifacts
 go-llama
-gpt4all
+/gpt4all
 go-stable-diffusion
+go-piper
+/go-bert
+go-ggllm
+/piper
+__pycache__/
+*.a
+get-sources
+
+go-ggml-transformers
 go-gpt2
 go-rwkv
 whisper.cpp
+/bloomz
+go-bert
 
 # LocalAI build binary
 LocalAI
@@ -15,8 +26,15 @@ local-ai
 # Ignore models
 models/*
 test-models/
+test-dir/
 
 release/
 
 # just in case
 .DS_Store
+.idea
+
+# Generated during build
+backend-assets/
+
+/ggml-metal.metal

.vscode/launch.json

@@ -25,7 +25,7 @@
             ],
             "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:${workspaceFolder}/bloomz",
+                "LIBRARY_PATH": "${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"
             }
         }

Dockerfile

@@ -1,15 +1,125 @@
-ARG GO_VERSION=1.20
-ARG BUILD_TYPE=
-FROM golang:$GO_VERSION
-ENV REBUILD=true
+ARG GO_VERSION=1.20-bullseye
+
+FROM golang:$GO_VERSION as requirements
+
+ARG BUILD_TYPE
+ARG CUDA_MAJOR_VERSION=11
+ARG CUDA_MINOR_VERSION=7
+ARG SPDLOG_VERSION="1.11.0"
+ARG PIPER_PHONEMIZE_VERSION='1.0.0'
+ARG TARGETARCH
+ARG TARGETVARIANT
+
+ENV BUILD_TYPE=${BUILD_TYPE}
+ENV EXTERNAL_GRPC_BACKENDS="huggingface-embeddings:/build/extra/grpc/huggingface/huggingface.py"
+ARG GO_TAGS="stablediffusion tts"
+
+RUN apt-get update && \
+    apt-get install -y ca-certificates cmake curl patch pip
+
+# Extras requirements
+COPY extra/requirements.txt /build/extra/requirements.txt
+RUN pip install -r /build/extra/requirements.txt && rm -rf /build/extra/requirements.txt
+
+# CuBLAS requirements
+RUN if [ "${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} \
+    ; fi
+ENV PATH /usr/local/cuda/bin:${PATH}
+
 WORKDIR /build
-RUN apt-get update && apt-get install -y cmake curl libgomp1 libopenblas-dev libopenblas-base libopencv-dev libopencv-core-dev libopencv-core4.5 ca-certificates
+
+# OpenBLAS requirements
+RUN apt-get install -y libopenblas-dev
+
+# Stable Diffusion requirements
+RUN apt-get install -y libopencv-dev && \
+    ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2
+
+# Use the variables in subsequent instructions
+RUN echo "Target Architecture: $TARGETARCH"
+RUN echo "Target Variant: $TARGETVARIANT"
+
+# piper requirements
+# Use pre-compiled Piper phonemization library (includes onnxruntime)
+#RUN if echo "${GO_TAGS}" | grep -q "tts"; then \
+RUN test -n "$TARGETARCH" \
+    || (echo 'warn: missing $TARGETARCH, either set this `ARG` manually, or run using `docker buildkit`')
+
+RUN curl -L "https://github.com/gabime/spdlog/archive/refs/tags/v${SPDLOG_VERSION}.tar.gz" | \
+    tar -xzvf - && \
+    mkdir -p "spdlog-${SPDLOG_VERSION}/build" && \
+    cd "spdlog-${SPDLOG_VERSION}/build" && \
+    cmake ..  && \
+    make -j8 && \
+    cmake --install . --prefix /usr && mkdir -p "lib/Linux-$(uname -m)" && \
+    cd /build && \
+    mkdir -p "lib/Linux-$(uname -m)/piper_phonemize" && \
+    curl -L "https://github.com/rhasspy/piper-phonemize/releases/download/v${PIPER_PHONEMIZE_VERSION}/libpiper_phonemize-${TARGETARCH:-$(go env GOARCH)}${TARGETVARIANT}.tar.gz" | \
+    tar -C "lib/Linux-$(uname -m)/piper_phonemize" -xzvf - && ls -liah /build/lib/Linux-$(uname -m)/piper_phonemize/ && \
+    cp -rfv /build/lib/Linux-$(uname -m)/piper_phonemize/lib/. /lib64/ && \
+    cp -rfv /build/lib/Linux-$(uname -m)/piper_phonemize/lib/. /usr/lib/ && \
+    cp -rfv /build/lib/Linux-$(uname -m)/piper_phonemize/include/. /usr/include/
+# \
+#    ; fi
+
+###################################
+###################################
+
+FROM requirements as builder
+
+ARG GO_TAGS="stablediffusion tts"
+
+ENV GO_TAGS=${GO_TAGS}
+ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
+ENV NVIDIA_REQUIRE_CUDA="cuda>=${CUDA_MAJOR_VERSION}.0"
+ENV NVIDIA_VISIBLE_DEVICES=all
+
+WORKDIR /build
+
+COPY Makefile .
+RUN make get-sources
+COPY go.mod .
+RUN make prepare
 COPY . .
-RUN ln -s /usr/include/opencv4/opencv2/ /usr/include/opencv2
-RUN make build
+COPY .git .
+
+RUN ESPEAK_DATA=/build/lib/Linux-$(uname -m)/piper_phonemize/lib/espeak-ng-data make build
+
+###################################
+###################################
+
+FROM requirements
+
+ARG FFMPEG
+
+ENV REBUILD=false
 ENV HEALTHCHECK_ENDPOINT=http://localhost:8080/readyz
+
+# Add FFmpeg
+RUN if [ "${FFMPEG}" = "true" ]; then \
+    apt-get install -y ffmpeg \
+    ; 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 . .
+RUN make prepare-sources
+COPY --from=builder /build/local-ai ./
+
 # Define the health check command
-HEALTHCHECK --interval=30s --timeout=360s --retries=10 \
+HEALTHCHECK --interval=1m --timeout=10m --retries=10 \
   CMD curl -f $HEALTHCHECK_ENDPOINT || exit 1
+
 EXPOSE 8080
 ENTRYPOINT [ "/build/entrypoint.sh" ]

Dockerfile.dev

@@ -1,20 +0,0 @@
-ARG GO_VERSION=1.20
-ARG DEBIAN_VERSION=11
-ARG BUILD_TYPE=
-
-FROM golang:$GO_VERSION as builder
-WORKDIR /build
-RUN apt-get update && apt-get install -y cmake libgomp1 libopenblas-dev libopenblas-base libopencv-dev libopencv-core-dev libopencv-core4.5 
-RUN ln -s /usr/include/opencv4/opencv2/ /usr/include/opencv2
-COPY . .
-RUN make build
-
-FROM debian:$DEBIAN_VERSION
-COPY --from=builder /build/local-ai /usr/bin/local-ai
-RUN apt-get update && apt-get install -y ca-certificates curl
-ENV HEALTHCHECK_ENDPOINT=http://localhost:8080/readyz
-# Define the health check command
-HEALTHCHECK --interval=30s --timeout=360s --retries=10 \
-  CMD curl -f $HEALTHCHECK_ENDPOINT || exit 1
-EXPOSE 8080
-ENTRYPOINT [ "/usr/bin/local-ai" ]

LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2023 go-skynet authors
+Copyright (c) 2023 Ettore Di Giacinto
 
 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

@@ -3,23 +3,62 @@ GOTEST=$(GOCMD) test
 GOVET=$(GOCMD) vet
 BINARY_NAME=local-ai
 
-GOLLAMA_VERSION?=8bd97d532e90cf34e755b3ea2d8aa17000443cf2
+# llama.cpp versions
+# Temporarly pinned to https://github.com/go-skynet/go-llama.cpp/pull/124
+GOLLAMA_VERSION?=f3a6ee0ef53d667f110d28fcf9b808bdca741c07
+
+GOLLAMA_GRAMMAR_VERSION?=cb8d7cd4cb95725a04504a9e3a26dd72a12b69ac
+# Temporary set a specific version of llama.cpp
+# containing: https://github.com/ggerganov/llama.cpp/pull/1773 and
+# rebased on top of master.
+# This pin can be dropped when the PR above is merged, and go-llama has merged changes as well
+# Set empty to use the version pinned by go-llama
+LLAMA_CPP_GRAMMAR_REPO?=https://github.com/mudler/llama.cpp
+LLAMA_CPP_GRAMMAR_VERSION?=48ce8722a05a018681634af801fd0fd45b3a87cc
+
+# gpt4all version
 GPT4ALL_REPO?=https://github.com/nomic-ai/gpt4all
-GPT4ALL_VERSION?=810a3b12ccd0b0ee82b55d7ddbcb1db7d345dd06
-GOGGMLTRANSFORMERS_VERSION?=c4c581f1853cf1b66276501c7c0dbea1e3e564b7
+GPT4ALL_VERSION?=5f0aaf8bdb166ea3b5bfd578c2b19f61b583e6a9
+
+# go-ggml-transformers version
+GOGGMLTRANSFORMERS_VERSION?=ffb09d7dd71e2cbc6c5d7d05357d230eea6f369a
+
+# go-rwkv version
 RWKV_REPO?=https://github.com/donomii/go-rwkv.cpp
-RWKV_VERSION?=07166da10cb2a9e8854395a4f210464dcea76e47
-WHISPER_CPP_VERSION?=77eab3fbfe5e5462021d92dd230076bba06eefbc
-BERT_VERSION?=cea1ed76a7f48ef386a8e369f6c82c48cdf2d551
-BLOOMZ_VERSION?=e9366e82abdfe70565644fbfae9651976714efd1
-BUILD_TYPE?=
+RWKV_VERSION?=c898cd0f62df8f2a7830e53d1d513bef4f6f792b
+
+# whisper.cpp version
+WHISPER_CPP_VERSION?=85ed71aaec8e0612a84c0b67804bde75aa75a273
+
+# bert.cpp version
+BERT_VERSION?=6abe312cded14042f6b7c3cd8edf082713334a4d
+
+# go-piper version
+PIPER_VERSION?=56b8a81b4760a6fbee1a82e62f007ae7e8f010a7
+
+# go-bloomz version
+BLOOMZ_VERSION?=1834e77b83faafe912ad4092ccf7f77937349e2f
+
+# stablediffusion version
+STABLEDIFFUSION_VERSION?=d89260f598afb809279bc72aa0107b4292587632
+
+# Go-ggllm
+GOGGLLM_VERSION?=862477d16eefb0805261c19c9b0d053e3b2b684b
+
+export BUILD_TYPE?=
 CGO_LDFLAGS?=
 CUDA_LIBPATH?=/usr/local/cuda/lib64/
-STABLEDIFFUSION_VERSION?=c0748eca3642d58bcf9521108bcee46959c647dc
 GO_TAGS?=
 BUILD_ID?=git
-LD_FLAGS=?=
+
+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?=
+ESPEAK_DATA?=
 
 OS := $(shell uname -s)
 ARCH := $(shell uname -m)
@@ -29,8 +68,14 @@ WHITE  := $(shell tput -Txterm setaf 7)
 CYAN   := $(shell tput -Txterm setaf 6)
 RESET  := $(shell tput -Txterm sgr0)
 
-C_INCLUDE_PATH=$(shell pwd)/go-llama:$(shell pwd)/go-stable-diffusion/:$(shell pwd)/gpt4all/gpt4all-bindings/golang/:$(shell pwd)/go-ggml-transformers:$(shell pwd)/go-rwkv:$(shell pwd)/whisper.cpp:$(shell pwd)/go-bert:$(shell pwd)/bloomz
-LIBRARY_PATH=$(shell pwd)/go-llama:$(shell pwd)/go-stable-diffusion/:$(shell pwd)/gpt4all/gpt4all-bindings/golang/:$(shell pwd)/go-ggml-transformers:$(shell pwd)/go-rwkv:$(shell pwd)/whisper.cpp:$(shell pwd)/go-bert:$(shell pwd)/bloomz
+ifndef UNAME_S
+UNAME_S := $(shell uname -s)
+endif
+
+# workaround for rwkv.cpp
+ifeq ($(UNAME_S),Darwin)
+        CGO_LDFLAGS += -lcblas -framework Accelerate 
+endif
 
 ifeq ($(BUILD_TYPE),openblas)
 	CGO_LDFLAGS+=-lopenblas
@@ -41,13 +86,29 @@ ifeq ($(BUILD_TYPE),cublas)
 	export LLAMA_CUBLAS=1
 endif
 
+ifeq ($(BUILD_TYPE),metal)
+	CGO_LDFLAGS+=-framework Foundation -framework Metal -framework MetalKit -framework MetalPerformanceShaders
+	export LLAMA_METAL=1
+endif
+
+ifeq ($(BUILD_TYPE),clblas)
+	CGO_LDFLAGS+=-lOpenCL -lclblast
+endif
+
 # glibc-static or glibc-devel-static required
 ifeq ($(STATIC),true)
 	LD_FLAGS=-linkmode external -extldflags -static
 endif
 
-ifeq ($(GO_TAGS),stablediffusion)
-	OPTIONAL_TARGETS+=go-stable-diffusion/libstablediffusion.a
+ifeq ($(findstring stablediffusion,$(GO_TAGS)),stablediffusion)
+#	OPTIONAL_TARGETS+=go-stable-diffusion/libstablediffusion.a
+	OPTIONAL_GRPC+=backend-assets/grpc/stablediffusion
+endif
+
+ifeq ($(findstring tts,$(GO_TAGS)),tts)
+#	OPTIONAL_TARGETS+=go-piper/libpiper_binding.a
+#	OPTIONAL_TARGETS+=backend-assets/espeak-ng-data
+	OPTIONAL_GRPC+=backend-assets/grpc/piper
 endif
 
 .PHONY: all test build vendor
@@ -58,31 +119,24 @@ all: help
 gpt4all:
 	git clone --recurse-submodules $(GPT4ALL_REPO) gpt4all
 	cd gpt4all && git checkout -b build $(GPT4ALL_VERSION) && git submodule update --init --recursive --depth 1
-	# This is hackish, but needed as both go-llama and go-gpt4allj have their own version of ggml..
-	@find ./gpt4all -type f -name "*.c" -exec sed -i'' -e 's/ggml_/ggml_gptj_/g' {} +
-	@find ./gpt4all -type f -name "*.cpp" -exec sed -i'' -e 's/ggml_/ggml_gptj_/g' {} +
-	@find ./gpt4all -type f -name "*.h" -exec sed -i'' -e 's/ggml_/ggml_gptj_/g' {} +
-	@find ./gpt4all -type f -name "*.cpp" -exec sed -i'' -e 's/gpt_/gptj_/g' {} +
-	@find ./gpt4all -type f -name "*.h" -exec sed -i'' -e 's/gpt_/gptj_/g' {} +
-	@find ./gpt4all -type f -name "*.h" -exec sed -i'' -e 's/set_console_color/set_gptj_console_color/g' {} +
-	@find ./gpt4all -type f -name "*.cpp" -exec sed -i'' -e 's/set_console_color/set_gptj_console_color/g' {} +
-	@find ./gpt4all -type f -name "*.cpp" -exec sed -i'' -e 's/llama_/gptjllama_/g' {} +
-	@find ./gpt4all -type f -name "*.go" -exec sed -i'' -e 's/llama_/gptjllama_/g' {} +
-	@find ./gpt4all -type f -name "*.h" -exec sed -i'' -e 's/llama_/gptjllama_/g' {} +
-	@find ./gpt4all -type f -name "*.txt" -exec sed -i'' -e 's/llama_/gptjllama_/g' {} +
-	@find ./gpt4all -type f -name "*.cpp" -exec sed -i'' -e 's/json_/json_gptj_/g' {} +
-	@find ./gpt4all -type f -name "*.cpp" -exec sed -i'' -e 's/void replace/void json_gptj_replace/g' {} +
-	@find ./gpt4all -type f -name "*.cpp" -exec sed -i'' -e 's/::replace/::json_gptj_replace/g' {} +
-	@find ./gpt4all -type f -name "*.cpp" -exec sed -i'' -e 's/regex_escape/gpt4allregex_escape/g' {} +
-	mv ./gpt4all/gpt4all-backend/llama.cpp/llama_util.h ./gpt4all/gpt4all-backend/llama.cpp/gptjllama_util.h
+
+## go-ggllm
+go-ggllm:
+	git clone --recurse-submodules https://github.com/mudler/go-ggllm.cpp go-ggllm
+	cd go-ggllm && git checkout -b build $(GOGGLLM_VERSION) && git submodule update --init --recursive --depth 1
+
+go-ggllm/libggllm.a: go-ggllm
+	$(MAKE) -C go-ggllm BUILD_TYPE=$(BUILD_TYPE) libggllm.a
+
+## go-piper
+go-piper:
+	git clone --recurse-submodules https://github.com/mudler/go-piper go-piper
+	cd go-piper && git checkout -b build $(PIPER_VERSION) && git submodule update --init --recursive --depth 1
 
 ## BERT embeddings
 go-bert:
 	git clone --recurse-submodules https://github.com/go-skynet/go-bert.cpp go-bert
 	cd go-bert && git checkout -b build $(BERT_VERSION) && git submodule update --init --recursive --depth 1
-	@find ./go-bert -type f -name "*.c" -exec sed -i'' -e 's/ggml_/ggml_bert_/g' {} +
-	@find ./go-bert -type f -name "*.cpp" -exec sed -i'' -e 's/ggml_/ggml_bert_/g' {} +
-	@find ./go-bert -type f -name "*.h" -exec sed -i'' -e 's/ggml_/ggml_bert_/g' {} +
 
 ## stable diffusion
 go-stable-diffusion:
@@ -96,21 +150,14 @@ go-stable-diffusion/libstablediffusion.a:
 go-rwkv:
 	git clone --recurse-submodules $(RWKV_REPO) go-rwkv
 	cd go-rwkv && git checkout -b build $(RWKV_VERSION) && git submodule update --init --recursive --depth 1
-	@find ./go-rwkv -type f -name "*.c" -exec sed -i'' -e 's/ggml_/ggml_rwkv_/g' {} +
-	@find ./go-rwkv -type f -name "*.cpp" -exec sed -i'' -e 's/ggml_/ggml_rwkv_/g' {} +
-	@find ./go-rwkv -type f -name "*.h" -exec sed -i'' -e 's/ggml_/ggml_rwkv_/g' {} +
 
 go-rwkv/librwkv.a: go-rwkv
-	cd go-rwkv && cd rwkv.cpp &&	cmake . -DRWKV_BUILD_SHARED_LIBRARY=OFF &&	cmake --build . && 	cp librwkv.a .. && cp ggml/src/libggml.a ..
+	cd go-rwkv && cd rwkv.cpp &&	cmake . -DRWKV_BUILD_SHARED_LIBRARY=OFF &&	cmake --build . && 	cp librwkv.a ..
 
 ## bloomz
 bloomz:
 	git clone --recurse-submodules https://github.com/go-skynet/bloomz.cpp bloomz
-	@find ./bloomz -type f -name "*.c" -exec sed -i'' -e 's/ggml_/ggml_bloomz_/g' {} +
-	@find ./bloomz -type f -name "*.cpp" -exec sed -i'' -e 's/ggml_/ggml_bloomz_/g' {} +
-	@find ./bloomz -type f -name "*.h" -exec sed -i'' -e 's/ggml_/ggml_bloomz_/g' {} +
-	@find ./bloomz -type f -name "*.cpp" -exec sed -i'' -e 's/gpt_/gpt_bloomz_/g' {} +
-	@find ./bloomz -type f -name "*.h" -exec sed -i'' -e 's/gpt_/gpt_bloomz_/g' {} +
+	cd bloomz && git checkout -b build $(BLOOMZ_VERSION) && git submodule update --init --recursive --depth 1
 
 bloomz/libbloomz.a: bloomz
 	cd bloomz && make libbloomz.a
@@ -118,6 +165,21 @@ bloomz/libbloomz.a: bloomz
 go-bert/libgobert.a: go-bert
 	$(MAKE) -C go-bert libgobert.a
 
+backend-assets/gpt4all: gpt4all/gpt4all-bindings/golang/libgpt4all.a
+	mkdir -p backend-assets/gpt4all
+	@cp gpt4all/gpt4all-bindings/golang/buildllm/*.so backend-assets/gpt4all/ || true
+	@cp gpt4all/gpt4all-bindings/golang/buildllm/*.dylib backend-assets/gpt4all/ || true
+	@cp gpt4all/gpt4all-bindings/golang/buildllm/*.dll backend-assets/gpt4all/ || true
+
+backend-assets/espeak-ng-data:
+	mkdir -p backend-assets/espeak-ng-data
+ifdef ESPEAK_DATA
+	@cp -rf $(ESPEAK_DATA)/. backend-assets/espeak-ng-data
+else
+	@echo "ESPEAK_DATA not set, skipping tts. Note that this will break the tts functionality."
+	@touch backend-assets/espeak-ng-data/keep
+endif
+
 gpt4all/gpt4all-bindings/golang/libgpt4all.a: gpt4all
 	$(MAKE) -C gpt4all/gpt4all-bindings/golang/ libgpt4all.a
 
@@ -125,27 +187,13 @@ gpt4all/gpt4all-bindings/golang/libgpt4all.a: gpt4all
 go-ggml-transformers:
 	git clone --recurse-submodules https://github.com/go-skynet/go-ggml-transformers.cpp go-ggml-transformers
 	cd go-ggml-transformers && git checkout -b build $(GOGPT2_VERSION) && git submodule update --init --recursive --depth 1
-	# This is hackish, but needed as both go-llama and go-gpt4allj have their own version of ggml..
-	@find ./go-ggml-transformers -type f -name "*.c" -exec sed -i'' -e 's/ggml_/ggml_gpt2_/g' {} +
-	@find ./go-ggml-transformers -type f -name "*.cpp" -exec sed -i'' -e 's/ggml_/ggml_gpt2_/g' {} +
-	@find ./go-ggml-transformers -type f -name "*.h" -exec sed -i'' -e 's/ggml_/ggml_gpt2_/g' {} +
-	@find ./go-ggml-transformers -type f -name "*.cpp" -exec sed -i'' -e 's/gpt_print_usage/gpt2_print_usage/g' {} +
-	@find ./go-ggml-transformers -type f -name "*.h" -exec sed -i'' -e 's/gpt_print_usage/gpt2_print_usage/g' {} +
-	@find ./go-ggml-transformers -type f -name "*.cpp" -exec sed -i'' -e 's/gpt_params_parse/gpt2_params_parse/g' {} +
-	@find ./go-ggml-transformers -type f -name "*.h" -exec sed -i'' -e 's/gpt_params_parse/gpt2_params_parse/g' {} +
-	@find ./go-ggml-transformers -type f -name "*.cpp" -exec sed -i'' -e 's/gpt_random_prompt/gpt2_random_prompt/g' {} +
-	@find ./go-ggml-transformers -type f -name "*.h" -exec sed -i'' -e 's/gpt_random_prompt/gpt2_random_prompt/g' {} +
-	@find ./go-ggml-transformers -type f -name "*.cpp" -exec sed -i'' -e 's/json_/json_gpt2_/g' {} +
 
 go-ggml-transformers/libtransformers.a: go-ggml-transformers
-	$(MAKE) -C go-ggml-transformers libtransformers.a
+	$(MAKE) -C go-ggml-transformers BUILD_TYPE=$(BUILD_TYPE) libtransformers.a
 
 whisper.cpp:
 	git clone https://github.com/ggerganov/whisper.cpp.git
 	cd whisper.cpp && git checkout -b build $(WHISPER_CPP_VERSION) && git submodule update --init --recursive --depth 1
-	@find ./whisper.cpp -type f -name "*.c" -exec sed -i'' -e 's/ggml_/ggml_whisper_/g' {} +
-	@find ./whisper.cpp -type f -name "*.cpp" -exec sed -i'' -e 's/ggml_/ggml_whisper_/g' {} +
-	@find ./whisper.cpp -type f -name "*.h" -exec sed -i'' -e 's/ggml_/ggml_whisper_/g' {} +
 
 whisper.cpp/libwhisper.a: whisper.cpp
 	cd whisper.cpp && make libwhisper.a
@@ -154,11 +202,28 @@ go-llama:
 	git clone --recurse-submodules https://github.com/go-skynet/go-llama.cpp go-llama
 	cd go-llama && git checkout -b build $(GOLLAMA_VERSION) && git submodule update --init --recursive --depth 1
 
+go-llama-grammar:
+	git clone --recurse-submodules https://github.com/go-skynet/go-llama.cpp go-llama-grammar
+	cd go-llama-grammar && git checkout -b build $(GOLLAMA_GRAMMAR_VERSION) && git submodule update --init --recursive --depth 1
+ifneq ($(LLAMA_CPP_GRAMMAR_REPO),)
+	cd go-llama-grammar && rm -rf llama.cpp && git clone $(LLAMA_CPP_GRAMMAR_REPO) llama.cpp && cd llama.cpp && git checkout -b build $(LLAMA_CPP_GRAMMAR_VERSION) && git submodule update --init --recursive --depth 1
+endif
+
 go-llama/libbinding.a: go-llama
 	$(MAKE) -C go-llama BUILD_TYPE=$(BUILD_TYPE) libbinding.a
 
+go-llama-grammar/libbinding.a: go-llama-grammar
+	$(MAKE) -C go-llama-grammar BUILD_TYPE=$(BUILD_TYPE) libbinding.a
+
+go-piper/libpiper_binding.a:
+	$(MAKE) -C go-piper libpiper_binding.a example/main
+
+get-sources: go-llama go-ggllm go-llama-grammar go-ggml-transformers gpt4all go-piper go-rwkv whisper.cpp go-bert bloomz go-stable-diffusion
+	touch $@
+
 replace:
 	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(shell pwd)/go-llama
+	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp-grammar=$(shell pwd)/go-llama-grammar
 	$(GOCMD) mod edit -replace github.com/nomic-ai/gpt4all/gpt4all-bindings/golang=$(shell pwd)/gpt4all/gpt4all-bindings/golang
 	$(GOCMD) mod edit -replace github.com/go-skynet/go-ggml-transformers.cpp=$(shell pwd)/go-ggml-transformers
 	$(GOCMD) mod edit -replace github.com/donomii/go-rwkv.cpp=$(shell pwd)/go-rwkv
@@ -166,13 +231,17 @@ replace:
 	$(GOCMD) mod edit -replace github.com/go-skynet/go-bert.cpp=$(shell pwd)/go-bert
 	$(GOCMD) mod edit -replace github.com/go-skynet/bloomz.cpp=$(shell pwd)/bloomz
 	$(GOCMD) mod edit -replace github.com/mudler/go-stable-diffusion=$(shell pwd)/go-stable-diffusion
+	$(GOCMD) mod edit -replace github.com/mudler/go-piper=$(shell pwd)/go-piper
+	$(GOCMD) mod edit -replace github.com/mudler/go-ggllm.cpp=$(shell pwd)/go-ggllm
 
-prepare-sources: go-llama go-ggml-transformers gpt4all go-rwkv whisper.cpp go-bert bloomz go-stable-diffusion replace
+prepare-sources: get-sources replace
 	$(GOCMD) mod download
 
 ## GENERIC
 rebuild: ## Rebuilds the project
+	$(GOCMD) clean -cache
 	$(MAKE) -C go-llama clean
+	$(MAKE) -C go-llama-grammar clean
 	$(MAKE) -C gpt4all/gpt4all-bindings/golang/ clean
 	$(MAKE) -C go-ggml-transformers clean
 	$(MAKE) -C go-rwkv clean
@@ -180,40 +249,47 @@ rebuild: ## Rebuilds the project
 	$(MAKE) -C go-stable-diffusion clean
 	$(MAKE) -C go-bert clean
 	$(MAKE) -C bloomz clean
+	$(MAKE) -C go-piper clean
+	$(MAKE) -C go-ggllm clean
 	$(MAKE) build
 
-prepare: prepare-sources gpt4all/gpt4all-bindings/golang/libgpt4all.a $(OPTIONAL_TARGETS) go-llama/libbinding.a go-bert/libgobert.a go-ggml-transformers/libtransformers.a go-rwkv/librwkv.a whisper.cpp/libwhisper.a bloomz/libbloomz.a  ## Prepares for building
+prepare: prepare-sources $(OPTIONAL_TARGETS) 
+	touch $@
 
 clean: ## Remove build related file
+	$(GOCMD) clean -cache
 	rm -fr ./go-llama
 	rm -rf ./gpt4all
+	rm -rf ./go-gpt2
 	rm -rf ./go-stable-diffusion
 	rm -rf ./go-ggml-transformers
+	rm -rf ./backend-assets
 	rm -rf ./go-rwkv
 	rm -rf ./go-bert
 	rm -rf ./bloomz
 	rm -rf ./whisper.cpp
+	rm -rf ./go-piper
+	rm -rf ./go-ggllm
 	rm -rf $(BINARY_NAME)
 	rm -rf release/
 
 ## Build:
 
-build: prepare ## Build the project
+build: 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)${RESET})
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=${C_INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} $(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -x -o $(BINARY_NAME) ./
+	$(info ${GREEN}I LD_FLAGS: ${YELLOW}$(LD_FLAGS)${RESET})
+
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o $(BINARY_NAME) ./
 
 dist: build
 	mkdir -p release
 	cp $(BINARY_NAME) release/$(BINARY_NAME)-$(BUILD_ID)-$(OS)-$(ARCH)
 
-generic-build: ## Build the project using generic
-	BUILD_TYPE="generic" $(MAKE) build
-
 ## Run
 run: prepare ## run local-ai
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=${C_INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} $(GOCMD) run ./main.go
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GOCMD) run ./
 
 test-models/testmodel:
 	mkdir test-models
@@ -222,13 +298,44 @@ test-models/testmodel:
 	wget https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin -O test-models/whisper-en
 	wget https://huggingface.co/skeskinen/ggml/resolve/main/all-MiniLM-L6-v2/ggml-model-q4_0.bin -O test-models/bert
 	wget https://cdn.openai.com/whisper/draft-20220913a/micro-machines.wav -O test-dir/audio.wav
-	wget https://huggingface.co/imxcstar/rwkv-4-raven-ggml/resolve/main/RWKV-4-Raven-1B5-v11-Eng99%25-Other1%25-20230425-ctx4096-16_Q4_2.bin -O test-models/rwkv
+	wget 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 https://raw.githubusercontent.com/saharNooby/rwkv.cpp/5eb8f09c146ea8124633ab041d9ea0b1f1db4459/rwkv/20B_tokenizer.json -O test-models/rwkv.tokenizer.json
 	cp tests/models_fixtures/* test-models
 
-test: prepare test-models/testmodel
+prepare-test: grpcs
+	cp -rf backend-assets api
 	cp tests/models_fixtures/* test-models
-	C_INCLUDE_PATH=${C_INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} 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 --flakeAttempts 5 -v -r ./api ./pkg
+
+test: prepare test-models/testmodel grpcs
+	@echo 'Running tests'
+	export GO_TAGS="tts stablediffusion"
+	$(MAKE) prepare-test
+	HUGGINGFACE_GRPC=$(abspath ./)/extra/grpc/huggingface/huggingface.py 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" --flake-attempts 5 -v -r ./api ./pkg
+	$(MAKE) test-gpt4all
+	$(MAKE) test-llama
+	$(MAKE) test-tts
+	$(MAKE) test-stablediffusion
+
+test-gpt4all: prepare-test
+	TEST_DIR=$(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 \
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama" --flake-attempts 5 -v -r ./api ./pkg
+
+test-tts: prepare-test
+	TEST_DIR=$(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.
@@ -241,3 +348,98 @@ 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)} \
 		}' $(MAKEFILE_LIST)
+
+protogen: protogen-go protogen-python
+
+protogen-go:
+	protoc --go_out=. --go_opt=paths=source_relative --go-grpc_out=. --go-grpc_opt=paths=source_relative \
+    pkg/grpc/proto/backend.proto
+
+protogen-python:
+	python -m grpc_tools.protoc -Ipkg/grpc/proto/ --python_out=extra/grpc/huggingface/ --grpc_python_out=extra/grpc/huggingface/ pkg/grpc/proto/backend.proto
+
+## GRPC
+
+backend-assets/grpc:
+	mkdir -p backend-assets/grpc
+
+backend-assets/grpc/falcon: backend-assets/grpc go-ggllm/libggllm.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggllm LIBRARY_PATH=$(shell pwd)/go-ggllm \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/falcon ./cmd/grpc/falcon/
+
+backend-assets/grpc/llama: backend-assets/grpc go-llama/libbinding.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-llama LIBRARY_PATH=$(shell pwd)/go-llama \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/llama ./cmd/grpc/llama/
+# TODO: every binary should have its own folder instead, so can have different metal implementations
+ifeq ($(BUILD_TYPE),metal)
+	cp go-llama/build/bin/ggml-metal.metal backend-assets/grpc/
+endif
+
+backend-assets/grpc/llama-grammar: backend-assets/grpc go-llama-grammar/libbinding.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-llama-grammar LIBRARY_PATH=$(shell pwd)/go-llama-grammar \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/llama-grammar ./cmd/grpc/llama-grammar/
+
+backend-assets/grpc/gpt4all: backend-assets/grpc backend-assets/gpt4all gpt4all/gpt4all-bindings/golang/libgpt4all.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/gpt4all/gpt4all-bindings/golang/ LIBRARY_PATH=$(shell pwd)/gpt4all/gpt4all-bindings/golang/ \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gpt4all ./cmd/grpc/gpt4all/
+
+backend-assets/grpc/dolly: backend-assets/grpc go-ggml-transformers/libtransformers.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggml-transformers LIBRARY_PATH=$(shell pwd)/go-ggml-transformers \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/dolly ./cmd/grpc/dolly/
+
+backend-assets/grpc/gpt2: backend-assets/grpc go-ggml-transformers/libtransformers.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggml-transformers LIBRARY_PATH=$(shell pwd)/go-ggml-transformers \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gpt2 ./cmd/grpc/gpt2/
+
+backend-assets/grpc/gptj: backend-assets/grpc go-ggml-transformers/libtransformers.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggml-transformers LIBRARY_PATH=$(shell pwd)/go-ggml-transformers \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gptj ./cmd/grpc/gptj/
+
+backend-assets/grpc/gptneox: backend-assets/grpc go-ggml-transformers/libtransformers.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggml-transformers LIBRARY_PATH=$(shell pwd)/go-ggml-transformers \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gptneox ./cmd/grpc/gptneox/
+
+backend-assets/grpc/mpt: backend-assets/grpc go-ggml-transformers/libtransformers.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggml-transformers LIBRARY_PATH=$(shell pwd)/go-ggml-transformers \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/mpt ./cmd/grpc/mpt/
+
+backend-assets/grpc/replit: backend-assets/grpc go-ggml-transformers/libtransformers.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggml-transformers LIBRARY_PATH=$(shell pwd)/go-ggml-transformers \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/replit ./cmd/grpc/replit/
+
+backend-assets/grpc/falcon-ggml: backend-assets/grpc go-ggml-transformers/libtransformers.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggml-transformers LIBRARY_PATH=$(shell pwd)/go-ggml-transformers \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/falcon-ggml ./cmd/grpc/falcon-ggml/
+
+backend-assets/grpc/starcoder: backend-assets/grpc go-ggml-transformers/libtransformers.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggml-transformers LIBRARY_PATH=$(shell pwd)/go-ggml-transformers \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/starcoder ./cmd/grpc/starcoder/
+
+backend-assets/grpc/rwkv: backend-assets/grpc go-rwkv/librwkv.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-rwkv LIBRARY_PATH=$(shell pwd)/go-rwkv \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/rwkv ./cmd/grpc/rwkv/
+
+backend-assets/grpc/bloomz: backend-assets/grpc bloomz/libbloomz.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/bloomz LIBRARY_PATH=$(shell pwd)/bloomz \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/bloomz ./cmd/grpc/bloomz/
+
+backend-assets/grpc/bert-embeddings: backend-assets/grpc go-bert/libgobert.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-bert LIBRARY_PATH=$(shell pwd)/go-bert \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/bert-embeddings ./cmd/grpc/bert-embeddings/
+
+backend-assets/grpc/langchain-huggingface: backend-assets/grpc
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/langchain-huggingface ./cmd/grpc/langchain-huggingface/
+
+backend-assets/grpc/stablediffusion: backend-assets/grpc go-stable-diffusion/libstablediffusion.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-stable-diffusion/ LIBRARY_PATH=$(shell pwd)/go-stable-diffusion/ \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/stablediffusion ./cmd/grpc/stablediffusion/
+
+backend-assets/grpc/piper: backend-assets/grpc backend-assets/espeak-ng-data go-piper/libpiper_binding.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" LIBRARY_PATH=$(shell pwd)/go-piper \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/piper ./cmd/grpc/piper/
+
+backend-assets/grpc/whisper: backend-assets/grpc whisper.cpp/libwhisper.a
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/whisper.cpp LIBRARY_PATH=$(shell pwd)/whisper.cpp \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/whisper ./cmd/grpc/whisper/
+
+grpcs: prepare backend-assets/grpc/langchain-huggingface backend-assets/grpc/llama-grammar backend-assets/grpc/falcon-ggml backend-assets/grpc/bert-embeddings backend-assets/grpc/falcon backend-assets/grpc/bloomz backend-assets/grpc/llama 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)

api/api.go

@@ -3,6 +3,13 @@ package api
 import (
 	"errors"
 
+	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/internal"
+	"github.com/go-skynet/LocalAI/pkg/assets"
+
 	"github.com/gofiber/fiber/v2"
 	"github.com/gofiber/fiber/v2/middleware/cors"
 	"github.com/gofiber/fiber/v2/middleware/logger"
@@ -11,18 +18,18 @@ import (
 	"github.com/rs/zerolog/log"
 )
 
-func App(opts ...AppOption) *fiber.App {
-	options := newOptions(opts...)
+func App(opts ...options.AppOption) (*fiber.App, error) {
+	options := options.NewOptions(opts...)
 
 	zerolog.SetGlobalLevel(zerolog.InfoLevel)
-	if options.debug {
+	if options.Debug {
 		zerolog.SetGlobalLevel(zerolog.DebugLevel)
 	}
 
 	// 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,
+		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
@@ -36,82 +43,122 @@ func App(opts ...AppOption) *fiber.App {
 
 			// Send custom error page
 			return ctx.Status(code).JSON(
-				ErrorResponse{
-					Error: &APIError{Message: err.Error(), Code: code},
+				openai.ErrorResponse{
+					Error: &openai.APIError{Message: err.Error(), Code: code},
 				},
 			)
 		},
 	})
 
-	if options.debug {
+	if options.Debug {
 		app.Use(logger.New(logger.Config{
 			Format: "[${ip}]:${port} ${status} - ${method} ${path}\n",
 		}))
 	}
 
-	cm := NewConfigMerger()
-	if err := cm.LoadConfigs(options.loader.ModelPath); err != nil {
+	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())
+
+	cm := config.NewConfigLoader()
+	if err := cm.LoadConfigs(options.Loader.ModelPath); err != nil {
 		log.Error().Msgf("error loading config files: %s", err.Error())
 	}
 
-	if options.configFile != "" {
-		if err := cm.LoadConfigFile(options.configFile); err != nil {
+	if options.ConfigFile != "" {
+		if err := cm.LoadConfigFile(options.ConfigFile); err != nil {
 			log.Error().Msgf("error loading config file: %s", err.Error())
 		}
 	}
 
-	if options.debug {
+	if options.Debug {
 		for _, v := range cm.ListConfigs() {
 			cfg, _ := cm.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 {
+			log.Warn().Msgf("Failed extracting backend assets files: %s (might be required for some backends to work properly, like gpt4all)", err)
+		}
+	}
+
 	// Default middleware config
 	app.Use(recover.New())
 
-	if options.cors {
-		if options.corsAllowOrigins == "" {
-			app.Use(cors.New())
+	if options.PreloadJSONModels != "" {
+		if err := localai.ApplyGalleryFromString(options.Loader.ModelPath, options.PreloadJSONModels, cm, options.Galleries); err != nil {
+			return nil, err
+		}
+	}
+
+	if options.PreloadModelsFromPath != "" {
+		if err := localai.ApplyGalleryFromFile(options.Loader.ModelPath, options.PreloadModelsFromPath, cm, options.Galleries); err != nil {
+			return nil, err
+		}
+	}
+
+	if options.CORS {
+		var c func(ctx *fiber.Ctx) error
+		if options.CORSAllowOrigins == "" {
+			c = cors.New()
 		} else {
-			app.Use(cors.New(cors.Config{
-				AllowOrigins: options.corsAllowOrigins,
-			}))
+			c = cors.New(cors.Config{AllowOrigins: options.CORSAllowOrigins})
 		}
+
+		app.Use(c)
 	}
 
 	// LocalAI API endpoints
-	applier := newGalleryApplier(options.loader.ModelPath)
-	applier.start(options.context, cm)
-	app.Post("/models/apply", applyModelGallery(options.loader.ModelPath, cm, applier.C))
-	app.Get("/models/jobs/:uuid", getOpStatus(applier))
+	galleryService := localai.NewGalleryService(options.Loader.ModelPath)
+	galleryService.Start(options.Context, cm)
+
+	app.Get("/version", func(c *fiber.Ctx) error {
+		return c.JSON(struct {
+			Version string `json:"version"`
+		}{Version: internal.PrintableVersion()})
+	})
+
+	app.Post("/models/apply", localai.ApplyModelGalleryEndpoint(options.Loader.ModelPath, cm, galleryService.C, options.Galleries))
+	app.Get("/models/available", localai.ListModelFromGalleryEndpoint(options.Galleries, options.Loader.ModelPath))
+	app.Get("/models/jobs/:uuid", localai.GetOpStatusEndpoint(galleryService))
 
 	// openAI compatible API endpoint
 
 	// chat
-	app.Post("/v1/chat/completions", chatEndpoint(cm, options))
-	app.Post("/chat/completions", chatEndpoint(cm, options))
+	app.Post("/v1/chat/completions", openai.ChatEndpoint(cm, options))
+	app.Post("/chat/completions", openai.ChatEndpoint(cm, options))
 
 	// edit
-	app.Post("/v1/edits", editEndpoint(cm, options))
-	app.Post("/edits", editEndpoint(cm, options))
+	app.Post("/v1/edits", openai.EditEndpoint(cm, options))
+	app.Post("/edits", openai.EditEndpoint(cm, options))
 
 	// completion
-	app.Post("/v1/completions", completionEndpoint(cm, options))
-	app.Post("/completions", completionEndpoint(cm, options))
+	app.Post("/v1/completions", openai.CompletionEndpoint(cm, options))
+	app.Post("/completions", openai.CompletionEndpoint(cm, options))
+	app.Post("/v1/engines/:model/completions", openai.CompletionEndpoint(cm, options))
 
 	// embeddings
-	app.Post("/v1/embeddings", embeddingsEndpoint(cm, options))
-	app.Post("/embeddings", embeddingsEndpoint(cm, options))
-	app.Post("/v1/engines/:model/embeddings", embeddingsEndpoint(cm, options))
+	app.Post("/v1/embeddings", openai.EmbeddingsEndpoint(cm, options))
+	app.Post("/embeddings", openai.EmbeddingsEndpoint(cm, options))
+	app.Post("/v1/engines/:model/embeddings", openai.EmbeddingsEndpoint(cm, options))
 
 	// audio
-	app.Post("/v1/audio/transcriptions", transcriptEndpoint(cm, options))
+	app.Post("/v1/audio/transcriptions", openai.TranscriptEndpoint(cm, options))
+	app.Post("/tts", localai.TTSEndpoint(cm, options))
 
 	// images
-	app.Post("/v1/images/generations", imageEndpoint(cm, options))
+	app.Post("/v1/images/generations", openai.ImageEndpoint(cm, options))
 
-	if options.imageDir != "" {
-		app.Static("/generated-images", options.imageDir)
+	if options.ImageDir != "" {
+		app.Static("/generated-images", options.ImageDir)
+	}
+
+	if options.AudioDir != "" {
+		app.Static("/generated-audio", options.AudioDir)
 	}
 
 	ok := func(c *fiber.Ctx) error {
@@ -123,8 +170,15 @@ func App(opts ...AppOption) *fiber.App {
 	app.Get("/readyz", ok)
 
 	// models
-	app.Get("/v1/models", listModels(options.loader, cm))
-	app.Get("/models", listModels(options.loader, cm))
+	app.Get("/v1/models", openai.ListModelsEndpoint(options.Loader, cm))
+	app.Get("/models", openai.ListModelsEndpoint(options.Loader, cm))
+
+	// 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.StopGRPC()
+	}()
 
-	return app
+	return app, nil
 }

api/api_test.go

@@ -3,8 +3,11 @@ package api_test
 import (
 	"bytes"
 	"context"
+	"embed"
 	"encoding/json"
+	"errors"
 	"fmt"
+	"io"
 	"io/ioutil"
 	"net/http"
 	"os"
@@ -12,7 +15,10 @@ import (
 	"runtime"
 
 	. "github.com/go-skynet/LocalAI/api"
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/pkg/gallery"
 	"github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/go-skynet/LocalAI/pkg/utils"
 	"github.com/gofiber/fiber/v2"
 	. "github.com/onsi/ginkgo/v2"
 	. "github.com/onsi/gomega"
@@ -20,9 +26,11 @@ import (
 
 	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]string `json:"overrides"`
@@ -51,6 +59,15 @@ func getModelStatus(url string) (response map[string]interface{}) {
 	}
 	return
 }
+
+func getModels(url string) (response []gallery.GalleryModel) {
+	utils.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"
@@ -95,6 +112,9 @@ func postModelApplyRequest(url string, request modelApplyRequest) (response map[
 	return
 }
 
+//go:embed backend-assets/*
+var backendAssets embed.FS
+
 var _ = Describe("API test", func() {
 
 	var app *fiber.App
@@ -105,6 +125,11 @@ var _ = Describe("API test", func() {
 	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
@@ -114,7 +139,36 @@ var _ = Describe("API test", func() {
 			modelLoader = model.NewModelLoader(tmpdir)
 			c, cancel = context.WithCancel(context.Background())
 
-			app = App(WithContext(c), WithModelLoader(modelLoader))
+			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 = ioutil.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"),
+				},
+			}
+
+			app, err = App(
+				append(commonOpts,
+					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("")
@@ -138,6 +192,53 @@ var _ = Describe("API test", func() {
 		})
 
 		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",
@@ -153,9 +254,8 @@ var _ = Describe("API test", func() {
 
 				Eventually(func() bool {
 					response := getModelStatus("http://127.0.0.1:9090/models/jobs/" + uuid)
-					fmt.Println(response)
 					return response["processed"].(bool)
-				}, "360s").Should(Equal(true))
+				}, "360s", "10s").Should(Equal(true))
 
 				dat, err := os.ReadFile(filepath.Join(tmpdir, "bert.yaml"))
 				Expect(err).ToNot(HaveOccurred())
@@ -178,9 +278,8 @@ var _ = Describe("API test", func() {
 
 				Eventually(func() bool {
 					response := getModelStatus("http://127.0.0.1:9090/models/jobs/" + uuid)
-					fmt.Println(response)
 					return response["processed"].(bool)
-				}, "360s").Should(Equal(true))
+				}, "360s", "10s").Should(Equal(true))
 
 				dat, err := os.ReadFile(filepath.Join(tmpdir, "bert.yaml"))
 				Expect(err).ToNot(HaveOccurred())
@@ -190,6 +289,215 @@ var _ = Describe("API test", func() {
 				Expect(err).ToNot(HaveOccurred())
 				Expect(content["backend"]).To(Equal("bert-embeddings"))
 			})
+
+			It("runs openllama", 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]string{"backend": "llama-grammar"},
+				})
+
+				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"), 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",
+					Overrides: map[string]string{},
+				})
+
+				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))
+
+				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",
+				},
+			}
+
+			app, err = App(
+				append(commonOpts,
+					options.WithContext(c),
+					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",
+			})
+
+			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))
+
 		})
 	})
 
@@ -198,7 +506,14 @@ var _ = Describe("API test", func() {
 			modelLoader = model.NewModelLoader(os.Getenv("MODELS_PATH"))
 			c, cancel = context.WithCancel(context.Background())
 
-			app = App(WithContext(c), WithModelLoader(modelLoader))
+			var err error
+			app, err = App(
+				append(commonOpts,
+					options.WithExternalBackend("huggingface", os.Getenv("HUGGINGFACE_GRPC")),
+					options.WithContext(c),
+					options.WithModelLoader(modelLoader),
+				)...)
+			Expect(err).ToNot(HaveOccurred())
 			go app.Listen("127.0.0.1:9090")
 
 			defaultConfig := openai.DefaultConfig("")
@@ -221,7 +536,7 @@ var _ = Describe("API test", func() {
 		It("returns the models list", func() {
 			models, err := client.ListModels(context.TODO())
 			Expect(err).ToNot(HaveOccurred())
-			Expect(len(models.Models)).To(Equal(10))
+			Expect(len(models.Models)).To(Equal(11))
 		})
 		It("can generate completions", func() {
 			resp, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "testmodel", Prompt: "abcdedfghikl"})
@@ -252,9 +567,10 @@ var _ = Describe("API test", func() {
 		})
 
 		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("error, status code: 500, message: could not load model - all backends returned error: 12 errors occurred:"))
+			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" {
@@ -298,15 +614,98 @@ var _ = Describe("API test", func() {
 			Expect(resp2.Data[0].Embedding).To(Equal(sunEmbedding))
 		})
 
+		Context("External gRPC calls", func() {
+			It("calculate embeddings with huggingface", 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", 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(Equal(" five."))
+				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)))
 			})
 		})
 	})
@@ -316,7 +715,14 @@ var _ = Describe("API test", func() {
 			modelLoader = model.NewModelLoader(os.Getenv("MODELS_PATH"))
 			c, cancel = context.WithCancel(context.Background())
 
-			app = App(WithContext(c), WithModelLoader(modelLoader), WithConfigFile(os.Getenv("CONFIG_FILE")))
+			var err error
+			app, err = App(
+				append(commonOpts,
+					options.WithContext(c),
+					options.WithModelLoader(modelLoader),
+					options.WithConfigFile(os.Getenv("CONFIG_FILE")))...,
+			)
+			Expect(err).ToNot(HaveOccurred())
 			go app.Listen("127.0.0.1:9090")
 
 			defaultConfig := openai.DefaultConfig("")
@@ -337,7 +743,7 @@ var _ = Describe("API test", func() {
 		It("can generate chat completions from config file", func() {
 			models, err := client.ListModels(context.TODO())
 			Expect(err).ToNot(HaveOccurred())
-			Expect(len(models.Models)).To(Equal(12))
+			Expect(len(models.Models)).To(Equal(13))
 		})
 		It("can generate chat completions from config file", func() {
 			resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: "list1", Messages: []openai.ChatCompletionMessage{openai.ChatCompletionMessage{Role: "user", Content: "abcdedfghikl"}}})

api/backend/embeddings.go

@@ -0,0 +1,109 @@
+package backend
+
+import (
+	"fmt"
+	"sync"
+
+	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 := []model.Option{
+		model.WithLoadGRPCLLMModelOpts(grpcOpts),
+		model.WithThreads(uint32(c.Threads)),
+		model.WithAssetDir(o.AssetsDestination),
+		model.WithModelFile(modelFile),
+		model.WithContext(o.Context),
+	}
+
+	for k, v := range o.ExternalGRPCBackends {
+		opts = append(opts, model.WithExternalBackend(k, v))
+	}
+
+	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.Client:
+		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) {
+		// 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()
+
+		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 +1,68 @@
+package backend
+
+import (
+	"fmt"
+	"sync"
+
+	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, dst string, loader *model.ModelLoader, c config.Config, o *options.Option) (func() error, error) {
+	if c.Backend != model.StableDiffusionBackend {
+		return nil, fmt.Errorf("endpoint only working with stablediffusion models")
+	}
+
+	opts := []model.Option{
+		model.WithBackendString(c.Backend),
+		model.WithAssetDir(o.AssetsDestination),
+		model.WithThreads(uint32(c.Threads)),
+		model.WithContext(o.Context),
+		model.WithModelFile(c.ImageGenerationAssets),
+	}
+
+	for k, v := range o.ExternalGRPCBackends {
+		opts = append(opts, model.WithExternalBackend(k, v))
+	}
+
+	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),
+				PositivePrompt: positive_prompt,
+				NegativePrompt: negative_prompt,
+				Dst:            dst,
+			})
+		return err
+	}
+
+	return func() error {
+		// This is still needed, see: https://github.com/ggerganov/llama.cpp/discussions/784
+		mutexMap.Lock()
+		l, ok := mutexes[c.Backend]
+		if !ok {
+			m := &sync.Mutex{}
+			mutexes[c.Backend] = m
+			l = m
+		}
+		mutexMap.Unlock()
+		l.Lock()
+		defer l.Unlock()
+
+		return fn()
+	}, nil
+}

api/backend/llm.go

@@ -0,0 +1,124 @@
+package backend
+
+import (
+	"os"
+	"regexp"
+	"strings"
+	"sync"
+
+	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"
+)
+
+func ModelInference(s string, loader *model.ModelLoader, c config.Config, o *options.Option, tokenCallback func(string) bool) (func() (string, error), error) {
+	modelFile := c.Model
+
+	grpcOpts := gRPCModelOpts(c)
+
+	var inferenceModel *grpc.Client
+	var err error
+
+	opts := []model.Option{
+		model.WithLoadGRPCLLMModelOpts(grpcOpts),
+		model.WithThreads(uint32(c.Threads)), // some models uses this to allocate threads during startup
+		model.WithAssetDir(o.AssetsDestination),
+		model.WithModelFile(modelFile),
+		model.WithContext(o.Context),
+	}
+
+	for k, v := range o.ExternalGRPCBackends {
+		opts = append(opts, model.WithExternalBackend(k, v))
+	}
+
+	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() (string, error) {
+		opts := gRPCPredictOpts(c, loader.ModelPath)
+		opts.Prompt = s
+		if tokenCallback != nil {
+			ss := ""
+			err := inferenceModel.PredictStream(o.Context, opts, func(s string) {
+				tokenCallback(s)
+				ss += s
+			})
+			return ss, err
+		} else {
+			reply, err := inferenceModel.Predict(o.Context, opts)
+			if err != nil {
+				return "", err
+			}
+			return reply.Message, err
+		}
+	}
+
+	return func() (string, error) {
+		// 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()
+
+		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))
+	}
+	return prediction
+
+}

api/backend/lock.go

@@ -0,0 +1,22 @@
+package backend
+
+import "sync"
+
+// mutex still needed, see: https://github.com/ggerganov/llama.cpp/discussions/784
+var mutexMap sync.Mutex
+var mutexes map[string]*sync.Mutex = make(map[string]*sync.Mutex)
+
+func Lock(s string) *sync.Mutex {
+	// This is still needed, see: https://github.com/ggerganov/llama.cpp/discussions/784
+	mutexMap.Lock()
+	l, ok := mutexes[s]
+	if !ok {
+		m := &sync.Mutex{}
+		mutexes[s] = m
+		l = m
+	}
+	mutexMap.Unlock()
+	l.Lock()
+
+	return l
+}

api/backend/options.go

@@ -0,0 +1,72 @@
+package backend
+
+import (
+	"os"
+	"path/filepath"
+
+	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
+
+	config "github.com/go-skynet/LocalAI/api/config"
+)
+
+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),
+		F16Memory:   c.F16,
+		MLock:       c.MMlock,
+		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,
+	}
+}
+
+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),
+		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,
+
+		Mirostat:          int32(c.Mirostat),
+		MirostatETA:       float32(c.MirostatETA),
+		MirostatTAU:       float32(c.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 +1,42 @@
+package backend
+
+import (
+	"context"
+	"fmt"
+
+	config "github.com/go-skynet/LocalAI/api/config"
+
+	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/pkg/grpc/proto"
+	"github.com/go-skynet/LocalAI/pkg/grpc/whisper/api"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+)
+
+func ModelTranscription(audio, language string, loader *model.ModelLoader, c config.Config, o *options.Option) (*api.Result, error) {
+	opts := []model.Option{
+		model.WithBackendString(model.WhisperBackend),
+		model.WithModelFile(c.Model),
+		model.WithContext(o.Context),
+		model.WithThreads(uint32(c.Threads)),
+		model.WithAssetDir(o.AssetsDestination),
+	}
+
+	for k, v := range o.ExternalGRPCBackends {
+		opts = append(opts, model.WithExternalBackend(k, v))
+	}
+
+	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 +1,72 @@
+package backend
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"path/filepath"
+
+	"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(text, modelFile string, loader *model.ModelLoader, o *options.Option) (string, *proto.Result, error) {
+	opts := []model.Option{
+		model.WithBackendString(model.PiperBackend),
+		model.WithModelFile(modelFile),
+		model.WithContext(o.Context),
+		model.WithAssetDir(o.AssetsDestination),
+	}
+
+	for k, v := range o.ExternalGRPCBackends {
+		opts = append(opts, model.WithExternalBackend(k, v))
+	}
+
+	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)
+
+	modelPath := filepath.Join(o.Loader.ModelPath, modelFile)
+
+	if err := utils.VerifyPath(modelPath, o.Loader.ModelPath); err != nil {
+		return "", nil, err
+	}
+
+	res, err := piperModel.TTS(context.Background(), &proto.TTSRequest{
+		Text:  text,
+		Model: modelPath,
+		Dst:   filePath,
+	})
+
+	return filePath, res, err
+}

api/config.go

@@ -1,329 +0,0 @@
-package api
-
-import (
-	"encoding/json"
-	"fmt"
-	"io/ioutil"
-	"os"
-	"path/filepath"
-	"strings"
-	"sync"
-
-	model "github.com/go-skynet/LocalAI/pkg/model"
-	"github.com/gofiber/fiber/v2"
-	"github.com/rs/zerolog/log"
-	"gopkg.in/yaml.v3"
-)
-
-type Config struct {
-	OpenAIRequest               `yaml:"parameters"`
-	Name                        string            `yaml:"name"`
-	StopWords                   []string          `yaml:"stopwords"`
-	Cutstrings                  []string          `yaml:"cutstrings"`
-	TrimSpace                   []string          `yaml:"trimspace"`
-	ContextSize                 int               `yaml:"context_size"`
-	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"`
-	MirostatETA                 float64           `yaml:"mirostat_eta"`
-	MirostatTAU                 float64           `yaml:"mirostat_tau"`
-	Mirostat                    int               `yaml:"mirostat"`
-	NGPULayers                  int               `yaml:"gpu_layers"`
-	ImageGenerationAssets       string            `yaml:"asset_dir"`
-	PromptStrings, InputStrings []string
-	InputToken                  [][]int
-}
-
-type TemplateConfig struct {
-	Completion string `yaml:"completion"`
-	Chat       string `yaml:"chat"`
-	Edit       string `yaml:"edit"`
-}
-
-type ConfigMerger struct {
-	configs map[string]Config
-	sync.Mutex
-}
-
-func NewConfigMerger() *ConfigMerger {
-	return &ConfigMerger{
-		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 ConfigMerger) 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 ConfigMerger) 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 ConfigMerger) GetConfig(m string) (Config, bool) {
-	cm.Lock()
-	defer cm.Unlock()
-	v, exists := cm.configs[m]
-	return v, exists
-}
-
-func (cm ConfigMerger) ListConfigs() []string {
-	cm.Lock()
-	defer cm.Unlock()
-	var res []string
-	for k := range cm.configs {
-		res = append(res, k)
-	}
-	return res
-}
-
-func (cm ConfigMerger) LoadConfigs(path string) error {
-	cm.Lock()
-	defer cm.Unlock()
-	files, err := ioutil.ReadDir(path)
-	if err != nil {
-		return err
-	}
-
-	for _, file := range files {
-		// Skip templates, YAML and .keep files
-		if !strings.Contains(file.Name(), ".yaml") {
-			continue
-		}
-		c, err := ReadConfig(filepath.Join(path, file.Name()))
-		if err == nil {
-			cm.configs[c.Name] = *c
-		}
-	}
-
-	return nil
-}
-
-func updateConfig(config *Config, input *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.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)
-			}
-		}
-	}
-
-	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.Mirostat = input.Mirostat
-	}
-
-	if input.MirostatETA != 0 {
-		config.MirostatETA = input.MirostatETA
-	}
-
-	if input.MirostatTAU != 0 {
-		config.MirostatTAU = input.MirostatTAU
-	}
-
-	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)
-			}
-		}
-	}
-
-	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 readInput(c *fiber.Ctx, loader *model.ModelLoader, randomModel bool) (string, *OpenAIRequest, error) {
-	input := new(OpenAIRequest)
-	// Get input data from the request body
-	if err := c.BodyParser(input); err != nil {
-		return "", nil, 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
-}
-
-func readConfig(modelFile string, input *OpenAIRequest, cm *ConfigMerger, loader *model.ModelLoader, debug bool, threads, ctx int, f16 bool) (*Config, *OpenAIRequest, error) {
-	// Load a config file if present after the model name
-	modelConfig := filepath.Join(loader.ModelPath, modelFile+".yaml")
-	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())
-		}
-	}
-
-	var config *Config
-	cfg, exists := cm.GetConfig(modelFile)
-	if !exists {
-		config = &Config{
-			OpenAIRequest: defaultRequest(modelFile),
-			ContextSize:   ctx,
-			Threads:       threads,
-			F16:           f16,
-			Debug:         debug,
-		}
-	} else {
-		config = &cfg
-	}
-
-	// Set the parameters for the language model prediction
-	updateConfig(config, input)
-
-	// Don't allow 0 as setting
-	if config.Threads == 0 {
-		if threads != 0 {
-			config.Threads = threads
-		} else {
-			config.Threads = 4
-		}
-	}
-
-	// Enforce debug flag if passed from CLI
-	if debug {
-		config.Debug = true
-	}
-
-	return config, input, nil
-}

api/config/config.go

@@ -0,0 +1,209 @@
+package api_config
+
+import (
+	"fmt"
+	"io/fs"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+
+	"gopkg.in/yaml.v3"
+)
+
+type Config struct {
+	PredictionOptions `yaml:"parameters"`
+	Name              string            `yaml:"name"`
+	StopWords         []string          `yaml:"stopwords"`
+	Cutstrings        []string          `yaml:"cutstrings"`
+	TrimSpace         []string          `yaml:"trimspace"`
+	ContextSize       int               `yaml:"context_size"`
+	F16               bool              `yaml:"f16"`
+	NUMA              bool              `yaml:"numa"`
+	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"`
+	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"`
+
+	TensorSplit           string `yaml:"tensor_split"`
+	MainGPU               string `yaml:"main_gpu"`
+	ImageGenerationAssets string `yaml:"asset_dir"`
+
+	PromptCachePath string `yaml:"prompt_cache_path"`
+	PromptCacheAll  bool   `yaml:"prompt_cache_all"`
+	PromptCacheRO   bool   `yaml:"prompt_cache_ro"`
+
+	Grammar string `yaml:"grammar"`
+
+	PromptStrings, InputStrings                []string
+	InputToken                                 [][]int
+	functionCallString, functionCallNameString string
+
+	FunctionsConfig Functions `yaml:"function"`
+}
+
+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 {
+	Completion string `yaml:"completion"`
+	Functions  string `yaml:"function"`
+	Chat       string `yaml:"chat"`
+	Edit       string `yaml:"edit"`
+}
+
+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) ListConfigs() []string {
+	cm.Lock()
+	defer cm.Unlock()
+	var res []string
+	for k := range cm.configs {
+		res = append(res, k)
+	}
+	return res
+}
+
+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") {
+			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 +1,37 @@
+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"`
+}

api/gallery.go

@@ -1,196 +0,0 @@
-package api
-
-import (
-	"context"
-	"fmt"
-	"io/ioutil"
-	"net/http"
-	"net/url"
-	"strings"
-	"sync"
-
-	"github.com/go-skynet/LocalAI/pkg/gallery"
-	"github.com/gofiber/fiber/v2"
-	"github.com/google/uuid"
-	"gopkg.in/yaml.v3"
-)
-
-type galleryOp struct {
-	req ApplyGalleryModelRequest
-	id  string
-}
-
-type galleryOpStatus struct {
-	Error     error  `json:"error"`
-	Processed bool   `json:"processed"`
-	Message   string `json:"message"`
-}
-
-type galleryApplier struct {
-	modelPath string
-	sync.Mutex
-	C        chan galleryOp
-	statuses map[string]*galleryOpStatus
-}
-
-func newGalleryApplier(modelPath string) *galleryApplier {
-	return &galleryApplier{
-		modelPath: modelPath,
-		C:         make(chan galleryOp),
-		statuses:  make(map[string]*galleryOpStatus),
-	}
-}
-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) start(c context.Context, cm *ConfigMerger) {
-	go func() {
-		for {
-			select {
-			case <-c.Done():
-				return
-			case op := <-g.C:
-				g.updatestatus(op.id, &galleryOpStatus{Message: "processing"})
-
-				updateError := func(e error) {
-					g.updatestatus(op.id, &galleryOpStatus{Error: e, Processed: true})
-				}
-
-				url, err := op.req.DecodeURL()
-				if err != nil {
-					updateError(err)
-					continue
-				}
-
-				// Send a GET request to the URL
-				response, err := http.Get(url)
-				if err != nil {
-					updateError(err)
-					continue
-				}
-				defer response.Body.Close()
-
-				// Read the response body
-				body, err := ioutil.ReadAll(response.Body)
-				if err != nil {
-					updateError(err)
-					continue
-				}
-
-				// Unmarshal YAML data into a Config struct
-				var config gallery.Config
-				err = yaml.Unmarshal(body, &config)
-				if err != nil {
-					updateError(fmt.Errorf("failed to unmarshal YAML: %v", err))
-					continue
-				}
-
-				config.Files = append(config.Files, op.req.AdditionalFiles...)
-
-				if err := gallery.Apply(g.modelPath, op.req.Name, &config, op.req.Overrides); err != nil {
-					updateError(err)
-					continue
-				}
-
-				// Reload models
-				if err := cm.LoadConfigs(g.modelPath); err != nil {
-					updateError(err)
-					continue
-				}
-
-				g.updatestatus(op.id, &galleryOpStatus{Processed: true, Message: "completed"})
-			}
-		}
-	}()
-}
-
-// endpoints
-
-type ApplyGalleryModelRequest struct {
-	URL             string                 `json:"url"`
-	Name            string                 `json:"name"`
-	Overrides       map[string]interface{} `json:"overrides"`
-	AdditionalFiles []gallery.File         `json:"files"`
-}
-
-const (
-	githubURI = "github:"
-)
-
-func (request ApplyGalleryModelRequest) DecodeURL() (string, error) {
-	input := request.URL
-	var rawURL string
-
-	if strings.HasPrefix(input, githubURI) {
-		parts := strings.Split(input, ":")
-		repoParts := strings.Split(parts[1], "@")
-		branch := "main"
-
-		if len(repoParts) > 1 {
-			branch = repoParts[1]
-		}
-
-		repoPath := strings.Split(repoParts[0], "/")
-		org := repoPath[0]
-		project := repoPath[1]
-		projectPath := strings.Join(repoPath[2:], "/")
-
-		rawURL = fmt.Sprintf("https://raw.githubusercontent.com/%s/%s/%s/%s", org, project, branch, projectPath)
-	} else if strings.HasPrefix(input, "http://") || strings.HasPrefix(input, "https://") {
-		// Handle regular URLs
-		u, err := url.Parse(input)
-		if err != nil {
-			return "", fmt.Errorf("invalid URL: %w", err)
-		}
-		rawURL = u.String()
-	} else {
-		return "", fmt.Errorf("invalid URL format")
-	}
-
-	return rawURL, nil
-}
-
-func getOpStatus(g *galleryApplier) func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-
-		status := g.getstatus(c.Params("uuid"))
-		if status == nil {
-			return fmt.Errorf("could not find any status for ID")
-		}
-
-		return c.JSON(status)
-	}
-}
-
-func applyModelGallery(modelPath string, cm *ConfigMerger, g chan galleryOp) func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-		input := new(ApplyGalleryModelRequest)
-		// 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
-		}
-		g <- galleryOp{
-			req: *input,
-			id:  uuid.String(),
-		}
-		return c.JSON(struct {
-			ID        string `json:"uuid"`
-			StatusURL string `json:"status"`
-		}{ID: uuid.String(), StatusURL: c.BaseURL() + "/models/jobs/" + uuid.String()})
-	}
-}

api/gallery_test.go

@@ -1,30 +0,0 @@
-package api_test
-
-import (
-	. "github.com/go-skynet/LocalAI/api"
-	. "github.com/onsi/ginkgo/v2"
-	. "github.com/onsi/gomega"
-)
-
-var _ = Describe("Gallery API tests", func() {
-	Context("requests", func() {
-		It("parses github with a branch", func() {
-			req := ApplyGalleryModelRequest{URL: "github:go-skynet/model-gallery/gpt4all-j.yaml@main"}
-			str, err := req.DecodeURL()
-			Expect(err).ToNot(HaveOccurred())
-			Expect(str).To(Equal("https://raw.githubusercontent.com/go-skynet/model-gallery/main/gpt4all-j.yaml"))
-		})
-		It("parses github without a branch", func() {
-			req := ApplyGalleryModelRequest{URL: "github:go-skynet/model-gallery/gpt4all-j.yaml"}
-			str, err := req.DecodeURL()
-			Expect(err).ToNot(HaveOccurred())
-			Expect(str).To(Equal("https://raw.githubusercontent.com/go-skynet/model-gallery/main/gpt4all-j.yaml"))
-		})
-		It("parses URLS", func() {
-			req := ApplyGalleryModelRequest{URL: "https://raw.githubusercontent.com/go-skynet/model-gallery/main/gpt4all-j.yaml"}
-			str, err := req.DecodeURL()
-			Expect(err).ToNot(HaveOccurred())
-			Expect(str).To(Equal("https://raw.githubusercontent.com/go-skynet/model-gallery/main/gpt4all-j.yaml"))
-		})
-	})
-})

api/localai/gallery.go

@@ -0,0 +1,224 @@
+package localai
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"strings"
+	"sync"
+
+	json "github.com/json-iterator/go"
+
+	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 {
+	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),
+	}
+}
+
+// prepareModel applies a
+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) 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", 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
+				}
+
+				g.updateStatus(op.id, &galleryOpStatus{Processed: true, Message: "completed", Progress: 100})
+			}
+		}
+	}()
+}
+
+type galleryModel struct {
+	gallery.GalleryModel
+	ID string `json:"id"`
+}
+
+func ApplyGalleryFromFile(modelPath, s string, cm *config.ConfigLoader, galleries []gallery.Gallery) error {
+	dat, err := os.ReadFile(s)
+	if err != nil {
+		return err
+	}
+	return ApplyGalleryFromString(modelPath, string(dat), cm, galleries)
+}
+
+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
+	}
+
+	for _, r := range requests {
+		utils.ResetDownloadTimers()
+		if r.ID == "" {
+			err = prepareModel(modelPath, r.GalleryModel, cm, utils.DisplayDownloadFunction)
+		} else {
+			err = gallery.InstallModelFromGallery(galleries, r.ID, modelPath, r.GalleryModel, utils.DisplayDownloadFunction)
+		}
+	}
+
+	return err
+}
+
+/// Endpoints
+
+func GetOpStatusEndpoint(g *galleryApplier) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+
+		status := g.getStatus(c.Params("uuid"))
+		if status == nil {
+			return fmt.Errorf("could not find any status for ID")
+		}
+
+		return c.JSON(status)
+	}
+}
+
+type GalleryModel struct {
+	ID string `json:"id"`
+	gallery.GalleryModel
+}
+
+func ApplyModelGalleryEndpoint(modelPath string, cm *config.ConfigLoader, g chan galleryOp, galleries []gallery.Gallery) 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
+		}
+		g <- galleryOp{
+			req:         input.GalleryModel,
+			id:          uuid.String(),
+			galleryName: input.ID,
+			galleries:   galleries,
+		}
+		return c.JSON(struct {
+			ID        string `json:"uuid"`
+			StatusURL string `json:"status"`
+		}{ID: uuid.String(), StatusURL: c.BaseURL() + "/models/jobs/" + uuid.String()})
+	}
+}
+
+func ListModelFromGalleryEndpoint(galleries []gallery.Gallery, basePath string) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		log.Debug().Msgf("Listing models from galleries: %+v", galleries)
+
+		models, err := gallery.AvailableGalleryModels(galleries, basePath)
+		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)
+	}
+}

api/localai/localai.go

@@ -0,0 +1,31 @@
+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"`
+}
+
+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.Input, input.Model, o.Loader, o)
+		if err != nil {
+			return err
+		}
+		return c.Download(filePath)
+	}
+}

api/openai.go

@@ -1,678 +0,0 @@
-package api
-
-import (
-	"bufio"
-	"bytes"
-	"encoding/base64"
-	"encoding/json"
-	"fmt"
-	"io"
-	"io/ioutil"
-	"net/http"
-	"os"
-	"path"
-	"path/filepath"
-	"strconv"
-	"strings"
-
-	"github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
-	model "github.com/go-skynet/LocalAI/pkg/model"
-	whisperutil "github.com/go-skynet/LocalAI/pkg/whisper"
-	llama "github.com/go-skynet/go-llama.cpp"
-	"github.com/gofiber/fiber/v2"
-	"github.com/rs/zerolog/log"
-	"github.com/valyala/fasthttp"
-)
-
-// 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,omitempty"`
-	FinishReason string   `json:"finish_reason,omitempty"`
-	Message      *Message `json:"message,omitempty"`
-	Delta        *Message `json:"delta,omitempty"`
-	Text         string   `json:"text,omitempty"`
-}
-
-type Message struct {
-	Role    string `json:"role,omitempty" yaml:"role"`
-	Content string `json:"content,omitempty" yaml:"content"`
-}
-
-type OpenAIModel struct {
-	ID     string `json:"id"`
-	Object string `json:"object"`
-}
-
-type OpenAIRequest struct {
-	Model string `json:"model" yaml:"model"`
-
-	// whisper
-	File     string `json:"file" validate:"required"`
-	Language string `json:"language"`
-	//whisper/image
-	ResponseFormat string `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"`
-
-	Stream bool `json:"stream"`
-	Echo   bool `json:"echo"`
-	// Common options between all the API calls
-	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"`
-
-	N int `json:"n"`
-
-	// 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"`
-
-	Seed int `json:"seed" yaml:"seed"`
-
-	// Image (not supported by OpenAI)
-	Mode int `json:"mode"`
-	Step int `json:"step"`
-}
-
-func defaultRequest(modelFile string) OpenAIRequest {
-	return OpenAIRequest{
-		TopP:        0.7,
-		TopK:        80,
-		Maxtokens:   512,
-		Temperature: 0.9,
-		Model:       modelFile,
-	}
-}
-
-// https://platform.openai.com/docs/api-reference/completions
-func completionEndpoint(cm *ConfigMerger, o *Option) func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-
-		model, input, err := readInput(c, o.loader, 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.ctxSize, o.f16)
-		if err != nil {
-			return fmt.Errorf("failed reading parameters from request:%w", err)
-		}
-
-		log.Debug().Msgf("Parameter Config: %+v", config)
-
-		templateFile := config.Model
-
-		if config.TemplateConfig.Completion != "" {
-			templateFile = config.TemplateConfig.Completion
-		}
-
-		var result []Choice
-		for _, i := range config.PromptStrings {
-			// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
-			templatedInput, err := o.loader.TemplatePrefix(templateFile, struct {
-				Input string
-			}{Input: i})
-			if err == nil {
-				i = templatedInput
-				log.Debug().Msgf("Template found, input modified to: %s", i)
-			}
-
-			r, err := ComputeChoices(i, input, config, o.loader, func(s string, c *[]Choice) {
-				*c = append(*c, Choice{Text: s})
-			}, nil)
-			if err != nil {
-				return err
-			}
-
-			result = append(result, r...)
-		}
-
-		resp := &OpenAIResponse{
-			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
-			Choices: result,
-			Object:  "text_completion",
-		}
-
-		jsonResult, _ := json.Marshal(resp)
-		log.Debug().Msgf("Response: %s", jsonResult)
-
-		// Return the prediction in the response body
-		return c.JSON(resp)
-	}
-}
-
-// https://platform.openai.com/docs/api-reference/embeddings
-func embeddingsEndpoint(cm *ConfigMerger, o *Option) func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-		model, input, err := readInput(c, o.loader, 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.ctxSize, o.f16)
-		if err != nil {
-			return fmt.Errorf("failed reading parameters from request:%w", err)
-		}
-
-		log.Debug().Msgf("Parameter Config: %+v", config)
-		items := []Item{}
-
-		for i, s := range config.InputToken {
-			// get the model function to call for the result
-			embedFn, err := ModelEmbedding("", s, o.loader, *config)
-			if err != nil {
-				return err
-			}
-
-			embeddings, err := embedFn()
-			if err != nil {
-				return err
-			}
-			items = append(items, Item{Embedding: embeddings, Index: i, Object: "embedding"})
-		}
-
-		for i, s := range config.InputStrings {
-			// get the model function to call for the result
-			embedFn, err := ModelEmbedding(s, []int{}, o.loader, *config)
-			if err != nil {
-				return err
-			}
-
-			embeddings, err := embedFn()
-			if err != nil {
-				return err
-			}
-			items = append(items, Item{Embedding: embeddings, Index: i, Object: "embedding"})
-		}
-
-		resp := &OpenAIResponse{
-			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)
-	}
-}
-
-func chatEndpoint(cm *ConfigMerger, o *Option) func(c *fiber.Ctx) error {
-
-	process := func(s string, req *OpenAIRequest, config *Config, loader *model.ModelLoader, responses chan OpenAIResponse) {
-		initialMessage := OpenAIResponse{
-			Model:   req.Model, // we have to return what the user sent here, due to OpenAI spec.
-			Choices: []Choice{{Delta: &Message{Role: "assistant"}}},
-			Object:  "chat.completion.chunk",
-		}
-		responses <- initialMessage
-
-		ComputeChoices(s, req, config, loader, func(s string, c *[]Choice) {}, func(s string) bool {
-			resp := OpenAIResponse{
-				Model:   req.Model, // we have to return what the user sent here, due to OpenAI spec.
-				Choices: []Choice{{Delta: &Message{Content: s}}},
-				Object:  "chat.completion.chunk",
-			}
-			log.Debug().Msgf("Sending goroutine: %s", s)
-
-			responses <- resp
-			return true
-		})
-		close(responses)
-	}
-	return func(c *fiber.Ctx) error {
-		model, input, err := readInput(c, o.loader, 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.ctxSize, o.f16)
-		if err != nil {
-			return fmt.Errorf("failed reading parameters from request:%w", err)
-		}
-
-		log.Debug().Msgf("Parameter Config: %+v", config)
-
-		var predInput string
-
-		mess := []string{}
-		for _, i := range input.Messages {
-			var content string
-			r := config.Roles[i.Role]
-			if r != "" {
-				content = fmt.Sprint(r, " ", i.Content)
-			} else {
-				content = i.Content
-			}
-
-			mess = append(mess, content)
-		}
-
-		predInput = strings.Join(mess, "\n")
-
-		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 := config.Model
-
-		if config.TemplateConfig.Chat != "" {
-			templateFile = config.TemplateConfig.Chat
-		}
-
-		// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
-		templatedInput, err := o.loader.TemplatePrefix(templateFile, struct {
-			Input string
-		}{Input: predInput})
-		if err == nil {
-			predInput = templatedInput
-			log.Debug().Msgf("Template found, input modified to: %s", predInput)
-		}
-
-		if input.Stream {
-			responses := make(chan 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 := &OpenAIResponse{
-					Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
-					Choices: []Choice{{FinishReason: "stop"}},
-				}
-				respData, _ := json.Marshal(resp)
-
-				w.WriteString(fmt.Sprintf("data: %s\n\n", respData))
-				w.WriteString("data: [DONE]\n\n")
-				w.Flush()
-			}))
-			return nil
-		}
-
-		result, err := ComputeChoices(predInput, input, config, o.loader, func(s string, c *[]Choice) {
-			*c = append(*c, Choice{Message: &Message{Role: "assistant", Content: s}})
-		}, nil)
-		if err != nil {
-			return err
-		}
-
-		resp := &OpenAIResponse{
-			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
-			Choices: result,
-			Object:  "chat.completion",
-		}
-		respData, _ := json.Marshal(resp)
-		log.Debug().Msgf("Response: %s", respData)
-
-		// Return the prediction in the response body
-		return c.JSON(resp)
-	}
-}
-
-func editEndpoint(cm *ConfigMerger, o *Option) func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-		model, input, err := readInput(c, o.loader, 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.ctxSize, o.f16)
-		if err != nil {
-			return fmt.Errorf("failed reading parameters from request:%w", err)
-		}
-
-		log.Debug().Msgf("Parameter Config: %+v", config)
-
-		templateFile := config.Model
-
-		if config.TemplateConfig.Edit != "" {
-			templateFile = config.TemplateConfig.Edit
-		}
-
-		var result []Choice
-		for _, i := range config.InputStrings {
-			// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
-			templatedInput, err := o.loader.TemplatePrefix(templateFile, struct {
-				Input       string
-				Instruction string
-			}{Input: i})
-			if err == nil {
-				i = templatedInput
-				log.Debug().Msgf("Template found, input modified to: %s", i)
-			}
-
-			r, err := ComputeChoices(i, input, config, o.loader, func(s string, c *[]Choice) {
-				*c = append(*c, Choice{Text: s})
-			}, nil)
-			if err != nil {
-				return err
-			}
-
-			result = append(result, r...)
-		}
-
-		resp := &OpenAIResponse{
-			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
-			Choices: result,
-			Object:  "edit",
-		}
-
-		jsonResult, _ := json.Marshal(resp)
-		log.Debug().Msgf("Response: %s", jsonResult)
-
-		// Return the prediction in the response body
-		return c.JSON(resp)
-	}
-}
-
-// 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 *ConfigMerger, o *Option) func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-		m, input, err := readInput(c, o.loader, 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)
-		}
-
-		log.Debug().Msgf("Parameter Config: %+v", config)
-
-		// XXX: Only stablediffusion is supported for now
-		if config.Backend == "" {
-			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 == "b64_json" {
-			b64JSON = true
-		}
-
-		var result []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 := 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 := ioutil.TempFile(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 := ImageGeneration(height, width, mode, step, input.Seed, positive_prompt, negative_prompt, output, o.loader, *config)
-				if err != nil {
-					return err
-				}
-				if err := fn(); err != nil {
-					return err
-				}
-
-				item := &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)
-			}
-		}
-
-		resp := &OpenAIResponse{
-			Data: result,
-		}
-
-		jsonResult, _ := json.Marshal(resp)
-		log.Debug().Msgf("Response: %s", jsonResult)
-
-		// Return the prediction in the response body
-		return c.JSON(resp)
-	}
-}
-
-// https://platform.openai.com/docs/api-reference/audio/create
-func transcriptEndpoint(cm *ConfigMerger, o *Option) func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-		m, input, err := readInput(c, o.loader, 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.ctxSize, 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)
-
-		whisperModel, err := o.loader.BackendLoader(model.WhisperBackend, config.Model, []llama.ModelOption{}, uint32(config.Threads))
-		if err != nil {
-			return err
-		}
-
-		if whisperModel == nil {
-			return fmt.Errorf("could not load whisper model")
-		}
-
-		w, ok := whisperModel.(whisper.Model)
-		if !ok {
-			return fmt.Errorf("loader returned non-whisper object")
-		}
-
-		tr, err := whisperutil.Transcript(w, dst, input.Language, uint(config.Threads))
-		if err != nil {
-			return err
-		}
-
-		log.Debug().Msgf("Trascribed: %+v", tr)
-		// TODO: handle different outputs here
-		return c.Status(http.StatusOK).JSON(fiber.Map{"text": tr})
-	}
-}
-
-func listModels(loader *model.ModelLoader, cm *ConfigMerger) 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 := []OpenAIModel{}
-		for _, m := range models {
-			mm[m] = nil
-			dataModels = append(dataModels, OpenAIModel{ID: m, Object: "model"})
-		}
-
-		for _, k := range cm.ListConfigs() {
-			if _, exists := mm[k]; !exists {
-				dataModels = append(dataModels, OpenAIModel{ID: k, Object: "model"})
-			}
-		}
-
-		return c.JSON(struct {
-			Object string        `json:"object"`
-			Data   []OpenAIModel `json:"data"`
-		}{
-			Object: "list",
-			Data:   dataModels,
-		})
-	}
-}

api/openai/api.go

@@ -0,0 +1,105 @@
+package openai
+
+import (
+	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 Message struct {
+	// The message role
+	Role string `json:"role,omitempty" yaml:"role"`
+	// The message content
+	Content *string `json:"content" yaml:"content"`
+	// 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 OpenAIRequest struct {
+	config.PredictionOptions
+
+	// whisper
+	File string `json:"file" validate:"required"`
+	//whisper/image
+	ResponseFormat string `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"`
+}

api/openai/chat.go

@@ -0,0 +1,322 @@
+package openai
+
+import (
+	"bufio"
+	"bytes"
+	"encoding/json"
+	"fmt"
+	"strings"
+
+	"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/pkg/grammar"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/gofiber/fiber/v2"
+	"github.com/rs/zerolog/log"
+	"github.com/valyala/fasthttp"
+)
+
+func ChatEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
+	emptyMessage := ""
+
+	process := func(s string, req *OpenAIRequest, config *config.Config, loader *model.ModelLoader, responses chan OpenAIResponse) {
+		initialMessage := OpenAIResponse{
+			Model:   req.Model, // we have to return what the user sent here, due to OpenAI spec.
+			Choices: []Choice{{Delta: &Message{Role: "assistant", Content: &emptyMessage}}},
+			Object:  "chat.completion.chunk",
+		}
+		responses <- initialMessage
+
+		ComputeChoices(s, req.N, config, o, loader, func(s string, c *[]Choice) {}, func(s string) bool {
+			resp := OpenAIResponse{
+				Model:   req.Model, // we have to return what the user sent here, due to OpenAI spec.
+				Choices: []Choice{{Delta: &Message{Content: &s}, Index: 0}},
+				Object:  "chat.completion.chunk",
+			}
+
+			responses <- resp
+			return true
+		})
+		close(responses)
+	}
+	return func(c *fiber.Ctx) error {
+		processFunctions := false
+		funcs := grammar.Functions{}
+		model, input, err := readInput(c, o.Loader, 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("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
+		}
+
+		// 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
+
+		mess := []string{}
+		for _, 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.Content != ""
+			if r != "" {
+				if contentExists {
+					content = fmt.Sprint(r, " ", *i.Content)
+				}
+				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.Content)
+				}
+				if i.FunctionCall != nil {
+					j, err := json.Marshal(i.FunctionCall)
+					if err == nil {
+						if contentExists {
+							content += "\n" + string(j)
+						} else {
+							content = string(j)
+						}
+					}
+				}
+			}
+
+			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 := config.Model
+
+		if config.TemplateConfig.Chat != "" && !processFunctions {
+			templateFile = config.TemplateConfig.Chat
+		}
+
+		if config.TemplateConfig.Functions != "" && processFunctions {
+			templateFile = config.TemplateConfig.Functions
+		}
+
+		// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
+		templatedInput, err := o.Loader.TemplatePrefix(templateFile, struct {
+			Input     string
+			Functions []grammar.Function
+		}{
+			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 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 := &OpenAIResponse{
+					Model: input.Model, // we have to return what the user sent here, due to OpenAI spec.
+					Choices: []Choice{
+						{
+							FinishReason: "stop",
+							Index:        0,
+							Delta:        &Message{Content: &emptyMessage},
+						}},
+					Object: "chat.completion.chunk",
+				}
+				respData, _ := json.Marshal(resp)
+
+				w.WriteString(fmt.Sprintf("data: %s\n\n", respData))
+				w.WriteString("data: [DONE]\n\n")
+				w.Flush()
+			}))
+			return nil
+		}
+
+		result, err := ComputeChoices(predInput, input.N, config, o, o.Loader, func(s string, c *[]Choice) {
+			if processFunctions {
+				// As we have to change the result before processing, we can't stream the answer (yet?)
+				ss := map[string]interface{}{}
+				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, Choice{Message: &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 = ""
+					predFunc, err := backend.ModelInference(predInput, 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
+					}
+
+					prediction = backend.Finetune(*config, predInput, prediction)
+					*c = append(*c, Choice{Message: &Message{Role: "assistant", Content: &prediction}})
+				} else {
+					// otherwise reply with the function call
+					*c = append(*c, Choice{
+						FinishReason: "function_call",
+						Message:      &Message{Role: "assistant", FunctionCall: ss},
+					})
+				}
+
+				return
+			}
+			*c = append(*c, Choice{FinishReason: "stop", Index: 0, Message: &Message{Role: "assistant", Content: &s}})
+		}, nil)
+		if err != nil {
+			return err
+		}
+
+		resp := &OpenAIResponse{
+			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
+			Choices: result,
+			Object:  "chat.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 +1,159 @@
+package openai
+
+import (
+	"bufio"
+	"bytes"
+	"encoding/json"
+	"errors"
+	"fmt"
+
+	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"
+	"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 {
+	process := func(s string, req *OpenAIRequest, config *config.Config, loader *model.ModelLoader, responses chan OpenAIResponse) {
+		ComputeChoices(s, req.N, config, o, loader, func(s string, c *[]Choice) {}, func(s string) bool {
+			resp := OpenAIResponse{
+				Model: req.Model, // we have to return what the user sent here, due to OpenAI spec.
+				Choices: []Choice{
+					{
+						Index: 0,
+						Text:  s,
+					},
+				},
+				Object: "text_completion",
+			}
+			log.Debug().Msgf("Sending goroutine: %s", s)
+
+			responses <- resp
+			return true
+		})
+		close(responses)
+	}
+
+	return func(c *fiber.Ctx) error {
+		model, input, err := readInput(c, o.Loader, true)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
+
+		log.Debug().Msgf("`input`: %+v", input)
+
+		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)
+
+		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 := 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]
+
+			// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
+			templatedInput, err := o.Loader.TemplatePrefix(templateFile, struct {
+				Input string
+			}{
+				Input: predInput,
+			})
+			if err == nil {
+				predInput = templatedInput
+				log.Debug().Msgf("Template found, input modified to: %s", predInput)
+			}
+
+			responses := make(chan 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 := &OpenAIResponse{
+					Model: input.Model, // we have to return what the user sent here, due to OpenAI spec.
+					Choices: []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 []Choice
+		for k, i := range config.PromptStrings {
+			// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
+			templatedInput, err := o.Loader.TemplatePrefix(templateFile, struct {
+				Input string
+			}{
+				Input: i,
+			})
+			if err == nil {
+				i = templatedInput
+				log.Debug().Msgf("Template found, input modified to: %s", i)
+			}
+
+			r, err := ComputeChoices(i, input.N, config, o, o.Loader, func(s string, c *[]Choice) {
+				*c = append(*c, Choice{Text: s, FinishReason: "stop", Index: k})
+			}, nil)
+			if err != nil {
+				return err
+			}
+
+			result = append(result, r...)
+		}
+
+		resp := &OpenAIResponse{
+			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
+			Choices: result,
+			Object:  "text_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 +1,67 @@
+package openai
+
+import (
+	"encoding/json"
+	"fmt"
+
+	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"
+)
+
+func EditEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
+	return func(c *fiber.Ctx) error {
+		model, input, err := readInput(c, o.Loader, 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)
+
+		templateFile := config.Model
+
+		if config.TemplateConfig.Edit != "" {
+			templateFile = config.TemplateConfig.Edit
+		}
+
+		var result []Choice
+		for _, i := range config.InputStrings {
+			// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
+			templatedInput, err := o.Loader.TemplatePrefix(templateFile, struct {
+				Input       string
+				Instruction string
+			}{Input: i})
+			if err == nil {
+				i = templatedInput
+				log.Debug().Msgf("Template found, input modified to: %s", i)
+			}
+
+			r, err := ComputeChoices(i, input.N, config, o, o.Loader, func(s string, c *[]Choice) {
+				*c = append(*c, Choice{Text: s})
+			}, nil)
+			if err != nil {
+				return err
+			}
+
+			result = append(result, r...)
+		}
+
+		resp := &OpenAIResponse{
+			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
+			Choices: result,
+			Object:  "edit",
+		}
+
+		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 +1,70 @@
+package openai
+
+import (
+	"encoding/json"
+	"fmt"
+
+	"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/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.Loader, 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 := []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, 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, Item{Embedding: embeddings, Index: i, Object: "embedding"})
+		}
+
+		resp := &OpenAIResponse{
+			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 +1,158 @@
+package openai
+
+import (
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"io/ioutil"
+	"os"
+	"path/filepath"
+	"strconv"
+	"strings"
+
+	"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"
+)
+
+// 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.Loader, 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)
+		}
+
+		log.Debug().Msgf("Parameter Config: %+v", config)
+
+		// XXX: Only stablediffusion is supported for now
+		if config.Backend == "" {
+			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 == "b64_json" {
+			b64JSON = true
+		}
+
+		var result []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 := 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 := ioutil.TempFile(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, output, o.Loader, *config, o)
+				if err != nil {
+					return err
+				}
+				if err := fn(); err != nil {
+					return err
+				}
+
+				item := &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)
+			}
+		}
+
+		resp := &OpenAIResponse{
+			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 +1,36 @@
+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"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+)
+
+func ComputeChoices(predInput string, n int, config *config.Config, o *options.Option, loader *model.ModelLoader, cb func(string, *[]Choice), tokenCallback func(string) bool) ([]Choice, error) {
+	result := []Choice{}
+
+	if n == 0 {
+		n = 1
+	}
+
+	// get the model function to call for the result
+	predFunc, err := backend.ModelInference(predInput, loader, *config, o, tokenCallback)
+	if err != nil {
+		return result, err
+	}
+
+	for i := 0; i < n; i++ {
+		prediction, err := predFunc()
+		if err != nil {
+			return result, err
+		}
+
+		prediction = backend.Finetune(*config, predInput, prediction)
+		cb(prediction, &result)
+
+		//result = append(result, Choice{Text: prediction})
+
+	}
+	return result, err
+}

api/openai/list.go

@@ -0,0 +1,37 @@
+package openai
+
+import (
+	config "github.com/go-skynet/LocalAI/api/config"
+	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 := []OpenAIModel{}
+		for _, m := range models {
+			mm[m] = nil
+			dataModels = append(dataModels, OpenAIModel{ID: m, Object: "model"})
+		}
+
+		for _, k := range cm.ListConfigs() {
+			if _, exists := mm[k]; !exists {
+				dataModels = append(dataModels, OpenAIModel{ID: k, Object: "model"})
+			}
+		}
+
+		return c.JSON(struct {
+			Object string        `json:"object"`
+			Data   []OpenAIModel `json:"data"`
+		}{
+			Object: "list",
+			Data:   dataModels,
+		})
+	}
+}

api/openai/request.go

@@ -0,0 +1,234 @@
+package openai
+
+import (
+	"encoding/json"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+
+	config "github.com/go-skynet/LocalAI/api/config"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/gofiber/fiber/v2"
+	"github.com/rs/zerolog/log"
+)
+
+func readInput(c *fiber.Ctx, loader *model.ModelLoader, randomModel bool) (string, *OpenAIRequest, error) {
+	input := new(OpenAIRequest)
+	// Get input data from the request body
+	if err := c.BodyParser(input); err != nil {
+		return "", nil, 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
+}
+
+func updateConfig(config *config.Config, input *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.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)
+			}
+		}
+	}
+
+	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.Mirostat = input.Mirostat
+	}
+
+	if input.MirostatETA != 0 {
+		config.MirostatETA = input.MirostatETA
+	}
+
+	if input.MirostatTAU != 0 {
+		config.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 *OpenAIRequest, cm *config.ConfigLoader, loader *model.ModelLoader, debug bool, threads, ctx int, f16 bool) (*config.Config, *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.Loader, 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.go

@@ -1,108 +0,0 @@
-package api
-
-import (
-	"context"
-
-	model "github.com/go-skynet/LocalAI/pkg/model"
-)
-
-type Option struct {
-	context                         context.Context
-	configFile                      string
-	loader                          *model.ModelLoader
-	uploadLimitMB, threads, ctxSize int
-	f16                             bool
-	debug, disableMessage           bool
-	imageDir                        string
-	cors                            bool
-	corsAllowOrigins                string
-}
-
-type AppOption func(*Option)
-
-func newOptions(o ...AppOption) *Option {
-	opt := &Option{
-		context:        context.Background(),
-		uploadLimitMB:  15,
-		threads:        1,
-		ctxSize:        512,
-		debug:          true,
-		disableMessage: true,
-	}
-	for _, oo := range o {
-		oo(opt)
-	}
-	return opt
-}
-
-func WithCors(b bool) AppOption {
-	return func(o *Option) {
-		o.cors = b
-	}
-}
-
-func WithCorsAllowOrigins(b string) AppOption {
-	return func(o *Option) {
-		o.corsAllowOrigins = b
-	}
-}
-
-func WithContext(ctx context.Context) AppOption {
-	return func(o *Option) {
-		o.context = ctx
-	}
-}
-
-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.ctxSize = 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 WithImageDir(imageDir string) AppOption {
-	return func(o *Option) {
-		o.imageDir = imageDir
-	}
-}

api/options/options.go

@@ -0,0 +1,186 @@
+package options
+
+import (
+	"context"
+	"embed"
+	"encoding/json"
+
+	"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
+
+	Galleries []gallery.Gallery
+
+	BackendAssets     embed.FS
+	AssetsDestination string
+
+	ExternalGRPCBackends map[string]string
+
+	AutoloadGalleries bool
+}
+
+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 WithCors(b bool) AppOption {
+	return func(o *Option) {
+		o.CORS = b
+	}
+}
+
+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")
+			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
+	}
+}

api/prediction.go

@@ -1,561 +0,0 @@
-package api
-
-import (
-	"fmt"
-	"regexp"
-	"strings"
-	"sync"
-
-	"github.com/donomii/go-rwkv.cpp"
-	model "github.com/go-skynet/LocalAI/pkg/model"
-	"github.com/go-skynet/LocalAI/pkg/stablediffusion"
-	"github.com/go-skynet/bloomz.cpp"
-	bert "github.com/go-skynet/go-bert.cpp"
-	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
-	llama "github.com/go-skynet/go-llama.cpp"
-	gpt4all "github.com/nomic-ai/gpt4all/gpt4all-bindings/golang"
-)
-
-// mutex still needed, see: https://github.com/ggerganov/llama.cpp/discussions/784
-var mutexMap sync.Mutex
-var mutexes map[string]*sync.Mutex = make(map[string]*sync.Mutex)
-
-func defaultLLamaOpts(c Config) []llama.ModelOption {
-	llamaOpts := []llama.ModelOption{}
-	if c.ContextSize != 0 {
-		llamaOpts = append(llamaOpts, llama.SetContext(c.ContextSize))
-	}
-	if c.F16 {
-		llamaOpts = append(llamaOpts, llama.EnableF16Memory)
-	}
-	if c.Embeddings {
-		llamaOpts = append(llamaOpts, llama.EnableEmbeddings)
-	}
-
-	if c.NGPULayers != 0 {
-		llamaOpts = append(llamaOpts, llama.SetGPULayers(c.NGPULayers))
-	}
-
-	return llamaOpts
-}
-
-func ImageGeneration(height, width, mode, step, seed int, positive_prompt, negative_prompt, dst string, loader *model.ModelLoader, c Config) (func() error, error) {
-	if c.Backend != model.StableDiffusionBackend {
-		return nil, fmt.Errorf("endpoint only working with stablediffusion models")
-	}
-	inferenceModel, err := loader.BackendLoader(c.Backend, c.ImageGenerationAssets, []llama.ModelOption{}, uint32(c.Threads))
-	if err != nil {
-		return nil, err
-	}
-
-	var fn func() error
-	switch model := inferenceModel.(type) {
-	case *stablediffusion.StableDiffusion:
-		fn = func() error {
-			return model.GenerateImage(height, width, mode, step, seed, positive_prompt, negative_prompt, dst)
-		}
-
-	default:
-		fn = func() error {
-			return fmt.Errorf("creation of images not supported by the backend")
-		}
-	}
-
-	return func() error {
-		// This is still needed, see: https://github.com/ggerganov/llama.cpp/discussions/784
-		mutexMap.Lock()
-		l, ok := mutexes[c.Backend]
-		if !ok {
-			m := &sync.Mutex{}
-			mutexes[c.Backend] = m
-			l = m
-		}
-		mutexMap.Unlock()
-		l.Lock()
-		defer l.Unlock()
-
-		return fn()
-	}, nil
-}
-
-func ModelEmbedding(s string, tokens []int, loader *model.ModelLoader, c Config) (func() ([]float32, error), error) {
-	if !c.Embeddings {
-		return nil, fmt.Errorf("endpoint disabled for this model by API configuration")
-	}
-
-	modelFile := c.Model
-
-	llamaOpts := defaultLLamaOpts(c)
-
-	var inferenceModel interface{}
-	var err error
-	if c.Backend == "" {
-		inferenceModel, err = loader.GreedyLoader(modelFile, llamaOpts, uint32(c.Threads))
-	} else {
-		inferenceModel, err = loader.BackendLoader(c.Backend, modelFile, llamaOpts, uint32(c.Threads))
-	}
-	if err != nil {
-		return nil, err
-	}
-
-	var fn func() ([]float32, error)
-	switch model := inferenceModel.(type) {
-	case *llama.LLama:
-		fn = func() ([]float32, error) {
-			predictOptions := buildLLamaPredictOptions(c)
-			if len(tokens) > 0 {
-				return model.TokenEmbeddings(tokens, predictOptions...)
-			}
-			return model.Embeddings(s, predictOptions...)
-		}
-	// bert embeddings
-	case *bert.Bert:
-		fn = func() ([]float32, error) {
-			if len(tokens) > 0 {
-				return model.TokenEmbeddings(tokens, bert.SetThreads(c.Threads))
-			}
-			return model.Embeddings(s, bert.SetThreads(c.Threads))
-		}
-	default:
-		fn = func() ([]float32, error) {
-			return nil, fmt.Errorf("embeddings not supported by the backend")
-		}
-	}
-
-	return func() ([]float32, error) {
-		// 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()
-
-		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
-}
-
-func buildLLamaPredictOptions(c Config) []llama.PredictOption {
-	// Generate the prediction using the language model
-	predictOptions := []llama.PredictOption{
-		llama.SetTemperature(c.Temperature),
-		llama.SetTopP(c.TopP),
-		llama.SetTopK(c.TopK),
-		llama.SetTokens(c.Maxtokens),
-		llama.SetThreads(c.Threads),
-	}
-
-	if c.Mirostat != 0 {
-		predictOptions = append(predictOptions, llama.SetMirostat(c.Mirostat))
-	}
-
-	if c.MirostatETA != 0 {
-		predictOptions = append(predictOptions, llama.SetMirostatETA(c.MirostatETA))
-	}
-
-	if c.MirostatTAU != 0 {
-		predictOptions = append(predictOptions, llama.SetMirostatTAU(c.MirostatTAU))
-	}
-
-	if c.Debug {
-		predictOptions = append(predictOptions, llama.Debug)
-	}
-
-	predictOptions = append(predictOptions, llama.SetStopWords(c.StopWords...))
-
-	if c.RepeatPenalty != 0 {
-		predictOptions = append(predictOptions, llama.SetPenalty(c.RepeatPenalty))
-	}
-
-	if c.Keep != 0 {
-		predictOptions = append(predictOptions, llama.SetNKeep(c.Keep))
-	}
-
-	if c.Batch != 0 {
-		predictOptions = append(predictOptions, llama.SetBatch(c.Batch))
-	}
-
-	if c.F16 {
-		predictOptions = append(predictOptions, llama.EnableF16KV)
-	}
-
-	if c.IgnoreEOS {
-		predictOptions = append(predictOptions, llama.IgnoreEOS)
-	}
-
-	if c.Seed != 0 {
-		predictOptions = append(predictOptions, llama.SetSeed(c.Seed))
-	}
-
-	return predictOptions
-}
-
-func ModelInference(s string, loader *model.ModelLoader, c Config, tokenCallback func(string) bool) (func() (string, error), error) {
-	supportStreams := false
-	modelFile := c.Model
-
-	llamaOpts := defaultLLamaOpts(c)
-
-	var inferenceModel interface{}
-	var err error
-	if c.Backend == "" {
-		inferenceModel, err = loader.GreedyLoader(modelFile, llamaOpts, uint32(c.Threads))
-	} else {
-		inferenceModel, err = loader.BackendLoader(c.Backend, modelFile, llamaOpts, uint32(c.Threads))
-	}
-	if err != nil {
-		return nil, err
-	}
-
-	var fn func() (string, error)
-
-	switch model := inferenceModel.(type) {
-	case *rwkv.RwkvState:
-		supportStreams = true
-
-		fn = func() (string, error) {
-			stopWord := "\n"
-			if len(c.StopWords) > 0 {
-				stopWord = c.StopWords[0]
-			}
-
-			if err := model.ProcessInput(s); err != nil {
-				return "", err
-			}
-
-			response := model.GenerateResponse(c.Maxtokens, stopWord, float32(c.Temperature), float32(c.TopP), tokenCallback)
-
-			return response, nil
-		}
-	case *transformers.GPTNeoX:
-		fn = func() (string, error) {
-			// Generate the prediction using the language model
-			predictOptions := []transformers.PredictOption{
-				transformers.SetTemperature(c.Temperature),
-				transformers.SetTopP(c.TopP),
-				transformers.SetTopK(c.TopK),
-				transformers.SetTokens(c.Maxtokens),
-				transformers.SetThreads(c.Threads),
-			}
-
-			if c.Batch != 0 {
-				predictOptions = append(predictOptions, transformers.SetBatch(c.Batch))
-			}
-
-			if c.Seed != 0 {
-				predictOptions = append(predictOptions, transformers.SetSeed(c.Seed))
-			}
-
-			return model.Predict(
-				s,
-				predictOptions...,
-			)
-		}
-	case *transformers.Replit:
-		fn = func() (string, error) {
-			// Generate the prediction using the language model
-			predictOptions := []transformers.PredictOption{
-				transformers.SetTemperature(c.Temperature),
-				transformers.SetTopP(c.TopP),
-				transformers.SetTopK(c.TopK),
-				transformers.SetTokens(c.Maxtokens),
-				transformers.SetThreads(c.Threads),
-			}
-
-			if c.Batch != 0 {
-				predictOptions = append(predictOptions, transformers.SetBatch(c.Batch))
-			}
-
-			if c.Seed != 0 {
-				predictOptions = append(predictOptions, transformers.SetSeed(c.Seed))
-			}
-
-			return model.Predict(
-				s,
-				predictOptions...,
-			)
-		}
-	case *transformers.Starcoder:
-		fn = func() (string, error) {
-			// Generate the prediction using the language model
-			predictOptions := []transformers.PredictOption{
-				transformers.SetTemperature(c.Temperature),
-				transformers.SetTopP(c.TopP),
-				transformers.SetTopK(c.TopK),
-				transformers.SetTokens(c.Maxtokens),
-				transformers.SetThreads(c.Threads),
-			}
-
-			if c.Batch != 0 {
-				predictOptions = append(predictOptions, transformers.SetBatch(c.Batch))
-			}
-
-			if c.Seed != 0 {
-				predictOptions = append(predictOptions, transformers.SetSeed(c.Seed))
-			}
-
-			return model.Predict(
-				s,
-				predictOptions...,
-			)
-		}
-	case *transformers.MPT:
-		fn = func() (string, error) {
-			// Generate the prediction using the language model
-			predictOptions := []transformers.PredictOption{
-				transformers.SetTemperature(c.Temperature),
-				transformers.SetTopP(c.TopP),
-				transformers.SetTopK(c.TopK),
-				transformers.SetTokens(c.Maxtokens),
-				transformers.SetThreads(c.Threads),
-			}
-
-			if c.Batch != 0 {
-				predictOptions = append(predictOptions, transformers.SetBatch(c.Batch))
-			}
-
-			if c.Seed != 0 {
-				predictOptions = append(predictOptions, transformers.SetSeed(c.Seed))
-			}
-
-			return model.Predict(
-				s,
-				predictOptions...,
-			)
-		}
-	case *bloomz.Bloomz:
-		fn = func() (string, error) {
-			// Generate the prediction using the language model
-			predictOptions := []bloomz.PredictOption{
-				bloomz.SetTemperature(c.Temperature),
-				bloomz.SetTopP(c.TopP),
-				bloomz.SetTopK(c.TopK),
-				bloomz.SetTokens(c.Maxtokens),
-				bloomz.SetThreads(c.Threads),
-			}
-
-			if c.Seed != 0 {
-				predictOptions = append(predictOptions, bloomz.SetSeed(c.Seed))
-			}
-
-			return model.Predict(
-				s,
-				predictOptions...,
-			)
-		}
-	case *transformers.GPTJ:
-		fn = func() (string, error) {
-			// Generate the prediction using the language model
-			predictOptions := []transformers.PredictOption{
-				transformers.SetTemperature(c.Temperature),
-				transformers.SetTopP(c.TopP),
-				transformers.SetTopK(c.TopK),
-				transformers.SetTokens(c.Maxtokens),
-				transformers.SetThreads(c.Threads),
-			}
-
-			if c.Batch != 0 {
-				predictOptions = append(predictOptions, transformers.SetBatch(c.Batch))
-			}
-
-			if c.Seed != 0 {
-				predictOptions = append(predictOptions, transformers.SetSeed(c.Seed))
-			}
-
-			return model.Predict(
-				s,
-				predictOptions...,
-			)
-		}
-	case *transformers.Dolly:
-		fn = func() (string, error) {
-			// Generate the prediction using the language model
-			predictOptions := []transformers.PredictOption{
-				transformers.SetTemperature(c.Temperature),
-				transformers.SetTopP(c.TopP),
-				transformers.SetTopK(c.TopK),
-				transformers.SetTokens(c.Maxtokens),
-				transformers.SetThreads(c.Threads),
-			}
-
-			if c.Batch != 0 {
-				predictOptions = append(predictOptions, transformers.SetBatch(c.Batch))
-			}
-
-			if c.Seed != 0 {
-				predictOptions = append(predictOptions, transformers.SetSeed(c.Seed))
-			}
-
-			return model.Predict(
-				s,
-				predictOptions...,
-			)
-		}
-	case *transformers.GPT2:
-		fn = func() (string, error) {
-			// Generate the prediction using the language model
-			predictOptions := []transformers.PredictOption{
-				transformers.SetTemperature(c.Temperature),
-				transformers.SetTopP(c.TopP),
-				transformers.SetTopK(c.TopK),
-				transformers.SetTokens(c.Maxtokens),
-				transformers.SetThreads(c.Threads),
-			}
-
-			if c.Batch != 0 {
-				predictOptions = append(predictOptions, transformers.SetBatch(c.Batch))
-			}
-
-			if c.Seed != 0 {
-				predictOptions = append(predictOptions, transformers.SetSeed(c.Seed))
-			}
-
-			return model.Predict(
-				s,
-				predictOptions...,
-			)
-		}
-	case *gpt4all.Model:
-		supportStreams = true
-
-		fn = func() (string, error) {
-			if tokenCallback != nil {
-				model.SetTokenCallback(tokenCallback)
-			}
-
-			// Generate the prediction using the language model
-			predictOptions := []gpt4all.PredictOption{
-				gpt4all.SetTemperature(c.Temperature),
-				gpt4all.SetTopP(c.TopP),
-				gpt4all.SetTopK(c.TopK),
-				gpt4all.SetTokens(c.Maxtokens),
-			}
-
-			if c.Batch != 0 {
-				predictOptions = append(predictOptions, gpt4all.SetBatch(c.Batch))
-			}
-
-			str, er := model.Predict(
-				s,
-				predictOptions...,
-			)
-			// Seems that if we don't free the callback explicitly we leave functions registered (that might try to send on closed channels)
-			// For instance otherwise the API returns: {"error":{"code":500,"message":"send on closed channel","type":""}}
-			// after a stream event has occurred
-			model.SetTokenCallback(nil)
-			return str, er
-		}
-	case *llama.LLama:
-		supportStreams = true
-		fn = func() (string, error) {
-
-			if tokenCallback != nil {
-				model.SetTokenCallback(tokenCallback)
-			}
-
-			predictOptions := buildLLamaPredictOptions(c)
-
-			str, er := model.Predict(
-				s,
-				predictOptions...,
-			)
-			// Seems that if we don't free the callback explicitly we leave functions registered (that might try to send on closed channels)
-			// For instance otherwise the API returns: {"error":{"code":500,"message":"send on closed channel","type":""}}
-			// after a stream event has occurred
-			model.SetTokenCallback(nil)
-			return str, er
-		}
-	}
-
-	return func() (string, error) {
-		// 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()
-
-		res, err := fn()
-		if tokenCallback != nil && !supportStreams {
-			tokenCallback(res)
-		}
-		return res, err
-	}, nil
-}
-
-func ComputeChoices(predInput string, input *OpenAIRequest, config *Config, loader *model.ModelLoader, cb func(string, *[]Choice), tokenCallback func(string) bool) ([]Choice, error) {
-	result := []Choice{}
-
-	n := input.N
-
-	if input.N == 0 {
-		n = 1
-	}
-
-	// get the model function to call for the result
-	predFunc, err := ModelInference(predInput, loader, *config, tokenCallback)
-	if err != nil {
-		return result, err
-	}
-
-	for i := 0; i < n; i++ {
-		prediction, err := predFunc()
-		if err != nil {
-			return result, err
-		}
-
-		prediction = Finetune(*config, predInput, prediction)
-		cb(prediction, &result)
-
-		//result = append(result, Choice{Text: prediction})
-
-	}
-	return result, err
-}
-
-var cutstrings map[string]*regexp.Regexp = make(map[string]*regexp.Regexp)
-var mu sync.Mutex = sync.Mutex{}
-
-func Finetune(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))
-	}
-	return prediction
-
-}

assets.go

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

cmd/grpc/bert-embeddings/main.go

@@ -0,0 +1,22 @@
+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"
+	bert "github.com/go-skynet/LocalAI/pkg/grpc/llm/bert"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &bert.Embeddings{}); err != nil {
+		panic(err)
+	}
+}

cmd/grpc/bloomz/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"
+
+	bloomz "github.com/go-skynet/LocalAI/pkg/grpc/llm/bloomz"
+
+	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, &bloomz.LLM{}); err != nil {
+		panic(err)
+	}
+}

cmd/grpc/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/pkg/grpc/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)
+	}
+}

cmd/grpc/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/pkg/grpc/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)
+	}
+}

cmd/grpc/falcon/main.go

@@ -0,0 +1,25 @@
+package main
+
+// GRPC Falcon server
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	falcon "github.com/go-skynet/LocalAI/pkg/grpc/llm/falcon"
+
+	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, &falcon.LLM{}); err != nil {
+		panic(err)
+	}
+}

cmd/grpc/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/pkg/grpc/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)
+	}
+}

cmd/grpc/gpt4all/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"
+
+	gpt4all "github.com/go-skynet/LocalAI/pkg/grpc/llm/gpt4all"
+
+	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, &gpt4all.LLM{}); err != nil {
+		panic(err)
+	}
+}

cmd/grpc/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/pkg/grpc/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)
+	}
+}

cmd/grpc/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/pkg/grpc/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)
+	}
+}

cmd/grpc/langchain-huggingface/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"
+
+	langchain "github.com/go-skynet/LocalAI/pkg/grpc/llm/langchain"
+
+	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, &langchain.LLM{}); err != nil {
+		panic(err)
+	}
+}

cmd/grpc/llama-grammar/main.go

@@ -0,0 +1,25 @@
+package main
+
+// GRPC Falcon server
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	llama "github.com/go-skynet/LocalAI/pkg/grpc/llm/llama-grammar"
+
+	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, &llama.LLM{}); err != nil {
+		panic(err)
+	}
+}

cmd/grpc/llama/main.go

@@ -0,0 +1,25 @@
+package main
+
+// GRPC Falcon server
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	llama "github.com/go-skynet/LocalAI/pkg/grpc/llm/llama"
+
+	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, &llama.LLM{}); err != nil {
+		panic(err)
+	}
+}

cmd/grpc/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/pkg/grpc/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)
+	}
+}

cmd/grpc/piper/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"
+
+	tts "github.com/go-skynet/LocalAI/pkg/grpc/tts"
+
+	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, &tts.Piper{}); err != nil {
+		panic(err)
+	}
+}

cmd/grpc/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/pkg/grpc/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)
+	}
+}

cmd/grpc/rwkv/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"
+
+	rwkv "github.com/go-skynet/LocalAI/pkg/grpc/llm/rwkv"
+
+	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, &rwkv.LLM{}); err != nil {
+		panic(err)
+	}
+}

cmd/grpc/stablediffusion/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"
+
+	image "github.com/go-skynet/LocalAI/pkg/grpc/image"
+
+	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.StableDiffusion{}); err != nil {
+		panic(err)
+	}
+}

cmd/grpc/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/pkg/grpc/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)
+	}
+}

cmd/grpc/whisper/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"
+
+	transcribe "github.com/go-skynet/LocalAI/pkg/grpc/transcribe"
+
+	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, &transcribe.Whisper{}); err != nil {
+		panic(err)
+	}
+}

docker-compose.yaml

@@ -5,7 +5,7 @@ services:
     image: quay.io/go-skynet/local-ai:latest
     build:
       context: .
-      dockerfile: Dockerfile.dev
+      dockerfile: Dockerfile
     ports:
       - 8080:8080
     env_file:

entrypoint.sh

@@ -1,9 +1,21 @@
 #!/bin/bash
+set -e
 
 cd /build
 
 if [ "$REBUILD" != "false" ]; then
-	make rebuild
+	rm -rf ./local-ai
+	ESPEAK_DATA=/build/lib/Linux-$(uname -m)/piper_phonemize/lib/espeak-ng-data make build -j${BUILD_PARALLELISM:-1}
+else
+	echo "@@@@@"
+	echo "Skipping rebuild"
+	echo "@@@@@"
+	echo "If you are experiencing issues with the pre-compiled builds, try setting REBUILD=true"
+	echo "If you are still experiencing issues with the build, try setting CMAKE_ARGS and disable the instructions set as needed:"
+	echo 'CMAKE_ARGS="-DLLAMA_F16C=OFF -DLLAMA_AVX512=OFF -DLLAMA_AVX2=OFF -DLLAMA_FMA=OFF"'
+	echo "see the documentation at: https://localai.io/basics/build/index.html"
+	echo "Note: See also https://github.com/go-skynet/LocalAI/issues/288"
+	echo "@@@@@"
 fi
 
 ./local-ai "$@"

examples/README.md

@@ -4,6 +4,13 @@ Here is a list of projects that can easily be integrated with the LocalAI backen
 
 ### Projects
 
+### AutoGPT
+
+_by [@mudler](https://github.com/mudler)_
+
+This example shows how to use AutoGPT with LocalAI.
+
+[Check it out here](https://github.com/go-skynet/LocalAI/tree/master/examples/autoGPT/)
 
 ### Chatbot-UI
 
@@ -15,6 +22,24 @@ This integration shows how to use LocalAI with [mckaywrigley/chatbot-ui](https:/
 
 [Check it out here](https://github.com/go-skynet/LocalAI/tree/master/examples/chatbot-ui/)
 
+There is also a separate example to show how to manually setup a model: [example](https://github.com/go-skynet/LocalAI/tree/master/examples/chatbot-ui-manual/)
+
+### K8sGPT
+
+_by [@mudler](https://github.com/mudler)_
+
+This example show how to use LocalAI inside Kubernetes with [k8sgpt](https://k8sgpt.ai).
+
+![Screenshot from 2023-06-19 23-58-47](https://github.com/go-skynet/go-ggml-transformers.cpp/assets/2420543/cab87409-ee68-44ae-8d53-41627fb49509)
+
+### Flowise
+
+_by [@mudler](https://github.com/mudler)_
+
+This example shows how to use [FlowiseAI/Flowise](https://github.com/FlowiseAI/Flowise) with LocalAI.
+
+[Check it out here](https://github.com/go-skynet/LocalAI/tree/master/examples/flowise/)
+
 ### Discord bot
 
 _by [@mudler](https://github.com/mudler)_
@@ -39,6 +64,14 @@ A ready to use example to show e2e how to integrate LocalAI with langchain
 
 [Check it out here](https://github.com/go-skynet/LocalAI/tree/master/examples/langchain-python/)
 
+### LocalAI functions
+
+_by [@mudler](https://github.com/mudler)_
+
+A ready to use example to show how to use OpenAI functions with LocalAI
+
+[Check it out here](https://github.com/go-skynet/LocalAI/tree/master/examples/functions/)
+
 ### LocalAI WebUI
 
 _by [@dhruvgera](https://github.com/dhruvgera)_
@@ -73,6 +106,14 @@ Run a slack bot which lets you talk directly with a model
 
 [Check it out here](https://github.com/go-skynet/LocalAI/tree/master/examples/slack-bot/)
 
+### Slack bot (Question answering)
+
+_by [@mudler](https://github.com/mudler)_
+
+Run a slack bot, ideally for teams, which lets you ask questions on a documentation website, or a github repository.
+
+[Check it out here](https://github.com/go-skynet/LocalAI/tree/master/examples/slack-qa-bot/)
+
 ### Question answering on documents with llama-index
 
 _by [@mudler](https://github.com/mudler)_
@@ -89,6 +130,16 @@ Shows how to integrate with `Langchain` and `Chroma` to enable question answerin
 
 [Check it out here](https://github.com/go-skynet/LocalAI/tree/master/examples/langchain-chroma/)
 
+### Telegram bot
+
+_by [@mudler](https://github.com/mudler)
+
+![Screenshot from 2023-06-09 00-36-26](https://github.com/go-skynet/LocalAI/assets/2420543/e98b4305-fa2d-41cf-9d2f-1bb2d75ca902)
+
+Use LocalAI to power a Telegram bot assistant, with Image generation and audio support!
+
+[Check it out here](https://github.com/go-skynet/LocalAI/tree/master/examples/telegram-bot/)
+
 ### Template for Runpod.io
 
 _by [@fHachenberg](https://github.com/fHachenberg)_

examples/autoGPT/.env

@@ -0,0 +1,5 @@
+OPENAI_API_KEY=sk---anystringhere
+OPENAI_API_BASE=http://api:8080/v1
+# Models to preload at start
+# Here we configure gpt4all as gpt-3.5-turbo and bert as embeddings
+PRELOAD_MODELS=[{"url": "github:go-skynet/model-gallery/gpt4all-j.yaml", "name": "gpt-3.5-turbo"}, { "url": "github:go-skynet/model-gallery/bert-embeddings.yaml", "name": "text-embedding-ada-002"}]

examples/autoGPT/README.md

@@ -0,0 +1,32 @@
+# AutoGPT
+
+Example of integration with [AutoGPT](https://github.com/Significant-Gravitas/Auto-GPT).
+
+## Run
+
+```bash
+# Clone LocalAI
+git clone https://github.com/go-skynet/LocalAI
+
+cd LocalAI/examples/autoGPT
+
+docker-compose run --rm auto-gpt
+```
+
+Note: The example automatically downloads the `gpt4all` model as it is under a permissive license. The GPT4All model does not seem to be enough to run AutoGPT. WizardLM-7b-uncensored seems to perform better (with `f16: true`).
+
+See the `.env` configuration file to set a different model with the [model-gallery](https://github.com/go-skynet/model-gallery) by editing `PRELOAD_MODELS`.
+
+## Without docker
+
+Run AutoGPT with `OPENAI_API_BASE` pointing to the LocalAI endpoint. If you run it locally for instance:
+
+```
+OPENAI_API_BASE=http://localhost:8080 python ...
+```
+
+Note: you need a model named `gpt-3.5-turbo` and `text-embedding-ada-002`. You can preload those in LocalAI at start by setting in the env:
+
+```
+PRELOAD_MODELS=[{"url": "github:go-skynet/model-gallery/gpt4all-j.yaml", "name": "gpt-3.5-turbo"}, { "url": "github:go-skynet/model-gallery/bert-embeddings.yaml", "name": "text-embedding-ada-002"}]
+```

examples/autoGPT/docker-compose.yaml

@@ -0,0 +1,42 @@
+version: "3.9"
+services:
+  api:
+    image: quay.io/go-skynet/local-ai:latest
+    ports:
+      - 8080:8080
+    env_file:
+      - .env
+    environment:
+      - DEBUG=true
+      - MODELS_PATH=/models
+    volumes:
+      - ./models:/models:cached
+    command: ["/usr/bin/local-ai" ]
+  auto-gpt:
+    image: significantgravitas/auto-gpt
+    depends_on:
+      api:
+        condition: service_healthy
+      redis:
+        condition: service_started
+    env_file:
+      - .env
+    environment:
+      MEMORY_BACKEND: ${MEMORY_BACKEND:-redis}
+      REDIS_HOST: ${REDIS_HOST:-redis}
+    profiles: ["exclude-from-up"]
+    volumes:
+      - ./auto_gpt_workspace:/app/autogpt/auto_gpt_workspace
+      - ./data:/app/data
+      ## allow auto-gpt to write logs to disk
+      - ./logs:/app/logs
+      ## uncomment following lines if you want to make use of these files
+      ## you must have them existing in the same folder as this docker-compose.yml
+      #- type: bind
+      #  source: ./azure.yaml
+      #  target: /app/azure.yaml
+      #- type: bind
+      #  source: ./ai_settings.yaml
+      #  target: /app/ai_settings.yaml
+  redis:
+    image: "redis/redis-stack-server:latest"

examples/chatbot-ui-manual/README.md

@@ -0,0 +1,48 @@
+# chatbot-ui
+
+Example of integration with [mckaywrigley/chatbot-ui](https://github.com/mckaywrigley/chatbot-ui).
+
+![Screenshot from 2023-04-26 23-59-55](https://user-images.githubusercontent.com/2420543/234715439-98d12e03-d3ce-4f94-ab54-2b256808e05e.png)
+
+## Setup
+
+```bash
+# Clone LocalAI
+git clone https://github.com/go-skynet/LocalAI
+
+cd LocalAI/examples/chatbot-ui
+
+# (optional) Checkout a specific LocalAI tag
+# git checkout -b build <TAG>
+
+# Download gpt4all-j to models/
+wget https://gpt4all.io/models/ggml-gpt4all-j.bin -O models/ggml-gpt4all-j
+
+# start with docker-compose
+docker-compose up -d --pull always
+# or you can build the images with:
+# docker-compose up -d --build
+```
+
+## Pointing chatbot-ui to a separately managed LocalAI service
+
+If you want to use the [chatbot-ui example](https://github.com/go-skynet/LocalAI/tree/master/examples/chatbot-ui) with an externally managed LocalAI service, you can alter the `docker-compose` file so that it looks like the below. You will notice the file is smaller, because we have removed the section that would normally start the LocalAI service. Take care to update the IP address (or FQDN) that the chatbot-ui service tries to access (marked `<<LOCALAI_IP>>` below):
+```
+version: '3.6'
+
+services:
+  chatgpt:
+    image: ghcr.io/mckaywrigley/chatbot-ui:main
+    ports:
+      - 3000:3000
+    environment:
+      - 'OPENAI_API_KEY=sk-XXXXXXXXXXXXXXXXXXXX'
+      - 'OPENAI_API_HOST=http://<<LOCALAI_IP>>:8080'
+```
+
+Once you've edited the Dockerfile, you can start it with `docker compose up`, then browse to `http://localhost:3000`.
+
+## Accessing chatbot-ui
+
+Open http://localhost:3000 for the Web UI.
+

examples/chatbot-ui-manual/docker-compose.yaml

@@ -0,0 +1,24 @@
+version: '3.6'
+
+services:
+  api:
+    image: quay.io/go-skynet/local-ai:latest
+    build:
+      context: ../../
+      dockerfile: Dockerfile
+    ports:
+      - 8080:8080
+    environment:
+      - DEBUG=true
+      - MODELS_PATH=/models
+    volumes:
+      - ./models:/models:cached
+    command: ["/usr/bin/local-ai" ]
+
+  chatgpt:
+    image: ghcr.io/mckaywrigley/chatbot-ui:main
+    ports:
+      - 3000:3000
+    environment:
+      - 'OPENAI_API_KEY=sk-XXXXXXXXXXXXXXXXXXXX'
+      - 'OPENAI_API_HOST=http://api:8080'

examples/chatbot-ui/README.md

@@ -4,22 +4,18 @@ Example of integration with [mckaywrigley/chatbot-ui](https://github.com/mckaywr
 
 ![Screenshot from 2023-04-26 23-59-55](https://user-images.githubusercontent.com/2420543/234715439-98d12e03-d3ce-4f94-ab54-2b256808e05e.png)
 
-## Setup
+## Run
 
+In this example LocalAI will download the gpt4all model and set it up as "gpt-3.5-turbo". See the `docker-compose.yaml`
 ```bash
 # Clone LocalAI
 git clone https://github.com/go-skynet/LocalAI
 
 cd LocalAI/examples/chatbot-ui
 
-# (optional) Checkout a specific LocalAI tag
-# git checkout -b build <TAG>
-
-# Download gpt4all-j to models/
-wget https://gpt4all.io/models/ggml-gpt4all-j.bin -O models/ggml-gpt4all-j
-
 # start with docker-compose
-docker-compose up -d --pull always
+docker-compose up --pull always
+
 # or you can build the images with:
 # docker-compose up -d --build
 ```

examples/chatbot-ui/docker-compose.yaml

@@ -3,19 +3,32 @@ version: '3.6'
 services:
   api:
     image: quay.io/go-skynet/local-ai:latest
+    # As initially LocalAI will download the models defined in PRELOAD_MODELS
+    # you might need to tweak the healthcheck values here according to your network connection.
+    # Here we give a timespan of 20m to download all the required files.
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8080/readyz"]
+      interval: 1m
+      timeout: 20m
+      retries: 20
     build:
       context: ../../
-      dockerfile: Dockerfile.dev
+      dockerfile: Dockerfile
     ports:
       - 8080:8080
     environment:
       - DEBUG=true
       - MODELS_PATH=/models
+      # You can preload different models here as well.
+      # See: https://github.com/go-skynet/model-gallery
+      - 'PRELOAD_MODELS=[{"url": "github:go-skynet/model-gallery/gpt4all-j.yaml", "name": "gpt-3.5-turbo"}]'
     volumes:
       - ./models:/models:cached
     command: ["/usr/bin/local-ai" ]
-
   chatgpt:
+    depends_on:
+      api:
+        condition: service_healthy
     image: ghcr.io/mckaywrigley/chatbot-ui:main
     ports:
       - 3000:3000

examples/discord-bot/docker-compose.yaml

@@ -5,7 +5,7 @@ services:
     image: quay.io/go-skynet/local-ai:latest
     build:
       context: ../../
-      dockerfile: Dockerfile.dev
+      dockerfile: Dockerfile
     ports:
       - 8080:8080
     environment:

examples/flowise/README.md

@@ -0,0 +1,30 @@
+# flowise
+
+Example of integration with [FlowiseAI/Flowise](https://github.com/FlowiseAI/Flowise).
+
+![Screenshot from 2023-05-30 18-01-03](https://github.com/go-skynet/LocalAI/assets/2420543/02458782-0549-4131-971c-95ee56ec1af8)
+
+You can check a demo video in the Flowise PR: https://github.com/FlowiseAI/Flowise/pull/123
+
+## Run
+
+In this example LocalAI will download the gpt4all model and set it up as "gpt-3.5-turbo". See the `docker-compose.yaml`
+```bash
+# Clone LocalAI
+git clone https://github.com/go-skynet/LocalAI
+
+cd LocalAI/examples/flowise
+
+# start with docker-compose
+docker-compose up --pull always
+
+```
+
+## Accessing flowise
+
+Open http://localhost:3000.
+
+## Using LocalAI
+
+Search for LocalAI in the integration, and use the `http://api:8080/` as URL.
+

examples/flowise/docker-compose.yaml

@@ -0,0 +1,37 @@
+version: '3.6'
+
+services:
+  api:
+    image: quay.io/go-skynet/local-ai:latest
+    # As initially LocalAI will download the models defined in PRELOAD_MODELS
+    # you might need to tweak the healthcheck values here according to your network connection.
+    # Here we give a timespan of 20m to download all the required files.
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8080/readyz"]
+      interval: 1m
+      timeout: 20m
+      retries: 20
+    build:
+      context: ../../
+      dockerfile: Dockerfile
+    ports:
+      - 8080:8080
+    environment:
+      - DEBUG=true
+      - MODELS_PATH=/models
+      # You can preload different models here as well.
+      # See: https://github.com/go-skynet/model-gallery
+      - 'PRELOAD_MODELS=[{"url": "github:go-skynet/model-gallery/gpt4all-j.yaml", "name": "gpt-3.5-turbo"}]'
+    volumes:
+      - ./models:/models:cached
+    command: ["/usr/bin/local-ai" ]
+  flowise:
+    depends_on:
+      api:
+        condition: service_healthy
+    image: flowiseai/flowise
+    ports:
+      - 3000:3000
+    volumes:
+        - ~/.flowise:/root/.flowise
+    command: /bin/sh -c "sleep 3; flowise start"

examples/functions/.env

@@ -0,0 +1,9 @@
+OPENAI_API_KEY=sk---anystringhere
+OPENAI_API_BASE=http://api:8080/v1
+# Models to preload at start
+# Here we configure gpt4all as gpt-3.5-turbo and bert as embeddings
+PRELOAD_MODELS=[{"url": "github:go-skynet/model-gallery/openllama-7b-open-instruct.yaml", "name": "gpt-3.5-turbo"}]
+
+## Change the default number of threads
+#THREADS=14
+

examples/functions/Dockerfile

@@ -0,0 +1,5 @@
+FROM python:3.10-bullseye
+COPY . /app
+WORKDIR /app
+RUN pip install --no-cache-dir -r requirements.txt
+ENTRYPOINT [ "python", "./functions-openai.py" ];

examples/functions/README.md

@@ -0,0 +1,18 @@
+# LocalAI functions
+
+Example of using LocalAI functions, see the [OpenAI](https://openai.com/blog/function-calling-and-other-api-updates) blog post.
+
+## Run
+
+```bash
+# Clone LocalAI
+git clone https://github.com/go-skynet/LocalAI
+
+cd LocalAI/examples/functions
+
+docker-compose run --rm functions
+```
+
+Note: The example automatically downloads the `openllama` model as it is under a permissive license.
+
+See the `.env` configuration file to set a different model with the [model-gallery](https://github.com/go-skynet/model-gallery) by editing `PRELOAD_MODELS`.

examples/functions/docker-compose.yaml

@@ -0,0 +1,23 @@
+version: "3.9"
+services:
+  api:
+    image: quay.io/go-skynet/local-ai:master
+    ports:
+      - 8080:8080
+    env_file:
+      - .env
+    environment:
+      - DEBUG=true
+      - MODELS_PATH=/models
+    volumes:
+      - ./models:/models:cached
+    command: ["/usr/bin/local-ai" ]
+  functions:
+    build:
+      context: .
+      dockerfile: Dockerfile
+    depends_on:
+      api:
+        condition: service_healthy
+    env_file:
+      - .env

examples/functions/functions-openai.py

@@ -0,0 +1,76 @@
+import openai
+import json
+
+# Example dummy function hard coded to return the same weather
+# In production, this could be your backend API or an external API
+def get_current_weather(location, unit="fahrenheit"):
+    """Get the current weather in a given location"""
+    weather_info = {
+        "location": location,
+        "temperature": "72",
+        "unit": unit,
+        "forecast": ["sunny", "windy"],
+    }
+    return json.dumps(weather_info)
+
+
+def run_conversation():
+    # Step 1: send the conversation and available functions to GPT
+    messages = [{"role": "user", "content": "What's the weather like in Boston?"}]
+    functions = [
+        {
+            "name": "get_current_weather",
+            "description": "Get the current weather in a given location",
+            "parameters": {
+                "type": "object",
+                "properties": {
+                    "location": {
+                        "type": "string",
+                        "description": "The city and state, e.g. San Francisco, CA",
+                    },
+                    "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]},
+                },
+                "required": ["location"],
+            },
+        }
+    ]
+    response = openai.ChatCompletion.create(
+        model="gpt-3.5-turbo",
+        messages=messages,
+        functions=functions,
+        function_call="auto",  # auto is default, but we'll be explicit
+    )
+    response_message = response["choices"][0]["message"]
+
+    # Step 2: check if GPT wanted to call a function
+    if response_message.get("function_call"):
+        # Step 3: call the function
+        # Note: the JSON response may not always be valid; be sure to handle errors
+        available_functions = {
+            "get_current_weather": get_current_weather,
+        }  # only one function in this example, but you can have multiple
+        function_name = response_message["function_call"]["name"]
+        fuction_to_call = available_functions[function_name]
+        function_args = json.loads(response_message["function_call"]["arguments"])
+        function_response = fuction_to_call(
+            location=function_args.get("location"),
+            unit=function_args.get("unit"),
+        )
+
+        # Step 4: send the info on the function call and function response to GPT
+        messages.append(response_message)  # extend conversation with assistant's reply
+        messages.append(
+            {
+                "role": "function",
+                "name": function_name,
+                "content": function_response,
+            }
+        )  # extend conversation with function response
+        second_response = openai.ChatCompletion.create(
+            model="gpt-3.5-turbo",
+            messages=messages,
+        )  # get a new response from GPT where it can see the function response
+        return second_response
+
+
+print(run_conversation())

examples/functions/requirements.txt

@@ -0,0 +1,2 @@
+langchain==0.0.234
+openai==0.27.8

examples/k8sgpt/README.md

@@ -0,0 +1,70 @@
+# k8sgpt example
+
+This example show how to use LocalAI with k8sgpt
+
+![Screenshot from 2023-06-19 23-58-47](https://github.com/go-skynet/go-ggml-transformers.cpp/assets/2420543/cab87409-ee68-44ae-8d53-41627fb49509)
+
+## Create the cluster locally with Kind (optional)
+
+If you want to test this locally without a remote Kubernetes cluster, you can use kind.
+
+Install [kind](https://kind.sigs.k8s.io/) and create a cluster:
+
+```
+kind create cluster
+```
+
+## Setup LocalAI
+
+We will use [helm](https://helm.sh/docs/intro/install/):
+
+```
+helm repo add go-skynet https://go-skynet.github.io/helm-charts/
+helm repo update
+
+# Clone LocalAI
+git clone https://github.com/go-skynet/LocalAI
+
+cd LocalAI/examples/k8sgpt
+
+# modify values.yaml preload_models with the models you want to install.
+# CHANGE the URL to a model in huggingface.
+helm install local-ai go-skynet/local-ai --create-namespace --namespace local-ai --values values.yaml
+```
+
+## Setup K8sGPT
+
+```
+# Install k8sgpt
+helm repo add k8sgpt https://charts.k8sgpt.ai/
+helm repo update
+helm install release k8sgpt/k8sgpt-operator -n k8sgpt-operator-system --create-namespace
+```
+
+Apply the k8sgpt-operator configuration:
+
+```
+kubectl apply -f - << EOF
+apiVersion: core.k8sgpt.ai/v1alpha1
+kind: K8sGPT
+metadata:
+  name: k8sgpt-local-ai
+  namespace: default
+spec:
+  backend: localai
+  baseUrl: http://local-ai.local-ai.svc.cluster.local:8080/v1
+  noCache: false
+  model: gpt-3.5-turbo
+  noCache: false
+  version: v0.3.0
+  enableAI: true
+EOF
+```
+
+## Test
+
+Apply a broken pod:
+
+```
+kubectl apply -f broken-pod.yaml
+```

examples/k8sgpt/broken-pod.yaml

@@ -0,0 +1,14 @@
+apiVersion: v1
+kind: Pod
+metadata:
+  name: broken-pod
+spec:
+  containers:
+    - name: broken-pod
+      image: nginx:1.a.b.c
+      livenessProbe:
+        httpGet:
+          path: /
+          port: 90
+        initialDelaySeconds: 3
+        periodSeconds: 3

examples/k8sgpt/values.yaml

@@ -0,0 +1,95 @@
+replicaCount: 1
+
+deployment:
+  # https://quay.io/repository/go-skynet/local-ai?tab=tags
+  image: quay.io/go-skynet/local-ai:latest
+  env:
+    threads: 4
+    debug: "true"
+    context_size: 512
+    preload_models: '[{ "url": "github:go-skynet/model-gallery/wizard.yaml", "name": "gpt-3.5-turbo", "overrides": { "parameters": { "model": "WizardLM-7B-uncensored.ggmlv3.q5_1" }},"files": [ { "uri": "https://huggingface.co//WizardLM-7B-uncensored-GGML/resolve/main/WizardLM-7B-uncensored.ggmlv3.q5_1.bin", "sha256": "d92a509d83a8ea5e08ba4c2dbaf08f29015932dc2accd627ce0665ac72c2bb2b", "filename": "WizardLM-7B-uncensored.ggmlv3.q5_1" }]}]'
+  modelsPath: "/models"
+
+resources:
+  {}
+  # We usually recommend not to specify default resources and to leave this as a conscious
+  # choice for the user. This also increases chances charts run on environments with little
+  # resources, such as Minikube. If you do want to specify resources, uncomment the following
+  # lines, adjust them as necessary, and remove the curly braces after 'resources:'.
+  # limits:
+  #   cpu: 100m
+  #   memory: 128Mi
+  # requests:
+  #   cpu: 100m
+  #   memory: 128Mi
+
+# Prompt templates to include
+# Note: the keys of this map will be the names of the prompt template files
+promptTemplates:
+  {}
+  # ggml-gpt4all-j.tmpl: |
+  #   The prompt below is a question to answer, a task to complete, or a conversation to respond to; decide which and write an appropriate response.
+  #   ### Prompt:
+  #   {{.Input}}
+  #   ### Response:
+
+# Models to download at runtime
+models:
+  # Whether to force download models even if they already exist
+  forceDownload: false
+
+  # The list of URLs to download models from
+  # Note: the name of the file will be the name of the loaded model
+  list:
+  #- url: "https://gpt4all.io/models/ggml-gpt4all-j.bin"
+      # basicAuth: base64EncodedCredentials
+
+  # Persistent storage for models and prompt templates.
+  # PVC and HostPath are mutually exclusive. If both are enabled,
+  # PVC configuration takes precedence. If neither are enabled, ephemeral
+  # storage is used.
+  persistence:
+    pvc:
+      enabled: false
+      size: 6Gi
+      accessModes:
+        - ReadWriteOnce
+
+      annotations: {}
+
+      # Optional
+      storageClass: ~
+
+    hostPath:
+      enabled: false
+      path: "/models"
+
+service:
+  type: ClusterIP
+  port: 8080
+  annotations: {}
+  # If using an AWS load balancer, you'll need to override the default 60s load balancer idle timeout
+  # service.beta.kubernetes.io/aws-load-balancer-connection-idle-timeout: "1200"
+
+ingress:
+  enabled: false
+  className: ""
+  annotations:
+    {}
+    # kubernetes.io/ingress.class: nginx
+    # kubernetes.io/tls-acme: "true"
+  hosts:
+    - host: chart-example.local
+      paths:
+        - path: /
+          pathType: ImplementationSpecific
+  tls: []
+  #  - secretName: chart-example-tls
+  #    hosts:
+  #      - chart-example.local
+
+nodeSelector: {}
+
+tolerations: []
+
+affinity: {}

examples/langchain-huggingface/README.md

@@ -0,0 +1,68 @@
+# Data query example
+
+Example of integration with HuggingFace Inference API with help of [langchaingo](https://github.com/tmc/langchaingo).
+
+## Setup
+
+Download the LocalAI and start the API:
+
+```bash
+# Clone LocalAI
+git clone https://github.com/go-skynet/LocalAI
+
+cd LocalAI/examples/langchain-huggingface
+
+docker-compose up -d
+```
+
+Node: Ensure you've set `HUGGINGFACEHUB_API_TOKEN` environment variable, you can generate it
+on [Settings / Access Tokens](https://huggingface.co/settings/tokens) page of HuggingFace site.
+
+This is an example `.env` file for LocalAI:
+
+```ini
+MODELS_PATH=/models
+CONTEXT_SIZE=512
+HUGGINGFACEHUB_API_TOKEN=hg_123456
+```
+
+## Using remote models
+
+Now you can use any remote models available via HuggingFace API, for example let's enable using of
+[gpt2](https://huggingface.co/gpt2) model in `gpt-3.5-turbo.yaml` config:
+
+```yml
+name: gpt-3.5-turbo
+parameters:
+  model: gpt2
+  top_k: 80
+  temperature: 0.2
+  top_p: 0.7
+context_size: 1024
+backend: "langchain-huggingface"
+stopwords:
+- "HUMAN:"
+- "GPT:"
+roles:
+  user: " "
+  system: " "
+template:
+  completion: completion
+  chat: gpt4all
+```
+
+Here is you can see in field `parameters.model` equal `gpt2` and `backend` equal `langchain-huggingface`.
+
+## How to use
+
+```shell
+# Now API is accessible at localhost:8080
+curl http://localhost:8080/v1/models
+# {"object":"list","data":[{"id":"gpt-3.5-turbo","object":"model"}]}
+
+curl http://localhost:8080/v1/completions -H "Content-Type: application/json" -d '{
+  "model": "gpt-3.5-turbo",
+  "prompt": "A long time ago in a galaxy far, far away",
+  "temperature": 0.7
+}'
+```

examples/langchain-huggingface/docker-compose.yml

@@ -0,0 +1,15 @@
+version: '3.6'
+
+services:
+  api:
+    image: quay.io/go-skynet/local-ai:latest
+    build:
+      context: ../../
+      dockerfile: Dockerfile
+    ports:
+      - 8080:8080
+    env_file:
+      - ../../.env
+    volumes:
+      - ./models:/models:cached
+    command: ["/usr/bin/local-ai"]

examples/langchain-huggingface/models/completion.tmpl

@@ -0,0 +1 @@
+{{.Input}}

examples/langchain-huggingface/models/gpt-3.5-turbo.yaml

@@ -0,0 +1,17 @@
+name: gpt-3.5-turbo
+parameters:
+  model: gpt2
+  top_k: 80
+  temperature: 0.2
+  top_p: 0.7
+context_size: 1024
+backend: "langchain-huggingface"
+stopwords:
+- "HUMAN:"
+- "GPT:"
+roles:
+  user: " "
+  system: " "
+template:
+  completion: completion
+  chat: gpt4all

examples/langchain-huggingface/models/gpt4all.tmpl

@@ -0,0 +1,4 @@
+The prompt below is a question to answer, a task to complete, or a conversation to respond to; decide which and write an appropriate response.
+### Prompt:
+{{.Input}}
+### Response:

examples/langchain-python/README.md

@@ -12,15 +12,8 @@ git clone https://github.com/go-skynet/LocalAI
 
 cd LocalAI/examples/langchain-python
 
-# (optional) Checkout a specific LocalAI tag
-# git checkout -b build <TAG>
-
-# Download gpt4all-j to models/
-wget https://gpt4all.io/models/ggml-gpt4all-j.bin -O models/ggml-gpt4all-j
-
 # start with docker-compose
-docker-compose up -d --build
-
+docker-compose up --pull always
 
 pip install langchain
 pip install openai

examples/langchain-python/docker-compose.yaml

@@ -3,14 +3,25 @@ version: '3.6'
 services:
   api:
     image: quay.io/go-skynet/local-ai:latest
+    # As initially LocalAI will download the models defined in PRELOAD_MODELS
+    # you might need to tweak the healthcheck values here according to your network connection.
+    # Here we give a timespan of 20m to download all the required files.
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8080/readyz"]
+      interval: 1m
+      timeout: 20m
+      retries: 20
     build:
       context: ../../
-      dockerfile: Dockerfile.dev
+      dockerfile: Dockerfile
     ports:
       - 8080:8080
     environment:
       - DEBUG=true
       - MODELS_PATH=/models
+      # You can preload different models here as well.
+      # See: https://github.com/go-skynet/model-gallery
+      - 'PRELOAD_MODELS=[{"url": "github:go-skynet/model-gallery/gpt4all-j.yaml", "name": "gpt-3.5-turbo"}]'
     volumes:
       - ./models:/models:cached
     command: ["/usr/bin/local-ai" ]