feat: update go-gpt2 (#359)

Signed-off-by: mudler <mudler@mocaccino.org>

.dockerignore

@@ -1,2 +1,4 @@
 models
-examples/chatbot-ui/models
+examples/chatbot-ui/models
+examples/rwkv/models
+examples/**/models

.env

@@ -1,5 +1,30 @@
+## Set number of threads.
+## Note: prefer the number of physical cores. Overbooking the CPU degrades performance notably.
 # THREADS=14
+
+## Specify a different bind address (defaults to ":8080")
+# ADDRESS=127.0.0.1:8080
+
+## Default models context size
 # CONTEXT_SIZE=512
+
+## Default path for models
 MODELS_PATH=/models
+
+## Enable debug mode
 # DEBUG=true
-# BUILD_TYPE=generic
+
+## Specify a build type. Available: cublas, openblas.
+# BUILD_TYPE=openblas
+
+## Uncomment and set to false to disable rebuilding from source
+# REBUILD=false
+
+## Enable image generation with stablediffusion (requires REBUILD=true)
+# GO_TAGS=stablediffusion
+
+## Path where to store generated images
+# IMAGE_PATH=/tmp
+
+## Specify a default upload limit in MB (whisper)
+# UPLOAD_LIMIT

.github/ISSUE_TEMPLATE/bug_report.md

@@ -0,0 +1,31 @@
+---
+name: Bug report
+about: Create a report to help us improve
+title: ''
+labels: bug
+assignees: mudler
+
+---
+
+<!-- Thanks for helping us to improve LocalAI! We welcome all bug reports. Please fill out each area of the template so we can better help you. Comments like this will be hidden when you post but you can delete them if you wish. -->
+
+**LocalAI version:**
+<!-- Container Image or LocalAI tag/commit -->
+
+**Environment, CPU architecture, OS, and Version:**
+<!-- Provide the output from "uname -a", HW specs, if it's a VM  -->
+
+**Describe the bug**
+<!-- A clear and concise description of what the bug is. -->
+
+**To Reproduce**
+<!-- Steps to reproduce the behavior, including the LocalAI command used, if any -->
+
+**Expected behavior**
+<!-- A clear and concise description of what you expected to happen. -->
+
+**Logs**
+<!-- If applicable, add logs while running LocalAI in debug mode (`--debug` or `DEBUG=true`) to help explain your problem.  -->
+
+**Additional context**
+<!-- Add any other context about the problem here. -->

.github/ISSUE_TEMPLATE/config.yml

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

.github/ISSUE_TEMPLATE/feature_request.md

@@ -0,0 +1,22 @@
+---
+name: Feature request
+about: Suggest an idea for this project
+title: ''
+labels: enhancement
+assignees: mudler
+
+---
+
+<!-- Thanks for helping us to improve LocalAI! We welcome all feature requests. Please fill out each area of the template so we can better help you. Comments like this will be hidden when you post but you can delete them if you wish. -->
+
+**Is your feature request related to a problem? Please describe.**
+<!-- A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]  -->
+
+**Describe the solution you'd like**
+<!-- A clear and concise description of what you want to happen.  -->
+
+**Describe alternatives you've considered**
+<!-- A clear and concise description of any alternative solutions or features you've considered.  -->
+
+**Additional context**
+<!-- Add any other context or screenshots about the feature request here. -->

.github/PULL_REQUEST_TEMPLATE.md

@@ -0,0 +1,23 @@
+**Description**
+
+This PR fixes #
+
+**Notes for Reviewers**
+
+
+**[Signed commits](../CONTRIBUTING.md#signing-off-on-commits-developer-certificate-of-origin)**
+- [ ] Yes, I signed my commits.
+ 
+
+<!--
+Thank you for contributing to LocalAI! 
+
+Contributing Conventions:
+
+1. Include descriptive PR titles with [<component-name>] prepended.
+2. Build and test your changes before submitting a PR. 
+3. Sign your commits
+
+By following the community's contribution conventions upfront, the review process will 
+be accelerated and your PR merged more quickly.
+-->

.github/bump_deps.sh

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

.github/release.yml

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

.github/stale.yml

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

.github/workflows/bump_deps.yaml

@@ -0,0 +1,51 @@
+name: Bump dependencies
+on:
+  schedule:
+    - cron: 0 20 * * *
+  workflow_dispatch:
+jobs:
+  bump:
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - repository: "go-skynet/go-llama.cpp"
+            variable: "GOLLAMA_VERSION"
+            branch: "master"
+          - repository: "go-skynet/go-ggml-transformers.cpp"
+            variable: "GOGGMLTRANSFORMERS_VERSION"
+            branch: "master"
+          - repository: "donomii/go-rwkv.cpp"
+            variable: "RWKV_VERSION"
+            branch: "main"
+          - repository: "ggerganov/whisper.cpp"
+            variable: "WHISPER_CPP_VERSION"
+            branch: "master"
+          - repository: "go-skynet/go-bert.cpp"
+            variable: "BERT_VERSION"
+            branch: "master"
+          - repository: "go-skynet/bloomz.cpp"
+            variable: "BLOOMZ_VERSION"
+            branch: "main"
+          - repository: "nomic-ai/gpt4all"
+            variable: "GPT4ALL_VERSION"
+            branch: "main"
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v3
+      - name: Bump dependencies 🔧
+        run: |
+          bash .github/bump_deps.sh ${{ matrix.repository }} ${{ matrix.branch }} ${{ matrix.variable }}
+      - name: Create Pull Request
+        uses: peter-evans/create-pull-request@v5
+        with:
+          token: ${{ secrets.UPDATE_BOT_TOKEN }}
+          push-to-fork: ci-forks/LocalAI
+          commit-message: ':arrow_up: Update ${{ matrix.repository }}'
+          title: ':arrow_up: Update ${{ matrix.repository }}'
+          branch: "update/${{ matrix.variable }}"
+          body: Bump of ${{ matrix.repository }} version
+          signoff: true
+
+
+

.github/workflows/image.yml

@@ -9,6 +9,10 @@ on:
     tags:
       - '*'
 
+concurrency:
+  group: ci-${{ github.head_ref || github.ref }}-${{ github.repository }}
+  cancel-in-progress: true
+
 jobs:
   docker:
     runs-on: ubuntu-latest
@@ -54,8 +58,8 @@ jobs:
         uses: docker/login-action@v2
         with:
           registry: quay.io
-          username: ${{ secrets.QUAY_USERNAME }}
-          password: ${{ secrets.QUAY_PASSWORD }}
+          username: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
+          password: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
       - name: Build
         if: github.event_name != 'pull_request'
         uses: docker/build-push-action@v4

.github/workflows/release.yaml

@@ -0,0 +1,84 @@
+name: Build and Release
+
+on: push
+
+permissions:
+  contents: write
+
+jobs:
+  build-linux:
+    strategy:
+      matrix:
+        include:
+          - build: 'avx2'
+            defines: ''
+          - build: 'avx'
+            defines: '-DLLAMA_AVX2=OFF'
+          - build: 'avx512'
+            defines: '-DLLAMA_AVX512=ON'
+    runs-on: ubuntu-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v3
+        with: 
+          submodules: true
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install build-essential ffmpeg
+      - name: Build
+        id: build
+        env:
+          CMAKE_ARGS: "${{ matrix.define }}"
+          BUILD_ID: "${{ matrix.build }}"
+        run: |
+          make dist
+      - uses: actions/upload-artifact@v3
+        with:
+          name: ${{ matrix.build }}
+          path: release/
+      - name: Release
+        uses: softprops/action-gh-release@v1
+        if: startsWith(github.ref, 'refs/tags/')
+        with:
+          files: |
+            release/*
+
+  build-macOS:
+    strategy:
+      matrix:
+        include:
+          - build: 'avx2'
+            defines: ''
+          - build: 'avx'
+            defines: '-DLLAMA_AVX2=OFF'
+          - build: 'avx512'
+            defines: '-DLLAMA_AVX512=ON'
+    runs-on: macOS-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v3
+        with: 
+          submodules: true
+
+      - name: Dependencies
+        run: |
+          brew update
+          brew install sdl2 ffmpeg
+      - name: Build
+        id: build
+        env:
+          CMAKE_ARGS: "${{ matrix.define }}"
+          BUILD_ID: "${{ matrix.build }}"
+        run: |
+          make dist
+      - uses: actions/upload-artifact@v3
+        with:
+          name: ${{ matrix.build }}
+          path: release/
+      - name: Release
+        uses: softprops/action-gh-release@v1
+        if: startsWith(github.ref, 'refs/tags/')
+        with:
+          files: |
+            release/*

.github/workflows/release.yml.disabled

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

.github/workflows/test.yml

@@ -9,6 +9,10 @@ on:
     tags:
       - '*'
 
+concurrency:
+  group: ci-tests-${{ github.head_ref || github.ref }}-${{ github.repository }}
+  cancel-in-progress: true
+
 jobs:
   ubuntu-latest:
     runs-on: ubuntu-latest
@@ -21,7 +25,7 @@ jobs:
       - name: Dependencies
         run: |
           sudo apt-get update
-          sudo apt-get install build-essential
+          sudo apt-get install build-essential ffmpeg
       - name: Test
         run: |
           make test
@@ -38,7 +42,7 @@ jobs:
       - name: Dependencies
         run: |
           brew update
-          brew install sdl2
+          brew install sdl2 ffmpeg
       - name: Test
         run: |
           make test

.gitignore

@@ -1,7 +1,10 @@
 # go-llama build artifacts
 go-llama
-go-gpt4all-j
+gpt4all
+go-stable-diffusion
 go-gpt2
+go-rwkv
+whisper.cpp
 
 # LocalAI build binary
 LocalAI
@@ -11,4 +14,9 @@ local-ai
 
 # Ignore models
 models/*
-test-models/
+test-models/
+
+release/
+
+# just in case
+.DS_Store

.goreleaser.yaml

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

.vscode/launch.json

@@ -2,7 +2,20 @@
     "version": "0.2.0",
     "configurations": [
         {
-            "name": "Launch Go",
+            "name": "Python: Current File",
+            "type": "python",
+            "request": "launch",
+            "program": "${file}",
+            "console": "integratedTerminal",
+            "justMyCode": false,
+            "cwd": "${workspaceFolder}/examples/langchain-chroma",
+            "env": {
+                "OPENAI_API_BASE": "http://localhost:8080/v1",
+                "OPENAI_API_KEY": "abc"
+            }
+        },
+        {
+            "name": "Launch LocalAI API",
             "type": "go",
             "request": "launch",
             "mode": "debug",
@@ -11,8 +24,8 @@
                 "api"
             ],
             "env": {
-                "C_INCLUDE_PATH": "/workspace/go-llama:/workspace/go-gpt4all-j:/workspace/go-gpt2",
-                "LIBRARY_PATH": "/workspace/go-llama:/workspace/go-gpt4all-j:/workspace/go-gpt2",
+                "C_INCLUDE_PATH": "${workspaceFolder}/go-llama:${workspaceFolder}/go-stable-diffusion/:${workspaceFolder}/gpt4all/gpt4all-bindings/golang/:${workspaceFolder}/go-gpt2:${workspaceFolder}/go-rwkv:${workspaceFolder}/whisper.cpp:${workspaceFolder}/go-bert:${workspaceFolder}/bloomz",
+                "LIBRARY_PATH": "$${workspaceFolder}/go-llama:${workspaceFolder}/go-stable-diffusion/:${workspaceFolder}/gpt4all/gpt4all-bindings/golang/:${workspaceFolder}/go-gpt2:${workspaceFolder}/go-rwkv:${workspaceFolder}/whisper.cpp:${workspaceFolder}/go-bert:${workspaceFolder}/bloomz",
                 "DEBUG": "true"
             }
         }

Dockerfile

@@ -1,13 +1,11 @@
 ARG GO_VERSION=1.20
-ARG DEBIAN_VERSION=11
 ARG BUILD_TYPE=
-
-FROM golang:$GO_VERSION as builder
+FROM golang:$GO_VERSION
+ENV REBUILD=true
 WORKDIR /build
-RUN apt-get update && apt-get install -y cmake
+RUN apt-get update && apt-get install -y cmake libgomp1 libopenblas-dev libopenblas-base libopencv-dev libopencv-core-dev libopencv-core4.5 ca-certificates
 COPY . .
+RUN ln -s /usr/include/opencv4/opencv2/ /usr/include/opencv2
 RUN make build
-
-FROM debian:$DEBIAN_VERSION
-COPY --from=builder /build/local-ai /usr/bin/local-ai
-ENTRYPOINT [ "/usr/bin/local-ai" ]
+EXPOSE 8080
+ENTRYPOINT [ "/build/entrypoint.sh" ]

Dockerfile.dev

@@ -0,0 +1,16 @@
+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
+EXPOSE 8080
+ENTRYPOINT [ "/usr/bin/local-ai" ]

Makefile

@@ -2,111 +2,228 @@ GOCMD=go
 GOTEST=$(GOCMD) test
 GOVET=$(GOCMD) vet
 BINARY_NAME=local-ai
-# renovate: datasource=github-tags depName=go-skynet/go-llama.cpp
-GOLLAMA_VERSION?=llama.cpp-25d7abb
-# renovate: datasource=git-refs packageNameTemplate=https://github.com/go-skynet/go-gpt4all-j.cpp currentValueTemplate=master depNameTemplate=go-gpt4all-j.cpp
-GOGPT4ALLJ_VERSION?=1f7bff57f66cb7062e40d0ac3abd2217815e5109
-# renovate: datasource=git-refs packageNameTemplate=https://github.com/go-skynet/go-gpt2.cpp currentValueTemplate=master depNameTemplate=go-gpt2.cpp
-GOGPT2_VERSION?=245a5bfe6708ab80dc5c733dcdbfbe3cfd2acdaa
 
+GOLLAMA_VERSION?=ccf23adfb278c0165d388389a5d60f3fe38e4854
+GPT4ALL_REPO?=https://github.com/nomic-ai/gpt4all
+GPT4ALL_VERSION?=8119ff4df0a99bde44255db2b8c7290b5582ac2b
+GOGGMLTRANSFORMERS_VERSION?=14fd6c9
+RWKV_REPO?=https://github.com/donomii/go-rwkv.cpp
+RWKV_VERSION?=07166da10cb2a9e8854395a4f210464dcea76e47
+WHISPER_CPP_VERSION?=041be06d5881d3c759cc4ed45d655804361237cd
+BERT_VERSION?=cea1ed76a7f48ef386a8e369f6c82c48cdf2d551
+BLOOMZ_VERSION?=e9366e82abdfe70565644fbfae9651976714efd1
+BUILD_TYPE?=
+CGO_LDFLAGS?=
+CUDA_LIBPATH?=/usr/local/cuda/lib64/
+STABLEDIFFUSION_VERSION?=c0748eca3642d58bcf9521108bcee46959c647dc
+GO_TAGS?=
+BUILD_ID?=git
+
+OPTIONAL_TARGETS?=
+
+OS := $(shell uname -s)
+ARCH := $(shell uname -m)
 GREEN  := $(shell tput -Txterm setaf 2)
 YELLOW := $(shell tput -Txterm setaf 3)
 WHITE  := $(shell tput -Txterm setaf 7)
 CYAN   := $(shell tput -Txterm setaf 6)
 RESET  := $(shell tput -Txterm sgr0)
 
-C_INCLUDE_PATH=$(shell pwd)/go-llama:$(shell pwd)/go-gpt4all-j:$(shell pwd)/go-gpt2
-LIBRARY_PATH=$(shell pwd)/go-llama:$(shell pwd)/go-gpt4all-j:$(shell pwd)/go-gpt2
+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
 
-# Use this if you want to set the default behavior
-ifndef BUILD_TYPE
-	BUILD_TYPE:=default
+ifeq ($(BUILD_TYPE),openblas)
+	CGO_LDFLAGS+=-lopenblas
 endif
 
-ifeq ($(BUILD_TYPE), "generic")
-	GENERIC_PREFIX:=generic-
-else
-	GENERIC_PREFIX:=
+ifeq ($(BUILD_TYPE),cublas)
+	CGO_LDFLAGS+=-lcublas -lcudart -L$(CUDA_LIBPATH)
+	export LLAMA_CUBLAS=1
+endif
+
+ifeq ($(GO_TAGS),stablediffusion)
+	OPTIONAL_TARGETS+=go-stable-diffusion/libstablediffusion.a
 endif
 
 .PHONY: all test build vendor
 
 all: help
 
+## GPT4ALL
+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
+
+## 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:
+	git clone --recurse-submodules https://github.com/mudler/go-stable-diffusion go-stable-diffusion
+	cd go-stable-diffusion && git checkout -b build $(STABLEDIFFUSION_VERSION) && git submodule update --init --recursive --depth 1
+
+go-stable-diffusion/libstablediffusion.a:
+	$(MAKE) -C go-stable-diffusion libstablediffusion.a
+
+## RWKV
+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 ..
+
+## 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' {} +
+
+bloomz/libbloomz.a: bloomz
+	cd bloomz && make libbloomz.a
+
+go-bert/libgobert.a: go-bert
+	$(MAKE) -C go-bert libgobert.a
+
+gpt4all/gpt4all-bindings/golang/libgpt4all.a: gpt4all
+	$(MAKE) -C gpt4all/gpt4all-bindings/golang/ libgpt4all.a
+
+## CEREBRAS GPT
+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
+
+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
+
+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/libbinding.a: go-llama 
+	$(MAKE) -C go-llama BUILD_TYPE=$(BUILD_TYPE) libbinding.a
+
+replace:
+	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(shell pwd)/go-llama
+	$(GOCMD) mod edit -replace github.com/nomic-ai/gpt4all/gpt4all-bindings/golang=$(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
+	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp=$(shell pwd)/whisper.cpp
+	$(GOCMD) mod edit -replace github.com/go-skynet/go-bert.cpp=$(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
+
+prepare-sources: go-llama go-ggml-transformers gpt4all go-rwkv whisper.cpp go-bert bloomz go-stable-diffusion replace
+	$(GOCMD) mod download
+
+## GENERIC
+rebuild: ## Rebuilds the project
+	$(MAKE) -C go-llama clean
+	$(MAKE) -C gpt4all/gpt4all-bindings/golang/ clean
+	$(MAKE) -C go-ggml-transformers clean
+	$(MAKE) -C go-rwkv clean
+	$(MAKE) -C whisper.cpp clean
+	$(MAKE) -C go-stable-diffusion clean
+	$(MAKE) -C go-bert clean
+	$(MAKE) -C bloomz 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
+
+clean: ## Remove build related file
+	rm -fr ./go-llama
+	rm -rf ./gpt4all
+	rm -rf ./go-stable-diffusion
+	rm -rf ./go-ggml-transformers
+	rm -rf ./go-rwkv
+	rm -rf ./go-bert
+	rm -rf ./bloomz
+	rm -rf ./whisper.cpp
+	rm -rf $(BINARY_NAME)
+	rm -rf release/
+
 ## Build:
 
 build: prepare ## Build the project
 	$(info ${GREEN}I local-ai build info:${RESET})
 	$(info ${GREEN}I BUILD_TYPE: ${YELLOW}$(BUILD_TYPE)${RESET})
-	C_INCLUDE_PATH=${C_INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} $(GOCMD) build -o $(BINARY_NAME) ./
+	$(info ${GREEN}I GO_TAGS: ${YELLOW}$(GO_TAGS)${RESET})
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=${C_INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} $(GOCMD) build -tags "$(GO_TAGS)" -x -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
 
-## GPT4ALL-J
-go-gpt4all-j:
-	git clone --recurse-submodules https://github.com/go-skynet/go-gpt4all-j.cpp go-gpt4all-j
-	cd go-gpt4all-j && git checkout -b build $(GOGPT4ALLJ_VERSION)
-	# This is hackish, but needed as both go-llama and go-gpt4allj have their own version of ggml..
-	@find ./go-gpt4all-j -type f -name "*.c" -exec sed -i'' -e 's/ggml_/ggml_gptj_/g' {} +
-	@find ./go-gpt4all-j -type f -name "*.cpp" -exec sed -i'' -e 's/ggml_/ggml_gptj_/g' {} +
-	@find ./go-gpt4all-j -type f -name "*.h" -exec sed -i'' -e 's/ggml_/ggml_gptj_/g' {} +
-	@find ./go-gpt4all-j -type f -name "*.cpp" -exec sed -i'' -e 's/gpt_/gptj_/g' {} +
-	@find ./go-gpt4all-j -type f -name "*.h" -exec sed -i'' -e 's/gpt_/gptj_/g' {} +
-	@find ./go-gpt4all-j -type f -name "*.cpp" -exec sed -i'' -e 's/json_/json_gptj_/g' {} +
-	@find ./go-gpt4all-j -type f -name "*.cpp" -exec sed -i'' -e 's/void replace/void json_gptj_replace/g' {} +
-	@find ./go-gpt4all-j -type f -name "*.cpp" -exec sed -i'' -e 's/::replace/::json_gptj_replace/g' {} +
-
-go-gpt4all-j/libgptj.a: go-gpt4all-j
-	$(MAKE) -C go-gpt4all-j $(GENERIC_PREFIX)libgptj.a
-
-# CEREBRAS GPT
-go-gpt2:
-	git clone --recurse-submodules https://github.com/go-skynet/go-gpt2.cpp go-gpt2
-	cd go-gpt2 && git checkout -b build $(GOGPT2_VERSION)
-	# This is hackish, but needed as both go-llama and go-gpt4allj have their own version of ggml..
-	@find ./go-gpt2 -type f -name "*.c" -exec sed -i'' -e 's/ggml_/ggml_gpt2_/g' {} +
-	@find ./go-gpt2 -type f -name "*.cpp" -exec sed -i'' -e 's/ggml_/ggml_gpt2_/g' {} +
-	@find ./go-gpt2 -type f -name "*.h" -exec sed -i'' -e 's/ggml_/ggml_gpt2_/g' {} +
-	@find ./go-gpt2 -type f -name "*.cpp" -exec sed -i'' -e 's/gpt_/gpt2_/g' {} +
-	@find ./go-gpt2 -type f -name "*.h" -exec sed -i'' -e 's/gpt_/gpt2_/g' {} +
-	@find ./go-gpt2 -type f -name "*.cpp" -exec sed -i'' -e 's/json_/json_gpt2_/g' {} +
-
-go-gpt2/libgpt2.a: go-gpt2
-	$(MAKE) -C go-gpt2 $(GENERIC_PREFIX)libgpt2.a
-	
-
-go-llama:
-	git clone -b $(GOLLAMA_VERSION) --recurse-submodules https://github.com/go-skynet/go-llama.cpp go-llama
-
-go-llama/libbinding.a: go-llama
-	$(MAKE) -C go-llama $(GENERIC_PREFIX)libbinding.a
-
-replace:
-	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(shell pwd)/go-llama
-	$(GOCMD) mod edit -replace github.com/go-skynet/go-gpt4all-j.cpp=$(shell pwd)/go-gpt4all-j
-	$(GOCMD) mod edit -replace github.com/go-skynet/go-gpt2.cpp=$(shell pwd)/go-gpt2
-
-prepare: go-llama/libbinding.a go-gpt4all-j/libgptj.a go-gpt2/libgpt2.a replace
-
-clean: ## Remove build related file
-	rm -fr ./go-llama
-	rm -rf ./go-gpt4all-j
-	rm -rf ./go-gpt2
-	rm -rf $(BINARY_NAME)
-
-## Run:
-run: prepare
-	C_INCLUDE_PATH=${C_INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} $(GOCMD) run ./main.go
+## Run
+run: prepare ## run local-ai
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=${C_INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} $(GOCMD) run ./main.go
 
 test-models/testmodel:
 	mkdir test-models
-	wget https://huggingface.co/concedo/cerebras-111M-ggml/resolve/main/cerberas-111m-q4_0.bin -O test-models/testmodel
-	cp tests/fixtures/* test-models
+	mkdir test-dir
+	wget https://huggingface.co/nnakasato/ggml-model-test/resolve/main/ggml-model-q4.bin -O 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://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
-	cp tests/fixtures/* test-models
-	@C_INCLUDE_PATH=${C_INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models $(GOCMD) test -v -timeout 20m ./...
+	cp tests/models_fixtures/* test-models
+	C_INCLUDE_PATH=${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 -v -r ./api ./pkg
 
 ## Help:
 help: ## Show this help.

api/api.go

@@ -3,23 +3,26 @@ package api
 import (
 	"errors"
 
-	model "github.com/go-skynet/LocalAI/pkg/model"
 	"github.com/gofiber/fiber/v2"
 	"github.com/gofiber/fiber/v2/middleware/cors"
+	"github.com/gofiber/fiber/v2/middleware/logger"
 	"github.com/gofiber/fiber/v2/middleware/recover"
 	"github.com/rs/zerolog"
 	"github.com/rs/zerolog/log"
 )
 
-func App(configFile string, loader *model.ModelLoader, threads, ctxSize int, f16 bool, debug, disableMessage bool) *fiber.App {
+func App(opts ...AppOption) *fiber.App {
+	options := newOptions(opts...)
+
 	zerolog.SetGlobalLevel(zerolog.InfoLevel)
-	if debug {
+	if options.debug {
 		zerolog.SetGlobalLevel(zerolog.DebugLevel)
 	}
 
 	// Return errors as JSON responses
 	app := fiber.New(fiber.Config{
-		DisableStartupMessage: 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
@@ -40,35 +43,80 @@ func App(configFile string, loader *model.ModelLoader, threads, ctxSize int, f16
 		},
 	})
 
-	cm := make(ConfigMerger)
-	if err := cm.LoadConfigs(loader.ModelPath); err != nil {
+	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.Error().Msgf("error loading config files: %s", err.Error())
 	}
 
-	if configFile != "" {
-		if err := cm.LoadConfigFile(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 debug {
-		for k, v := range cm {
-			log.Debug().Msgf("Model: %s (config: %+v)", k, v)
+	if options.debug {
+		for _, v := range cm.ListConfigs() {
+			cfg, _ := cm.GetConfig(v)
+			log.Debug().Msgf("Model: %s (config: %+v)", v, cfg)
 		}
 	}
 	// Default middleware config
 	app.Use(recover.New())
-	app.Use(cors.New())
+
+	if options.cors {
+		if options.corsAllowOrigins == "" {
+			app.Use(cors.New())
+		} else {
+			app.Use(cors.New(cors.Config{
+				AllowOrigins: options.corsAllowOrigins,
+			}))
+		}
+	}
+
+	// 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))
 
 	// openAI compatible API endpoint
-	app.Post("/v1/chat/completions", openAIEndpoint(cm, true, debug, loader, threads, ctxSize, f16))
-	app.Post("/chat/completions", openAIEndpoint(cm, true, debug, loader, threads, ctxSize, f16))
 
-	app.Post("/v1/completions", openAIEndpoint(cm, false, debug, loader, threads, ctxSize, f16))
-	app.Post("/completions", openAIEndpoint(cm, false, debug, loader, threads, ctxSize, f16))
+	// chat
+	app.Post("/v1/chat/completions", chatEndpoint(cm, options))
+	app.Post("/chat/completions", chatEndpoint(cm, options))
 
-	app.Get("/v1/models", listModels(loader, cm))
-	app.Get("/models", listModels(loader, cm))
+	// edit
+	app.Post("/v1/edits", editEndpoint(cm, options))
+	app.Post("/edits", editEndpoint(cm, options))
+
+	// completion
+	app.Post("/v1/completions", completionEndpoint(cm, options))
+	app.Post("/completions", 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))
+
+	// audio
+	app.Post("/v1/audio/transcriptions", transcriptEndpoint(cm, options))
+
+	// images
+	app.Post("/v1/images/generations", imageEndpoint(cm, options))
+
+	if options.imageDir != "" {
+		app.Static("/generated-images", options.imageDir)
+	}
+
+	// models
+	app.Get("/v1/models", listModels(options.loader, cm))
+	app.Get("/models", listModels(options.loader, cm))
 
 	return app
 }

api/api_test.go

@@ -1,32 +1,212 @@
 package api_test
 
 import (
+	"bytes"
 	"context"
+	"encoding/json"
+	"fmt"
+	"io/ioutil"
+	"net/http"
 	"os"
+	"path/filepath"
+	"runtime"
 
 	. "github.com/go-skynet/LocalAI/api"
 	"github.com/go-skynet/LocalAI/pkg/model"
 	"github.com/gofiber/fiber/v2"
 	. "github.com/onsi/ginkgo/v2"
 	. "github.com/onsi/gomega"
+	"gopkg.in/yaml.v3"
 
+	openaigo "github.com/otiai10/openaigo"
 	"github.com/sashabaranov/go-openai"
 )
 
+type modelApplyRequest struct {
+	URL       string            `json:"url"`
+	Name      string            `json:"name"`
+	Overrides map[string]string `json:"overrides"`
+}
+
+func getModelStatus(url string) (response map[string]interface{}) {
+	// Create the HTTP request
+	resp, err := http.Get(url)
+	if err != nil {
+		fmt.Println("Error creating request:", err)
+		return
+	}
+	defer resp.Body.Close()
+
+	body, err := ioutil.ReadAll(resp.Body)
+	if err != nil {
+		fmt.Println("Error reading response body:", err)
+		return
+	}
+
+	// Unmarshal the response into a map[string]interface{}
+	err = json.Unmarshal(body, &response)
+	if err != nil {
+		fmt.Println("Error unmarshaling JSON response:", err)
+		return
+	}
+	return
+}
+func postModelApplyRequest(url string, request modelApplyRequest) (response map[string]interface{}) {
+
+	//url := "http://localhost:AI/models/apply"
+
+	// Create the request payload
+
+	payload, err := json.Marshal(request)
+	if err != nil {
+		fmt.Println("Error marshaling JSON:", err)
+		return
+	}
+
+	// Create the HTTP request
+	req, err := http.NewRequest("POST", url, bytes.NewBuffer(payload))
+	if err != nil {
+		fmt.Println("Error creating request:", err)
+		return
+	}
+	req.Header.Set("Content-Type", "application/json")
+
+	// Make the request
+	client := &http.Client{}
+	resp, err := client.Do(req)
+	if err != nil {
+		fmt.Println("Error making request:", err)
+		return
+	}
+	defer resp.Body.Close()
+
+	body, err := ioutil.ReadAll(resp.Body)
+	if err != nil {
+		fmt.Println("Error reading response body:", err)
+		return
+	}
+
+	// Unmarshal the response into a map[string]interface{}
+	err = json.Unmarshal(body, &response)
+	if err != nil {
+		fmt.Println("Error unmarshaling JSON response:", err)
+		return
+	}
+	return
+}
+
 var _ = Describe("API test", func() {
 
 	var app *fiber.App
 	var modelLoader *model.ModelLoader
 	var client *openai.Client
-	Context("API query", func() {
+	var client2 *openaigo.Client
+	var c context.Context
+	var cancel context.CancelFunc
+	var tmpdir string
+
+	Context("API with ephemeral models", func() {
 		BeforeEach(func() {
-			modelLoader = model.NewModelLoader(os.Getenv("MODELS_PATH"))
-			app = App("", modelLoader, 1, 512, false, true, true)
+			var err error
+			tmpdir, err = os.MkdirTemp("", "")
+			Expect(err).ToNot(HaveOccurred())
+
+			modelLoader = model.NewModelLoader(tmpdir)
+			c, cancel = context.WithCancel(context.Background())
+
+			app = App(WithContext(c), WithModelLoader(modelLoader))
 			go app.Listen("127.0.0.1:9090")
 
 			defaultConfig := openai.DefaultConfig("")
 			defaultConfig.BaseURL = "http://127.0.0.1:9090/v1"
 
+			client2 = openaigo.NewClient("")
+			client2.BaseURL = defaultConfig.BaseURL
+
+			// Wait for API to be ready
+			client = openai.NewClientWithConfig(defaultConfig)
+			Eventually(func() error {
+				_, err := client.ListModels(context.TODO())
+				return err
+			}, "2m").ShouldNot(HaveOccurred())
+		})
+
+		AfterEach(func() {
+			cancel()
+			app.Shutdown()
+			os.RemoveAll(tmpdir)
+		})
+
+		Context("Applying models", func() {
+			It("overrides models", func() {
+				response := postModelApplyRequest("http://127.0.0.1:9090/models/apply", modelApplyRequest{
+					URL:  "https://raw.githubusercontent.com/go-skynet/model-gallery/main/bert-embeddings.yaml",
+					Name: "bert",
+					Overrides: map[string]string{
+						"backend": "llama",
+					},
+				})
+
+				Expect(response["uuid"]).ToNot(BeEmpty(), fmt.Sprint(response))
+
+				uuid := response["uuid"].(string)
+
+				Eventually(func() bool {
+					response := getModelStatus("http://127.0.0.1:9090/models/jobs/" + uuid)
+					fmt.Println(response)
+					return response["processed"].(bool)
+				}, "360s").Should(Equal(true))
+
+				dat, err := os.ReadFile(filepath.Join(tmpdir, "bert.yaml"))
+				Expect(err).ToNot(HaveOccurred())
+
+				content := map[string]interface{}{}
+				err = yaml.Unmarshal(dat, &content)
+				Expect(err).ToNot(HaveOccurred())
+				Expect(content["backend"]).To(Equal("llama"))
+			})
+			It("apply models without overrides", func() {
+				response := postModelApplyRequest("http://127.0.0.1:9090/models/apply", modelApplyRequest{
+					URL:       "https://raw.githubusercontent.com/go-skynet/model-gallery/main/bert-embeddings.yaml",
+					Name:      "bert",
+					Overrides: map[string]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)
+					fmt.Println(response)
+					return response["processed"].(bool)
+				}, "360s").Should(Equal(true))
+
+				dat, err := os.ReadFile(filepath.Join(tmpdir, "bert.yaml"))
+				Expect(err).ToNot(HaveOccurred())
+
+				content := map[string]interface{}{}
+				err = yaml.Unmarshal(dat, &content)
+				Expect(err).ToNot(HaveOccurred())
+				Expect(content["backend"]).To(Equal("bert-embeddings"))
+			})
+		})
+	})
+
+	Context("API query", func() {
+		BeforeEach(func() {
+			modelLoader = model.NewModelLoader(os.Getenv("MODELS_PATH"))
+			c, cancel = context.WithCancel(context.Background())
+
+			app = App(WithContext(c), WithModelLoader(modelLoader))
+			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 {
@@ -35,13 +215,13 @@ var _ = Describe("API test", func() {
 			}, "2m").ShouldNot(HaveOccurred())
 		})
 		AfterEach(func() {
+			cancel()
 			app.Shutdown()
 		})
 		It("returns the models list", func() {
 			models, err := client.ListModels(context.TODO())
 			Expect(err).ToNot(HaveOccurred())
-			Expect(len(models.Models)).To(Equal(3))
-			Expect(models.Models[0].ID).To(Equal("testmodel"))
+			Expect(len(models.Models)).To(Equal(10))
 		})
 		It("can generate completions", func() {
 			resp, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "testmodel", Prompt: "abcdedfghikl"})
@@ -74,20 +254,75 @@ var _ = Describe("API test", func() {
 		It("returns errors", func() {
 			_, 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: llama: model does not exist"))
+			Expect(err.Error()).To(ContainSubstring("error, status code: 500, message: could not load model - all backends returned error: 12 errors occurred:"))
+		})
+		It("transcribes audio", func() {
+			if runtime.GOOS != "linux" {
+				Skip("test supported only on linux")
+			}
+			resp, err := client.CreateTranscription(
+				context.Background(),
+				openai.AudioRequest{
+					Model:    openai.Whisper1,
+					FilePath: filepath.Join(os.Getenv("TEST_DIR"), "audio.wav"),
+				},
+			)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(resp.Text).To(ContainSubstring("This is the Micro Machine Man presenting"))
 		})
 
+		It("calculate embeddings", func() {
+			if runtime.GOOS != "linux" {
+				Skip("test supported only on linux")
+			}
+			resp, err := client.CreateEmbeddings(
+				context.Background(),
+				openai.EmbeddingRequest{
+					Model: openai.AdaEmbeddingV2,
+					Input: []string{"sun", "cat"},
+				},
+			)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(len(resp.Data[0].Embedding)).To(BeNumerically("==", 384))
+			Expect(len(resp.Data[1].Embedding)).To(BeNumerically("==", 384))
+
+			sunEmbedding := resp.Data[0].Embedding
+			resp2, err := client.CreateEmbeddings(
+				context.Background(),
+				openai.EmbeddingRequest{
+					Model: openai.AdaEmbeddingV2,
+					Input: []string{"sun"},
+				},
+			)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(resp2.Data[0].Embedding).To(Equal(sunEmbedding))
+		})
+
+		Context("backends", func() {
+			It("runs rwkv", 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."))
+			})
+		})
 	})
 
 	Context("Config file", func() {
 		BeforeEach(func() {
 			modelLoader = model.NewModelLoader(os.Getenv("MODELS_PATH"))
-			app = App(os.Getenv("CONFIG_FILE"), modelLoader, 1, 512, false, true, true)
+			c, cancel = context.WithCancel(context.Background())
+
+			app = App(WithContext(c), WithModelLoader(modelLoader), WithConfigFile(os.Getenv("CONFIG_FILE")))
 			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 {
@@ -96,14 +331,13 @@ var _ = Describe("API test", func() {
 			}, "2m").ShouldNot(HaveOccurred())
 		})
 		AfterEach(func() {
+			cancel()
 			app.Shutdown()
 		})
 		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(5))
-			Expect(models.Models[0].ID).To(Equal("testmodel"))
+			Expect(len(models.Models)).To(Equal(12))
 		})
 		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"}}})
@@ -117,5 +351,17 @@ var _ = Describe("API test", func() {
 			Expect(len(resp.Choices)).To(Equal(1))
 			Expect(resp.Choices[0].Message.Content).ToNot(BeEmpty())
 		})
+		It("can generate edit completions from config file", func() {
+			request := openaigo.EditCreateRequestBody{
+				Model:       "list2",
+				Instruction: "foo",
+				Input:       "bar",
+			}
+			resp, err := client2.CreateEdit(context.Background(), request)
+			Expect(err).ToNot(HaveOccurred())
+			Expect(len(resp.Choices)).To(Equal(1))
+			Expect(resp.Choices[0].Text).ToNot(BeEmpty())
+		})
+
 	})
 })

api/config.go

@@ -1,36 +1,59 @@
 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"`
-	TemplateConfig TemplateConfig    `yaml:"template"`
+	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 map[string]Config
+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)
@@ -58,28 +81,51 @@ func ReadConfig(file string) (*Config, error) {
 }
 
 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[cc.Name] = *cc
+		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[c.Name] = *c
+	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
@@ -92,9 +138,192 @@ func (cm ConfigMerger) LoadConfigs(path string) error {
 		}
 		c, err := ReadConfig(filepath.Join(path, file.Name()))
 		if err == nil {
-			cm[c.Name] = *c
+			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/gallery.go

@@ -0,0 +1,196 @@
+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

@@ -0,0 +1,30 @@
+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/openai.go

@@ -2,15 +2,23 @@ package api
 
 import (
 	"bufio"
+	"bytes"
+	"encoding/base64"
 	"encoding/json"
 	"fmt"
+	"io"
+	"io/ioutil"
+	"net/http"
 	"os"
+	"path"
 	"path/filepath"
-	"regexp"
+	"strconv"
 	"strings"
-	"sync"
 
+	"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"
@@ -28,12 +36,31 @@ 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 {
@@ -57,10 +84,21 @@ type OpenAIModel struct {
 type OpenAIRequest struct {
 	Model string `json:"model" yaml:"model"`
 
-	// Prompt is read only by completion API calls
-	Prompt string `json:"prompt" yaml:"prompt"`
+	// 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"`
 
-	Stop string `json:"stop" yaml:"stop"`
+	// 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"`
@@ -82,7 +120,15 @@ type OpenAIRequest struct {
 	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 {
@@ -95,167 +141,192 @@ func defaultRequest(modelFile string) OpenAIRequest {
 	}
 }
 
-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
-	}
+// 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 {
 
-	if input.Temperature != 0 {
-		config.Temperature = input.Temperature
-	}
+		model, input, err := readInput(c, o.loader, true)
+		if err != nil {
+			return fmt.Errorf("failed reading parameters from request:%w", err)
+		}
 
-	if input.Maxtokens != 0 {
-		config.Maxtokens = input.Maxtokens
-	}
+		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)
+		}
 
-	if input.Stop != "" {
-		config.StopWords = append(config.StopWords, input.Stop)
-	}
+		log.Debug().Msgf("Parameter Config: %+v", config)
 
-	if input.RepeatPenalty != 0 {
-		config.RepeatPenalty = input.RepeatPenalty
-	}
+		templateFile := config.Model
 
-	if input.Keep != 0 {
-		config.Keep = input.Keep
-	}
+		if config.TemplateConfig.Completion != "" {
+			templateFile = config.TemplateConfig.Completion
+		}
 
-	if input.Batch != 0 {
-		config.Batch = input.Batch
-	}
+		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)
+			}
 
-	if input.F16 {
-		config.F16 = input.F16
-	}
+			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
+			}
 
-	if input.IgnoreEOS {
-		config.IgnoreEOS = input.IgnoreEOS
-	}
+			result = append(result, r...)
+		}
 
-	if input.Seed != 0 {
-		config.Seed = input.Seed
+		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)
 	}
 }
 
-var cutstrings map[string]*regexp.Regexp = make(map[string]*regexp.Regexp)
-var mu sync.Mutex = sync.Mutex{}
-
-// https://platform.openai.com/docs/api-reference/completions
-func openAIEndpoint(cm ConfigMerger, chat, debug bool, loader *model.ModelLoader, threads, ctx int, f16 bool) func(c *fiber.Ctx) error {
+// 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 {
-
-		input := new(OpenAIRequest)
-		// Get input data from the request body
-		if err := c.BodyParser(input); err != nil {
-			return err
+		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; charset=utf-8")
+			//	c.Set("Content-Type", "text/event-stream")
 			c.Set("Cache-Control", "no-cache")
 			c.Set("Connection", "keep-alive")
 			c.Set("Transfer-Encoding", "chunked")
 		}
 
-		modelFile := input.Model
-		received, _ := json.Marshal(input)
-
-		log.Debug().Msgf("Request received: %s", string(received))
-
-		// Set model from bearer token, if available
-		bearer := strings.TrimLeft(c.Get("authorization"), "Bearer ")
-		bearerExists := bearer != "" && loader.ExistsInModelPath(bearer)
-
-		// If no model was specified, take the first available
-		if modelFile == "" && !bearerExists {
-			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 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
-		}
-
-		// 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 fmt.Errorf("failed loading model config (%s) %s", modelConfig, err.Error())
-			}
-		}
-
-		var config *Config
-		cfg, exists := cm[modelFile]
-		if !exists {
-			config = &Config{
-				OpenAIRequest: defaultRequest(modelFile),
-			}
-		} else {
-			config = &cfg
-		}
-
-		// Set the parameters for the language model prediction
-		updateConfig(config, input)
-
-		if threads != 0 {
-			config.Threads = threads
-		}
-		if ctx != 0 {
-			config.ContextSize = ctx
-		}
-		if f16 {
-			config.F16 = true
-		}
-
-		if debug {
-			config.Debug = true
-		}
-
-		log.Debug().Msgf("Parameter Config: %+v", config)
-
-		predInput := input.Prompt
-		if chat {
-			mess := []string{}
-			for _, i := range input.Messages {
-				r := config.Roles[i.Role]
-				if r == "" {
-					r = i.Role
-				}
-
-				content := fmt.Sprint(r, " ", i.Content)
-				mess = append(mess, content)
-			}
-
-			predInput = strings.Join(mess, "\n")
-		}
-
 		templateFile := config.Model
-		if config.TemplateConfig.Chat != "" && chat {
+
+		if config.TemplateConfig.Chat != "" {
 			templateFile = config.TemplateConfig.Chat
 		}
 
-		if config.TemplateConfig.Completion != "" && !chat {
-			templateFile = config.TemplateConfig.Completion
-		}
-
 		// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
-		templatedInput, err := loader.TemplatePrefix(templateFile, struct {
+		templatedInput, err := o.loader.TemplatePrefix(templateFile, struct {
 			Input string
 		}{Input: predInput})
 		if err == nil {
@@ -263,109 +334,320 @@ func openAIEndpoint(cm ConfigMerger, chat, debug bool, loader *model.ModelLoader
 			log.Debug().Msgf("Template found, input modified to: %s", predInput)
 		}
 
-		result := []Choice{}
+		if input.Stream {
+			responses := make(chan OpenAIResponse)
 
-		n := input.N
+			go process(predInput, input, config, o.loader, responses)
 
-		if input.N == 0 {
-			n = 1
+			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
 		}
 
-		// get the model function to call for the result
-		predFunc, err := ModelInference(predInput, loader, *config)
+		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
 		}
 
-		finetunePrediction := func(prediction string) string {
-			if config.Echo {
-				prediction = predInput + 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
-		}
-
-		for i := 0; i < n; i++ {
-			prediction, err := predFunc()
-			if err != nil {
-				return err
-			}
-
-			prediction = finetunePrediction(prediction)
-
-			if chat {
-				if input.Stream {
-					result = append(result, Choice{Delta: &Message{Role: "assistant", Content: prediction}})
-				} else {
-					result = append(result, Choice{Message: &Message{Role: "assistant", Content: prediction}})
-				}
-			} else {
-				result = append(result, Choice{Text: prediction})
-			}
-		}
-
 		resp := &OpenAIResponse{
 			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
 			Choices: result,
+			Object:  "chat.completion",
 		}
-		if input.Stream && chat {
-			resp.Object = "chat.completion.chunk"
-		} else if chat {
-			resp.Object = "chat.completion"
-		} else {
-			resp.Object = "text_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)
 
-		if input.Stream {
-			log.Debug().Msgf("Handling stream request")
-			c.Context().SetBodyStreamWriter(fasthttp.StreamWriter(func(w *bufio.Writer) {
-				fmt.Fprintf(w, "event: data\n")
-				w.Flush()
-
-				fmt.Fprintf(w, "data: %s\n\n", jsonResult)
-				w.Flush()
-
-				fmt.Fprintf(w, "event: data\n")
-				w.Flush()
-
-				resp := &OpenAIResponse{
-					Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
-					Choices: []Choice{Choice{FinishReason: "stop"}},
-				}
-				respData, _ := json.Marshal(resp)
-
-				fmt.Fprintf(w, "data: %s\n\n", respData)
-				w.Flush()
-
-				//	fmt.Fprintf(w, "data: [DONE]\n\n")
-				//		w.Flush()
-			}))
-			return nil
-		} else {
-			// Return the prediction in the response body
-			return c.JSON(resp)
-		}
+		// Return the prediction in the response body
+		return c.JSON(resp)
 	}
 }
 
-func listModels(loader *model.ModelLoader, cm ConfigMerger) func(ctx *fiber.Ctx) error {
+// 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 {
@@ -379,7 +661,7 @@ func listModels(loader *model.ModelLoader, cm ConfigMerger) func(ctx *fiber.Ctx)
 			dataModels = append(dataModels, OpenAIModel{ID: m, Object: "model"})
 		}
 
-		for k := range cm {
+		for _, k := range cm.ListConfigs() {
 			if _, exists := mm[k]; !exists {
 				dataModels = append(dataModels, OpenAIModel{ID: k, Object: "model"})
 			}

api/options.go

@@ -0,0 +1,108 @@
+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/prediction.go

@@ -2,28 +2,25 @@ package api
 
 import (
 	"fmt"
+	"regexp"
+	"strings"
 	"sync"
 
+	"github.com/donomii/go-rwkv.cpp"
 	model "github.com/go-skynet/LocalAI/pkg/model"
-	gpt2 "github.com/go-skynet/go-gpt2.cpp"
-	gptj "github.com/go-skynet/go-gpt4all-j.cpp"
+	"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 ModelInference(s string, loader *model.ModelLoader, c Config) (func() (string, error), error) {
-	var model *llama.LLama
-	var gptModel *gptj.GPTJ
-	var gpt2Model *gpt2.GPT2
-	var stableLMModel *gpt2.StableLM
-
-	modelFile := c.Model
-
-	// Try to load the model
-	var llamaerr, gpt2err, gptjerr, stableerr error
+func defaultLLamaOpts(c Config) []llama.ModelOption {
 	llamaOpts := []llama.ModelOption{}
 	if c.ContextSize != 0 {
 		llamaOpts = append(llamaOpts, llama.SetContext(c.ContextSize))
@@ -31,136 +28,238 @@ func ModelInference(s string, loader *model.ModelLoader, c Config) (func() (stri
 	if c.F16 {
 		llamaOpts = append(llamaOpts, llama.EnableF16Memory)
 	}
+	if c.Embeddings {
+		llamaOpts = append(llamaOpts, llama.EnableEmbeddings)
+	}
 
-	// TODO: this is ugly, better identifying the model somehow! however, it is a good stab for a first implementation..
-	model, llamaerr = loader.LoadLLaMAModel(modelFile, llamaOpts...)
-	if llamaerr != nil {
-		gptModel, gptjerr = loader.LoadGPTJModel(modelFile)
-		if gptjerr != nil {
-			gpt2Model, gpt2err = loader.LoadGPT2Model(modelFile)
-			if gpt2err != nil {
-				stableLMModel, stableerr = loader.LoadStableLMModel(modelFile)
-				if stableerr != nil {
-					return nil, fmt.Errorf("llama: %s gpt: %s gpt2: %s stableLM: %s", llamaerr.Error(), gptjerr.Error(), gpt2err.Error(), stableerr.Error()) // llama failed first, so we want to catch both errors
-				}
+	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 {
-	case stableLMModel != nil:
+	switch model := inferenceModel.(type) {
+	case *rwkv.RwkvState:
+		supportStreams = true
+
 		fn = func() (string, error) {
-			// Generate the prediction using the language model
-			predictOptions := []gpt2.PredictOption{
-				gpt2.SetTemperature(c.Temperature),
-				gpt2.SetTopP(c.TopP),
-				gpt2.SetTopK(c.TopK),
-				gpt2.SetTokens(c.Maxtokens),
-				gpt2.SetThreads(c.Threads),
+			stopWord := "\n"
+			if len(c.StopWords) > 0 {
+				stopWord = c.StopWords[0]
 			}
 
-			if c.Batch != 0 {
-				predictOptions = append(predictOptions, gpt2.SetBatch(c.Batch))
+			if err := model.ProcessInput(s); err != nil {
+				return "", err
 			}
 
-			if c.Seed != 0 {
-				predictOptions = append(predictOptions, gpt2.SetSeed(c.Seed))
-			}
+			response := model.GenerateResponse(c.Maxtokens, stopWord, float32(c.Temperature), float32(c.TopP), tokenCallback)
 
-			return stableLMModel.Predict(
-				s,
-				predictOptions...,
-			)
+			return response, nil
 		}
-	case gpt2Model != nil:
+	case *transformers.GPTNeoX:
 		fn = func() (string, error) {
 			// Generate the prediction using the language model
-			predictOptions := []gpt2.PredictOption{
-				gpt2.SetTemperature(c.Temperature),
-				gpt2.SetTopP(c.TopP),
-				gpt2.SetTopK(c.TopK),
-				gpt2.SetTokens(c.Maxtokens),
-				gpt2.SetThreads(c.Threads),
+			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, gpt2.SetBatch(c.Batch))
+				predictOptions = append(predictOptions, transformers.SetBatch(c.Batch))
 			}
 
 			if c.Seed != 0 {
-				predictOptions = append(predictOptions, gpt2.SetSeed(c.Seed))
-			}
-
-			return gpt2Model.Predict(
-				s,
-				predictOptions...,
-			)
-		}
-	case gptModel != nil:
-		fn = func() (string, error) {
-			// Generate the prediction using the language model
-			predictOptions := []gptj.PredictOption{
-				gptj.SetTemperature(c.Temperature),
-				gptj.SetTopP(c.TopP),
-				gptj.SetTopK(c.TopK),
-				gptj.SetTokens(c.Maxtokens),
-				gptj.SetThreads(c.Threads),
-			}
-
-			if c.Batch != 0 {
-				predictOptions = append(predictOptions, gptj.SetBatch(c.Batch))
-			}
-
-			if c.Seed != 0 {
-				predictOptions = append(predictOptions, gptj.SetSeed(c.Seed))
-			}
-
-			return gptModel.Predict(
-				s,
-				predictOptions...,
-			)
-		}
-	case model != nil:
-		fn = func() (string, error) {
-			// 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.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))
+				predictOptions = append(predictOptions, transformers.SetSeed(c.Seed))
 			}
 
 			return model.Predict(
@@ -168,6 +267,220 @@ func ModelInference(s string, loader *model.ModelLoader, c Config) (func() (stri
 				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) {
@@ -183,6 +496,66 @@ func ModelInference(s string, loader *model.ModelLoader, c Config) (func() (stri
 		l.Lock()
 		defer l.Unlock()
 
-		return fn()
+		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
+
+}

charts/local-ai/Chart.yaml

@@ -1,6 +0,0 @@
-apiVersion: v2
-appVersion: 0.1.0
-description: A Helm chart for LocalAI
-name: local-ai
-type: application
-version: 1.0.0

charts/local-ai/templates/_helpers.tpl

@@ -1,44 +0,0 @@
-{{/*
-Expand the name of the chart.
-*/}}
-{{- define "local-ai.name" -}}
-{{- default .Chart.Name .Values.nameOverride | trunc 63 | trimSuffix "-" }}
-{{- end }}
-
-{{/*
-Create a default fully qualified app name.
-We truncate at 63 chars because some Kubernetes name fields are limited to this (by the DNS naming spec).
-If release name contains chart name it will be used as a full name.
-*/}}
-{{- define "local-ai.fullname" -}}
-{{- if .Values.fullnameOverride }}
-{{- .Values.fullnameOverride | trunc 63 | trimSuffix "-" }}
-{{- else }}
-{{- $name := default .Chart.Name .Values.nameOverride }}
-{{- if contains $name .Release.Name }}
-{{- .Release.Name | trunc 63 | trimSuffix "-" }}
-{{- else }}
-{{- printf "%s-%s" .Release.Name $name | trunc 63 | trimSuffix "-" }}
-{{- end }}
-{{- end }}
-{{- end }}
-
-{{/*
-Create chart name and version as used by the chart label.
-*/}}
-{{- define "local-ai.chart" -}}
-{{- printf "%s-%s" .Chart.Name .Chart.Version | replace "+" "_" | trunc 63 | trimSuffix "-" }}
-{{- end }}
-
-{{/*
-Common labels
-*/}}
-{{- define "local-ai.labels" -}}
-helm.sh/chart: {{ include "local-ai.chart" . }}
-app.kubernetes.io/name: {{ include "local-ai.name" . }}
-app.kubernetes.io/instance: "{{ .Release.Name }}"
-app.kubernetes.io/managed-by: {{ .Release.Service }}
-{{- if .Chart.AppVersion }}
-app.kubernetes.io/version: {{ .Chart.AppVersion | quote }}
-{{- end }}
-{{- end }}

charts/local-ai/templates/data-volume.yaml

@@ -1,39 +0,0 @@
-{{- if .Values.dataVolume.enabled }}
-apiVersion: cdi.kubevirt.io/v1beta1
-kind: DataVolume
-metadata:
-  name: {{ template "local-ai.fullname" . }}
-  namespace: {{ .Release.Namespace | quote }}
-  labels:
-    {{- include "local-ai.labels" . | nindent 4 }}
-spec:
-  contentType: archive
-  source:
-    {{ .Values.dataVolume.source.type }}:
-      url: {{ .Values.dataVolume.source.url }}
-      secretRef: {{ template "local-ai.fullname" . }}
-      {{- if and (eq .Values.dataVolume.source.type "http") .Values.dataVolume.source.secretExtraHeaders }}
-      secretExtraHeaders: {{ .Values.dataVolume.source.secretExtraHeaders }}
-      {{- end }}
-      {{- if .Values.dataVolume.source.caCertConfigMap }}
-      caCertConfigMap: {{ .Values.dataVolume.source.caCertConfigMap }}
-      {{- end }}
-  pvc:
-    accessModes: {{ .Values.dataVolume.pvc.accessModes }}
-    resources:
-      requests:
-        storage: {{ .Values.dataVolume.pvc.size }}
----
-{{- if .Values.dataVolume.secret.enabled }}
-apiVersion: v1
-kind: Secret
-metadata:
-  name: {{ template "local-ai.fullname" . }}
-  namespace: {{ .Release.Namespace | quote }}
-  labels:
-    {{- include "local-ai.labels" . | nindent 4 }}
-data:
-  accessKeyId: {{ .Values.dataVolume.secret.username }}
-  secretKey: {{ .Values.dataVolume.secret.password }}
-{{- end }}
-{{- end }}

charts/local-ai/templates/deployment.yaml

@@ -1,39 +0,0 @@
-apiVersion: apps/v1
-kind: Deployment
-metadata:
-  name: {{ template "local-ai.fullname" . }}
-  namespace: {{ .Release.Namespace | quote }}
-  labels:
-    {{- include "local-ai.labels" . | nindent 4 }}
-spec:
-  selector:
-    matchLabels:
-      app.kubernetes.io/name: {{ include "local-ai.name" . }}
-      app.kubernetes.io/instance: {{ .Release.Name }}
-  replicas: 1
-  template:
-    metadata:
-      name: {{ template "local-ai.fullname" . }}
-      labels:
-        app.kubernetes.io/name: {{ include "local-ai.name" . }}
-        app.kubernetes.io/instance: {{ .Release.Name }}
-    spec:
-      containers:
-        - name: {{ template "local-ai.fullname" . }}
-          image: {{ .Values.deployment.image }}
-          env:
-          - name: THREADS
-            value: {{ .Values.deployment.env.threads | quote }}
-          - name: CONTEXT_SIZE
-            value: {{ .Values.deployment.env.contextSize | quote }}
-          - name: MODELS_PATH
-            value: {{ .Values.deployment.env.modelsPath }}
-{{- if .Values.deployment.volume.enabled }}
-          volumeMounts:
-          - mountPath: {{ .Values.deployment.env.modelsPath }}
-            name: models
-      volumes:
-      - name: models
-        persistentVolumeClaim:
-          claimName: {{ template "local-ai.fullname" . }}
-{{- end }}

charts/local-ai/templates/service.yaml

@@ -1,19 +0,0 @@
-apiVersion: v1
-kind: Service
-metadata:
-  name: {{ template "local-ai.fullname" . }}
-  namespace: {{ .Release.Namespace | quote }}
-  labels:
-    {{- include "local-ai.labels" . | nindent 4 }}
-{{- if .Values.service.annotations }}
-  annotations:
-  {{ toYaml .Values.service.annotations | indent 4 }}
-{{- end }}
-spec:
-  selector:
-    app.kubernetes.io/name: {{ include "local-ai.name" . }}
-  type: "{{ .Values.service.type }}"
-  ports:
-    - protocol: TCP
-      port: 8080
-      targetPort: 8080

charts/local-ai/values.yaml

@@ -1,38 +0,0 @@
-deployment:
-  image: quay.io/go-skynet/local-ai:latest
-  env:
-    threads: 14
-    contextSize: 512
-    modelsPath: "/models"
-  volume:
-    enabled: false
-
-service:
-  type: ClusterIP
-  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"
-
-# Optionally create a PVC containing a model binary, sourced from an arbitrary HTTP server or S3 bucket
-# (requires https://github.com/kubevirt/containerized-data-importer)
-dataVolume:
-  enabled: false
-  source:
-    type: "http" # Source type. One of: [ http | s3 ]
-    url: "http://<model_server>/<model_archive>" # e.g. koala-7B-4bit-128g.GGML.tar
-
-    # CertConfigMap is an optional ConfigMap reference, containing a Certificate Authority (CA) public key
-    # and a base64 encoded pem certificate
-    caCertConfigMap: ""
-
-    # SecretExtraHeaders is an optional list of Secret references, each containing an extra HTTP header
-    # that may include sensitive information. Only applicable for the http source type.
-    secretExtraHeaders: []
-  pvc:
-    accessModes:
-    - ReadWriteOnce
-    size: 5Gi
-  secret:
-    enabled: false
-    username: "" # base64 encoded
-    password: "" # base64 encoded

docker-compose.yaml

@@ -5,11 +5,11 @@ services:
     image: quay.io/go-skynet/local-ai:latest
     build:
       context: .
-      dockerfile: Dockerfile
+      dockerfile: Dockerfile.dev
     ports:
       - 8080:8080
     env_file:
       - .env
     volumes:
       - ./models:/models:cached
-    command: ["/usr/bin/local-ai" ]
+    command: ["/usr/bin/local-ai" ]

entrypoint.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+
+cd /build
+
+if [ "$REBUILD" != "false" ]; then
+	make rebuild
+fi
+
+./local-ai "$@"

examples/README.md

@@ -2,10 +2,101 @@
 
 Here is a list of projects that can easily be integrated with the LocalAI backend. 
 
-## Projects
+### Projects
 
-- [chatbot-ui](https://github.com/go-skynet/LocalAI/tree/master/examples/chatbot-ui/) (by [@mudler](https://github.com/mudler))
+
+### Chatbot-UI
+
+_by [@mkellerman](https://github.com/mkellerman)_
+
+![Screenshot from 2023-04-26 23-59-55](https://user-images.githubusercontent.com/2420543/234715439-98d12e03-d3ce-4f94-ab54-2b256808e05e.png)
+
+This integration shows how to use LocalAI with [mckaywrigley/chatbot-ui](https://github.com/mckaywrigley/chatbot-ui).
+
+[Check it out here](https://github.com/go-skynet/LocalAI/tree/master/examples/chatbot-ui/)
+
+### Discord bot
+
+_by [@mudler](https://github.com/mudler)_
+
+Run a discord bot which lets you talk directly with a model
+
+[Check it out here](https://github.com/go-skynet/LocalAI/tree/master/examples/discord-bot/), or for a live demo you can talk with our bot in #random-bot in our discord server.
+
+### Langchain
+
+_by [@dave-gray101](https://github.com/dave-gray101)_
+
+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/)
+
+### Langchain Python
+
+_by [@mudler](https://github.com/mudler)_
+
+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 WebUI
+
+_by [@dhruvgera](https://github.com/dhruvgera)_
+
+![image](https://user-images.githubusercontent.com/42107491/235344183-44b5967d-ba22-4331-804c-8da7004a5d35.png)
+
+A light, community-maintained web interface for LocalAI
+
+[Check it out here](https://github.com/go-skynet/LocalAI/tree/master/examples/localai-webui/)
+
+### How to run rwkv models
+
+_by [@mudler](https://github.com/mudler)_
+
+A full example on how to run RWKV models with LocalAI
+
+[Check it out here](https://github.com/go-skynet/LocalAI/tree/master/examples/rwkv/)
+
+### PrivateGPT
+
+_by [@mudler](https://github.com/mudler)_
+
+A full example on how to run PrivateGPT with LocalAI
+
+[Check it out here](https://github.com/go-skynet/LocalAI/tree/master/examples/privateGPT/)
+
+### Slack bot
+
+_by [@mudler](https://github.com/mudler)_
+
+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/)
+
+### Question answering on documents with llama-index
+
+_by [@mudler](https://github.com/mudler)_
+
+Shows how to integrate with [Llama-Index](https://gpt-index.readthedocs.io/en/stable/getting_started/installation.html) to enable question answering on a set of documents.
+
+[Check it out here](https://github.com/go-skynet/LocalAI/tree/master/examples/query_data/)
+
+### Question answering on documents with langchain and chroma
+
+_by [@mudler](https://github.com/mudler)_
+
+Shows how to integrate with `Langchain` and `Chroma` to enable question answering on a set of documents.
+
+[Check it out here](https://github.com/go-skynet/LocalAI/tree/master/examples/langchain-chroma/)
+
+### Template for Runpod.io
+
+_by [@fHachenberg](https://github.com/fHachenberg)_
+
+Allows to run any LocalAI-compatible model as a backend on the servers of https://runpod.io
+
+[Check it out here](https://runpod.io/gsc?template=uv9mtqnrd0&ref=984wlcra)
 
 ## Want to contribute?
 
-Create an issue, and put `Example: <description>` in the title! We will post your examples here.
+Create an issue, and put `Example: <description>` in the title! We will post your examples here.

examples/chatbot-ui/README.md

@@ -19,8 +19,30 @@ cd LocalAI/examples/chatbot-ui
 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 -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/docker-compose.yaml

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

examples/chatbot-ui/models/gpt-3.5-turbo.yaml

@@ -5,7 +5,6 @@ parameters:
   temperature: 0.2
   top_p: 0.7
 context_size: 1024
-threads: 14
 stopwords:
 - "HUMAN:"
 - "GPT:"

examples/discord-bot/.env.example

@@ -0,0 +1,6 @@
+OPENAI_API_KEY=x
+DISCORD_BOT_TOKEN=x
+DISCORD_CLIENT_ID=x
+OPENAI_API_BASE=http://api:8080
+ALLOWED_SERVER_IDS=x
+SERVER_TO_MODERATION_CHANNEL=1:1

examples/discord-bot/README.md

@@ -0,0 +1,76 @@
+# discord-bot
+
+![Screenshot from 2023-05-01 07-58-19](https://user-images.githubusercontent.com/2420543/235413924-0cb2e75b-f2d6-4119-8610-44386e44afb8.png)
+
+## Setup
+
+```bash
+# Clone LocalAI
+git clone https://github.com/go-skynet/LocalAI
+
+cd LocalAI/examples/discord-bot
+
+# (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
+
+# Set the discord bot options (see: https://github.com/go-skynet/gpt-discord-bot#setup)
+cp -rfv .env.example .env
+vim .env
+
+# start with docker-compose
+docker-compose up -d --build
+```
+
+Note: see setup options here: https://github.com/go-skynet/gpt-discord-bot#setup
+
+Open up the URL in the console and give permission to the bot in your server. Start a thread with `/chat ..`
+
+## Kubernetes
+
+- install the local-ai chart first
+- change OPENAI_API_BASE to point to the API address and apply the discord-bot manifest:
+
+```yaml
+apiVersion: v1
+kind: Namespace
+metadata:
+  name: discord-bot
+---
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: localai
+  namespace: discord-bot
+  labels:
+    app: localai
+spec:
+  selector:
+    matchLabels:
+      app: localai
+  replicas: 1
+  template:
+    metadata:
+      labels:
+        app: localai
+      name: localai
+    spec:
+      containers:
+        - name: localai-discord
+          env:
+          - name: OPENAI_API_KEY
+            value: "x"
+          - name: DISCORD_BOT_TOKEN
+            value: ""
+          - name: DISCORD_CLIENT_ID
+            value: ""
+          - name: OPENAI_API_BASE
+            value: "http://local-ai.default.svc.cluster.local:8080"
+          - name: ALLOWED_SERVER_IDS
+            value: "xx"
+          - name: SERVER_TO_MODERATION_CHANNEL
+            value: "1:1"
+          image: quay.io/go-skynet/gpt-discord-bot:main
+```

examples/discord-bot/docker-compose.yaml

@@ -0,0 +1,21 @@
+version: '3.6'
+
+services:
+  api:
+    image: quay.io/go-skynet/local-ai:latest
+    build:
+      context: ../../
+      dockerfile: Dockerfile.dev
+    ports:
+      - 8080:8080
+    environment:
+      - DEBUG=true
+      - MODELS_PATH=/models
+    volumes:
+      - ./models:/models:cached
+    command: ["/usr/bin/local-ai" ]
+
+  bot:
+    image: quay.io/go-skynet/gpt-discord-bot:main
+    env_file:
+    - .env

examples/discord-bot/models

@@ -0,0 +1 @@
+../chatbot-ui/models/

examples/langchain-chroma/.env.example

@@ -0,0 +1,5 @@
+THREADS=4
+CONTEXT_SIZE=512
+MODELS_PATH=/models
+DEBUG=true
+# BUILD_TYPE=generic

examples/langchain-chroma/.gitignore

@@ -0,0 +1,4 @@
+db/
+state_of_the_union.txt
+models/bert
+models/ggml-gpt4all-j

examples/langchain-chroma/README.md

@@ -0,0 +1,63 @@
+# Data query example
+
+This example makes use of [langchain and chroma](https://blog.langchain.dev/langchain-chroma/) to enable question answering on a set of documents.
+
+## Setup
+
+Download the models and start the API:
+
+```bash
+# Clone LocalAI
+git clone https://github.com/go-skynet/LocalAI
+
+cd LocalAI/examples/langchain-chroma
+
+wget https://huggingface.co/skeskinen/ggml/resolve/main/all-MiniLM-L6-v2/ggml-model-q4_0.bin -O models/bert
+wget https://gpt4all.io/models/ggml-gpt4all-j.bin -O models/ggml-gpt4all-j
+
+# configure your .env
+# NOTE: ensure that THREADS does not exceed your machine's CPU cores
+mv .env.example .env
+
+# start with docker-compose
+docker-compose up -d --build
+
+# tail the logs & wait until the build completes
+docker logs -f langchain-chroma-api-1
+```
+
+### Python requirements
+
+```
+pip install -r requirements.txt
+```
+
+### Create a storage
+
+In this step we will create a local vector database from our document set, so later we can ask questions on it with the LLM.
+
+Note: **OPENAI_API_KEY** is not required. However the library might fail if no API_KEY is passed by, so an arbitrary string can be used.
+
+```bash
+export OPENAI_API_BASE=http://localhost:8080/v1
+export OPENAI_API_KEY=sk-
+
+wget https://raw.githubusercontent.com/hwchase17/chat-your-data/master/state_of_the_union.txt
+python store.py
+```
+
+After it finishes, a directory "db" will be created with the vector index database.
+
+## Query
+
+We can now query the dataset. 
+
+```bash
+export OPENAI_API_BASE=http://localhost:8080/v1
+export OPENAI_API_KEY=sk-
+
+python query.py
+# President Trump recently stated during a press conference regarding tax reform legislation that "we're getting rid of all these loopholes." He also mentioned that he wants to simplify the system further through changes such as increasing the standard deduction amount and making other adjustments aimed at reducing taxpayers' overall burden.    
+```
+
+Keep in mind now things are hit or miss!

examples/langchain-chroma/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-chroma/models/embeddings.yaml

@@ -0,0 +1,6 @@
+name: text-embedding-ada-002
+parameters:
+  model: bert
+threads: 4
+backend: bert-embeddings
+embeddings: true

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

@@ -0,0 +1,16 @@
+name: gpt-3.5-turbo
+parameters:
+  model: ggml-gpt4all-j
+  top_k: 80
+  temperature: 0.2
+  top_p: 0.7
+context_size: 1024
+stopwords:
+- "HUMAN:"
+- "GPT:"
+roles:
+  user: " "
+  system: " "
+template:
+  completion: completion
+  chat: gpt4all

examples/langchain-chroma/query.py

@@ -0,0 +1,23 @@
+
+import os
+from langchain.vectorstores import Chroma
+from langchain.embeddings import OpenAIEmbeddings
+from langchain.chat_models import ChatOpenAI
+from langchain.chains import RetrievalQA
+from langchain.vectorstores.base import VectorStoreRetriever
+
+base_path = os.environ.get('OPENAI_API_BASE', 'http://localhost:8080/v1')
+
+# Load and process the text
+embedding = OpenAIEmbeddings()
+persist_directory = 'db'
+
+# Now we can load the persisted database from disk, and use it as normal. 
+llm = ChatOpenAI(temperature=0, model_name="gpt-3.5-turbo", openai_api_base=base_path)
+vectordb = Chroma(persist_directory=persist_directory, embedding_function=embedding)
+retriever = VectorStoreRetriever(vectorstore=vectordb)
+qa = RetrievalQA.from_llm(llm=llm, retriever=retriever)
+
+query = "What the president said about taxes ?"
+print(qa.run(query))
+

examples/langchain-chroma/requirements.txt

@@ -0,0 +1,4 @@
+langchain==0.0.160
+openai==0.27.6
+chromadb==0.3.21
+llama-index==0.6.2

examples/langchain-chroma/store.py

@@ -0,0 +1,25 @@
+
+import os
+from langchain.vectorstores import Chroma
+from langchain.embeddings import OpenAIEmbeddings
+from langchain.text_splitter import CharacterTextSplitter
+from langchain.document_loaders import TextLoader
+
+base_path = os.environ.get('OPENAI_API_BASE', 'http://localhost:8080/v1')
+
+# Load and process the text
+loader = TextLoader('state_of_the_union.txt')
+documents = loader.load()
+
+text_splitter = CharacterTextSplitter(chunk_size=300, chunk_overlap=70)
+texts = text_splitter.split_documents(documents)
+
+# Embed and store the texts
+# Supplying a persist_directory will store the embeddings on disk
+persist_directory = 'db'
+
+embedding = OpenAIEmbeddings(model="text-embedding-ada-002")
+vectordb = Chroma.from_documents(documents=texts, embedding=embedding, persist_directory=persist_directory)
+
+vectordb.persist()
+vectordb = None

examples/langchain-python/README.md

@@ -0,0 +1,36 @@
+## Langchain-python
+
+Langchain example from [quickstart](https://python.langchain.com/en/latest/getting_started/getting_started.html).
+
+To interact with langchain, you can just set the `OPENAI_API_BASE` URL and provide a token with a random string.
+
+See the example below:
+
+```
+# Clone LocalAI
+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
+
+
+pip install langchain
+pip install openai
+
+export OPENAI_API_BASE=http://localhost:8080
+# Note: **OPENAI_API_KEY** is not required. However the library might fail if no API_KEY is passed by, so an arbitrary string can be used.
+export OPENAI_API_KEY=sk-
+
+python test.py
+# A good company name for a company that makes colorful socks would be "Colorsocks".
+
+python agent.py
+```

examples/langchain-python/agent.py

@@ -0,0 +1,44 @@
+## This is a fork/based from https://gist.github.com/wiseman/4a706428eaabf4af1002a07a114f61d6
+
+from io import StringIO
+import sys
+import os
+from typing import Dict, Optional
+
+from langchain.agents import load_tools
+from langchain.agents import initialize_agent
+from langchain.agents.tools import Tool
+from langchain.llms import OpenAI
+
+base_path = os.environ.get('OPENAI_API_BASE', 'http://localhost:8080/v1')
+model_name = os.environ.get('MODEL_NAME', 'gpt-3.5-turbo')
+
+class PythonREPL:
+    """Simulates a standalone Python REPL."""
+
+    def __init__(self):
+        pass        
+
+    def run(self, command: str) -> str:
+        """Run command and returns anything printed."""
+        old_stdout = sys.stdout
+        sys.stdout = mystdout = StringIO()
+        try:
+            exec(command, globals())
+            sys.stdout = old_stdout
+            output = mystdout.getvalue()
+        except Exception as e:
+            sys.stdout = old_stdout
+            output = str(e)
+        return output
+
+llm = OpenAI(temperature=0.0, openai_api_base=base_path, model_name=model_name)
+python_repl = Tool(
+        "Python REPL",
+        PythonREPL().run,
+        """A Python shell. Use this to execute python commands. Input should be a valid python command.
+        If you expect output it should be printed out.""",
+    )
+tools = [python_repl]
+agent = initialize_agent(tools, llm, agent="zero-shot-react-description", verbose=True)
+agent.run("What is the 10th fibonacci number?")

examples/langchain-python/docker-compose.yaml

@@ -0,0 +1,16 @@
+version: '3.6'
+
+services:
+  api:
+    image: quay.io/go-skynet/local-ai:latest
+    build:
+      context: ../../
+      dockerfile: Dockerfile.dev
+    ports:
+      - 8080:8080
+    environment:
+      - DEBUG=true
+      - MODELS_PATH=/models
+    volumes:
+      - ./models:/models:cached
+    command: ["/usr/bin/local-ai" ]

examples/langchain-python/models

@@ -0,0 +1 @@
+../chatbot-ui/models

examples/langchain-python/test.py

@@ -0,0 +1,6 @@
+
+from langchain.llms import OpenAI
+
+llm = OpenAI(temperature=0.9,model_name="gpt-3.5-turbo")
+text = "What would be a good company name for a company that makes colorful socks?"
+print(llm(text))

examples/langchain/.gitignore

@@ -0,0 +1,2 @@
+models/ggml-koala-13B-4bit-128g
+models/ggml-gpt4all-j

examples/langchain/JS.Dockerfile

@@ -0,0 +1,6 @@
+FROM node:latest
+COPY ./langchainjs-localai-example /app
+WORKDIR /app
+RUN npm install
+RUN npm run build
+ENTRYPOINT [ "npm", "run", "start" ]

examples/langchain/PY.Dockerfile

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

examples/langchain/README.md

@@ -0,0 +1,30 @@
+# langchain
+
+Example of using langchain, with the standard OpenAI llm module, and LocalAI. Has docker compose profiles for both the Typescript and Python versions.
+
+**Please Note** - This is a tech demo example at this time. ggml-gpt4all-j has pretty terrible results for most langchain applications with the settings used in this example.
+
+## Setup
+
+```bash
+# Clone LocalAI
+git clone https://github.com/go-skynet/LocalAI
+
+cd LocalAI/examples/langchain
+
+# (optional) - Edit the example code in typescript.
+# vi ./langchainjs-localai-example/index.ts
+
+# Download gpt4all-j to models/
+wget https://gpt4all.io/models/ggml-gpt4all-j.bin -O models/ggml-gpt4all-j
+
+# start with docker-compose for typescript!
+docker-compose --profile ts up --build
+
+# or start with docker-compose for python!
+docker-compose --profile py up --build
+```
+
+## Copyright
+
+Some of the example code in index.mts and full_demo.py is adapted from the langchainjs project and is Copyright (c) Harrison Chase. Used under the terms of the MIT license, as is the remainder of this code.

examples/langchain/docker-compose.yaml

@@ -0,0 +1,43 @@
+version: '3.6'
+
+services:
+  api:
+    image: quay.io/go-skynet/local-ai:latest
+    build:
+      context: ../../
+      dockerfile: Dockerfile.dev
+    ports:
+      - 8080:8080
+    environment:
+      - DEBUG=true
+      - MODELS_PATH=/models
+    volumes:
+      - ./models:/models:cached
+    command: ["/usr/bin/local-ai" ]
+
+  js:
+    build:
+      context: .
+      dockerfile: JS.Dockerfile
+    profiles:
+      - js
+      - ts
+    depends_on:
+    - "api"
+    environment:
+      - 'OPENAI_API_KEY=sk-XXXXXXXXXXXXXXXXXXXX'
+      - 'OPENAI_API_BASE=http://api:8080/v1'
+      - 'MODEL_NAME=gpt-3.5-turbo' #gpt-3.5-turbo' # ggml-gpt4all-j' # ggml-koala-13B-4bit-128g'
+
+  py:
+    build:
+      context: .
+      dockerfile: PY.Dockerfile
+    profiles:
+      - py
+    depends_on:
+    - "api"
+    environment:
+      - 'OPENAI_API_KEY=sk-XXXXXXXXXXXXXXXXXXXX'
+      - 'OPENAI_API_BASE=http://api:8080/v1'
+      - 'MODEL_NAME=gpt-3.5-turbo' #gpt-3.5-turbo' # ggml-gpt4all-j' # ggml-koala-13B-4bit-128g'

examples/langchain/langchainjs-localai-example/.gitignore

@@ -0,0 +1,2 @@
+node_modules/
+dist/

examples/langchain/langchainjs-localai-example/.vscode/launch.json

@@ -0,0 +1,20 @@
+{
+    // Use IntelliSense to learn about possible attributes.
+    // Hover to view descriptions of existing attributes.
+    // For more information, visit: https://go.microsoft.com/fwlink/?linkid=830387
+    "version": "0.2.0",
+    "configurations": [
+        {
+            "type": "node",
+            "request": "launch",
+            "name": "Launch Program",
+            // "skipFiles": [
+            //     "<node_internals>/**"
+            // ],
+            "program": "${workspaceFolder}\\dist\\index.mjs",
+            "outFiles": [
+                "${workspaceFolder}/**/*.js"
+            ]
+        }
+    ]
+}

examples/langchain/langchainjs-localai-example/package.json

@@ -0,0 +1,21 @@
+{
+  "name": "langchainjs-localai-example",
+  "version": "0.1.0",
+  "description": "Trivial Example of using langchain + the OpenAI API + LocalAI together",
+  "main": "index.mjs",
+  "scripts": {
+    "build": "tsc --build",
+    "clean": "tsc --build --clean",
+    "start": "node --trace-warnings dist/index.mjs"
+  },
+  "author": "dave@gray101.com",
+  "license": "MIT",
+  "devDependencies": {
+    "@types/node": "^18.16.4",
+    "typescript": "^5.0.4"
+  },
+  "dependencies": {
+    "langchain": "^0.0.67",
+    "typeorm": "^0.3.15"
+  }
+}

examples/langchain/langchainjs-localai-example/src/index.mts

@@ -0,0 +1,79 @@
+import { OpenAIChat } from "langchain/llms/openai";
+import { loadQAStuffChain } from "langchain/chains";
+import { Document } from "langchain/document";
+import { initializeAgentExecutorWithOptions } from "langchain/agents";
+import {Calculator} from "langchain/tools/calculator";
+
+const pathToLocalAi = process.env['OPENAI_API_BASE'] || 'http://api:8080/v1';
+const fakeApiKey = process.env['OPENAI_API_KEY'] || '-';
+const modelName = process.env['MODEL_NAME'] || 'gpt-3.5-turbo';
+
+function getModel(): OpenAIChat {
+  return new OpenAIChat({
+    prefixMessages: [
+      {
+        role: "system",
+        content: "You are a helpful assistant that answers in pirate language",
+      },
+    ],
+    modelName: modelName,
+    maxTokens: 50,
+    openAIApiKey: fakeApiKey,
+    maxRetries: 2
+  }, {
+    basePath: pathToLocalAi,
+    apiKey: fakeApiKey,
+  });
+}
+
+// Minimal example.
+export const run = async () => {
+  const model = getModel();
+  console.log(`about to model.call at ${new Date().toUTCString()}`);
+  const res = await model.call(
+    "What would be a good company name a company that makes colorful socks?"
+  );
+  console.log(`${new Date().toUTCString()}`);
+  console.log({ res });
+};
+
+await run();
+
+// This example uses the `StuffDocumentsChain`
+export const run2 = async () => {
+  const model = getModel();
+  const chainA = loadQAStuffChain(model);
+  const docs = [
+    new Document({ pageContent: "Harrison went to Harvard." }),
+    new Document({ pageContent: "Ankush went to Princeton." }),
+  ];
+  const resA = await chainA.call({
+    input_documents: docs,
+    question: "Where did Harrison go to college?",
+  });
+  console.log({ resA });
+};
+
+await run2();
+
+// Quickly thrown together example of using tools + agents.
+// This seems like it should work, but it doesn't yet.
+export const temporarilyBrokenToolTest = async () => {
+  const model = getModel();
+
+  const executor = await initializeAgentExecutorWithOptions([new Calculator(true)], model, {
+    agentType: "zero-shot-react-description",
+  });
+
+  console.log("Loaded agent.");
+
+  const input = `What is the value of (500 *2) + 350 - 13?`;
+
+  console.log(`Executing with input "${input}"...`);
+
+  const result = await executor.call({ input });
+
+  console.log(`Got output ${result.output}`);
+}
+
+await temporarilyBrokenToolTest();

examples/langchain/langchainjs-localai-example/tsconfig.json

@@ -0,0 +1,15 @@
+{
+  "compilerOptions": {
+    "target": "es2022",
+    "lib": ["ES2022", "DOM"],
+    "module": "ES2022",
+    "moduleResolution": "node",
+    "strict": true,
+    "esModuleInterop": true,
+    "allowSyntheticDefaultImports": true,
+    "isolatedModules": true,
+    "outDir": "./dist"
+  },
+  "include": ["src", "test"],
+  "exclude": ["node_modules", "dist"]
+}

examples/langchain/langchainpy-localai-example/.vscode/launch.json

@@ -0,0 +1,24 @@
+{
+    "version": "0.2.0",
+    "configurations": [
+        {
+            "name": "Python: Current File",
+            "type": "python",
+            "request": "launch",
+            "program": "${file}",
+            "console": "integratedTerminal",
+            "redirectOutput": true,
+            "justMyCode": false
+        },
+        {
+            "name": "Python: Attach to Port 5678",
+            "type": "python",
+            "request": "attach",
+            "connect": {
+                "host": "localhost",
+                "port": 5678
+              },
+            "justMyCode": false
+        }
+    ]
+}

examples/langchain/langchainpy-localai-example/.vscode/settings.json

@@ -0,0 +1,3 @@
+{
+    "python.defaultInterpreterPath": "${workspaceFolder}/.venv/Scripts/python"
+}

examples/langchain/langchainpy-localai-example/full_demo.py

@@ -0,0 +1,46 @@
+import os
+import logging
+
+from langchain.chat_models import ChatOpenAI
+from langchain import PromptTemplate, LLMChain
+from langchain.prompts.chat import (
+    ChatPromptTemplate,
+    SystemMessagePromptTemplate,
+    AIMessagePromptTemplate,
+    HumanMessagePromptTemplate,
+)
+from langchain.schema import (
+    AIMessage,
+    HumanMessage,
+    SystemMessage
+)
+
+# This logging incantation makes it easy to see that you're actually reaching your LocalAI instance rather than OpenAI.
+logging.basicConfig(level=logging.DEBUG)
+
+print('Langchain + LocalAI PYTHON Tests')
+
+base_path = os.environ.get('OPENAI_API_BASE', 'http://api:8080/v1')
+key = os.environ.get('OPENAI_API_KEY', '-')
+model_name = os.environ.get('MODEL_NAME', 'gpt-3.5-turbo')
+
+
+chat = ChatOpenAI(temperature=0, openai_api_base=base_path, openai_api_key=key, model_name=model_name, max_tokens=100)
+
+print("Created ChatOpenAI for ", chat.model_name)
+
+template = "You are a helpful assistant that translates {input_language} to {output_language}. The next message will be a sentence in {input_language}. Respond ONLY with the translation in {output_language}. Do not respond in {input_language}!"
+system_message_prompt = SystemMessagePromptTemplate.from_template(template)
+human_template = "{text}"
+human_message_prompt = HumanMessagePromptTemplate.from_template(human_template)
+
+chat_prompt = ChatPromptTemplate.from_messages([system_message_prompt, human_message_prompt])
+
+print("ABOUT to execute")
+
+# get a chat completion from the formatted messages
+response = chat(chat_prompt.format_prompt(input_language="English", output_language="French", text="I love programming.").to_messages())
+
+print(response)
+
+print(".");

examples/langchain/langchainpy-localai-example/requirements.txt

@@ -0,0 +1,32 @@
+aiohttp==3.8.4
+aiosignal==1.3.1
+async-timeout==4.0.2
+attrs==23.1.0
+certifi==2022.12.7
+charset-normalizer==3.1.0
+colorama==0.4.6
+dataclasses-json==0.5.7
+debugpy==1.6.7
+frozenlist==1.3.3
+greenlet==2.0.2
+idna==3.4
+langchain==0.0.159
+marshmallow==3.19.0
+marshmallow-enum==1.5.1
+multidict==6.0.4
+mypy-extensions==1.0.0
+numexpr==2.8.4
+numpy==1.24.3
+openai==0.27.6
+openapi-schema-pydantic==1.2.4
+packaging==23.1
+pydantic==1.10.7
+PyYAML==6.0
+requests==2.29.0
+SQLAlchemy==2.0.12
+tenacity==8.2.2
+tqdm==4.65.0
+typing-inspect==0.8.0
+typing_extensions==4.5.0
+urllib3==1.26.15
+yarl==1.9.2

examples/langchain/langchainpy-localai-example/simple_demo.py

@@ -0,0 +1,6 @@
+
+from langchain.llms import OpenAI
+
+llm = OpenAI(temperature=0.9,model_name="gpt-3.5-turbo")
+text = "What would be a good company name for a company that makes colorful socks?"
+print(llm(text))

examples/langchain/models/completion.tmpl

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

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

@@ -0,0 +1,17 @@
+name: gpt-3.5-turbo
+parameters:
+  model: ggml-gpt4all-j # ggml-koala-13B-4bit-128g
+  top_k: 80
+  temperature: 0.2
+  top_p: 0.7
+context_size: 1024
+stopwords:
+- "HUMAN:"
+- "GPT:"
+roles:
+  user: " "
+  system: " "
+backend: "gptj"
+template:
+  completion: completion
+  chat: gpt4all

examples/langchain/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/localai-webui/README.md

@@ -0,0 +1,26 @@
+# localai-webui
+
+Example of integration with [dhruvgera/localai-frontend](https://github.com/Dhruvgera/LocalAI-frontend).
+
+![image](https://user-images.githubusercontent.com/42107491/235344183-44b5967d-ba22-4331-804c-8da7004a5d35.png)
+
+## Setup
+
+```bash
+# Clone LocalAI
+git clone https://github.com/go-skynet/LocalAI
+
+cd LocalAI/examples/localai-webui
+
+# (optional) Checkout a specific LocalAI tag
+# git checkout -b build <TAG>
+
+# Download any desired models to models/ in the parent LocalAI project dir
+# For example: wget https://gpt4all.io/models/ggml-gpt4all-j.bin
+
+# start with docker-compose
+docker-compose up -d --build
+```
+
+Open http://localhost:3000 for the Web UI.
+

examples/localai-webui/docker-compose.yml

@@ -0,0 +1,20 @@
+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"]
+
+  frontend:
+    image: quay.io/go-skynet/localai-frontend:master
+    ports:
+      - 3000:3000

examples/privateGPT/README.md

@@ -0,0 +1,25 @@
+# privateGPT
+
+This example is a re-adaptation of https://github.com/imartinez/privateGPT to work with LocalAI and OpenAI endpoints. We have a fork with the changes required to work with privateGPT here https://github.com/go-skynet/privateGPT ( PR: https://github.com/imartinez/privateGPT/pull/408 ).
+
+Follow the instructions in https://github.com/go-skynet/privateGPT:
+
+```bash
+git clone git@github.com:go-skynet/privateGPT.git
+cd privateGPT
+pip install -r requirements.txt
+```
+
+Rename `example.env` to `.env` and edit the variables appropriately.
+
+This is an example `.env` file for LocalAI:
+
+```
+PERSIST_DIRECTORY=db
+# Set to OpenAI here
+MODEL_TYPE=OpenAI
+EMBEDDINGS_MODEL_NAME=all-MiniLM-L6-v2
+MODEL_N_CTX=1000
+# LocalAI URL
+OPENAI_API_BASE=http://localhost:8080/v1
+```

examples/query_data/.gitignore

@@ -0,0 +1 @@
+storage/

examples/query_data/README.md

@@ -0,0 +1,69 @@
+# Data query example
+
+This example makes use of [Llama-Index](https://gpt-index.readthedocs.io/en/stable/getting_started/installation.html) to enable question answering on a set of documents.
+
+It loosely follows [the quickstart](https://gpt-index.readthedocs.io/en/stable/guides/primer/usage_pattern.html).
+
+Summary of the steps:
+
+- prepare the dataset (and store it into `data`)
+- prepare a vector index database to run queries on
+- run queries
+
+## Requirements
+
+You will need a training data set. Copy that over `data`.
+
+## Setup
+
+Start the API:
+
+```bash
+# Clone LocalAI
+git clone https://github.com/go-skynet/LocalAI
+
+cd LocalAI/examples/query_data
+
+wget https://huggingface.co/skeskinen/ggml/resolve/main/all-MiniLM-L6-v2/ggml-model-q4_0.bin -O models/bert
+wget https://gpt4all.io/models/ggml-gpt4all-j.bin -O models/ggml-gpt4all-j
+
+# start with docker-compose
+docker-compose up -d --build
+```
+
+### Create a storage
+
+In this step we will create a local vector database from our document set, so later we can ask questions on it with the LLM.
+
+Note: **OPENAI_API_KEY** is not required. However the library might fail if no API_KEY is passed by, so an arbitrary string can be used.
+
+```bash
+export OPENAI_API_BASE=http://localhost:8080/v1
+export OPENAI_API_KEY=sk-
+
+python store.py
+```
+
+After it finishes, a directory "storage" will be created with the vector index database.
+
+## Query
+
+We can now query the dataset. 
+
+```bash
+export OPENAI_API_BASE=http://localhost:8080/v1
+export OPENAI_API_KEY=sk-
+
+python query.py
+```
+
+## Update
+
+To update our vector database, run `update.py`
+
+```bash
+export OPENAI_API_BASE=http://localhost:8080/v1
+export OPENAI_API_KEY=sk-
+
+python update.py
+```

examples/query_data/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/query_data/models/completion.tmpl

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

examples/query_data/models/embeddings.yaml

@@ -0,0 +1,6 @@
+name: text-embedding-ada-002
+parameters:
+  model: bert
+threads: 14
+backend: bert-embeddings
+embeddings: true

examples/query_data/models/gpt-3.5-turbo.yaml

@@ -0,0 +1,16 @@
+name: gpt-3.5-turbo
+parameters:
+  model: ggml-gpt4all-j
+  top_k: 80
+  temperature: 0.2
+  top_p: 0.7
+context_size: 1024
+stopwords:
+- "HUMAN:"
+- "GPT:"
+roles:
+  user: " "
+  system: " "
+template:
+  completion: completion
+  chat: gpt4all

examples/query_data/query.py

@@ -0,0 +1,35 @@
+import os
+
+# Uncomment to specify your OpenAI API key here (local testing only, not in production!), or add corresponding environment variable (recommended)
+# os.environ['OPENAI_API_KEY']= ""
+
+from llama_index import   LLMPredictor, PromptHelper, ServiceContext
+from langchain.llms.openai import OpenAI
+from llama_index import StorageContext, load_index_from_storage
+
+base_path = os.environ.get('OPENAI_API_BASE', 'http://localhost:8080/v1')
+
+# This example uses text-davinci-003 by default; feel free to change if desired
+llm_predictor = LLMPredictor(llm=OpenAI(temperature=0, model_name="gpt-3.5-turbo", openai_api_base=base_path))
+
+# Configure prompt parameters and initialise helper
+max_input_size = 500
+num_output = 256
+max_chunk_overlap = 20
+
+prompt_helper = PromptHelper(max_input_size, num_output, max_chunk_overlap)
+
+# Load documents from the 'data' directory
+service_context = ServiceContext.from_defaults(llm_predictor=llm_predictor, prompt_helper=prompt_helper)
+
+# rebuild storage context
+storage_context = StorageContext.from_defaults(persist_dir='./storage')
+
+# load index
+index = load_index_from_storage(storage_context,     service_context=service_context,    )
+
+query_engine = index.as_query_engine()
+
+data = input("Question: ")
+response = query_engine.query(data)
+print(response)

examples/query_data/store.py

@@ -0,0 +1,27 @@
+import os
+
+# Uncomment to specify your OpenAI API key here (local testing only, not in production!), or add corresponding environment variable (recommended)
+# os.environ['OPENAI_API_KEY']= ""
+
+from llama_index import GPTVectorStoreIndex, SimpleDirectoryReader, LLMPredictor, PromptHelper, ServiceContext
+from langchain.llms.openai import OpenAI
+from llama_index import StorageContext, load_index_from_storage
+
+base_path = os.environ.get('OPENAI_API_BASE', 'http://localhost:8080/v1')
+
+# This example uses text-davinci-003 by default; feel free to change if desired
+llm_predictor = LLMPredictor(llm=OpenAI(temperature=0, model_name="gpt-3.5-turbo", openai_api_base=base_path))
+
+# Configure prompt parameters and initialise helper
+max_input_size = 400
+num_output = 400
+max_chunk_overlap = 30
+
+prompt_helper = PromptHelper(max_input_size, num_output, max_chunk_overlap)
+
+# Load documents from the 'data' directory
+documents = SimpleDirectoryReader('data').load_data()
+service_context = ServiceContext.from_defaults(llm_predictor=llm_predictor, prompt_helper=prompt_helper, chunk_size_limit = 400)
+index = GPTVectorStoreIndex.from_documents(documents, service_context=service_context)
+index.storage_context.persist(persist_dir="./storage")
+

examples/query_data/update.py

@@ -0,0 +1,32 @@
+import os
+
+# Uncomment to specify your OpenAI API key here (local testing only, not in production!), or add corresponding environment variable (recommended)
+# os.environ['OPENAI_API_KEY']= ""
+
+from llama_index import   LLMPredictor, PromptHelper, SimpleDirectoryReader, ServiceContext
+from langchain.llms.openai import OpenAI
+from llama_index import StorageContext, load_index_from_storage
+
+base_path = os.environ.get('OPENAI_API_BASE', 'http://localhost:8080/v1')
+
+# This example uses text-davinci-003 by default; feel free to change if desired
+llm_predictor = LLMPredictor(llm=OpenAI(temperature=0, model_name="gpt-3.5-turbo", openai_api_base=base_path))
+
+# Configure prompt parameters and initialise helper
+max_input_size = 512
+num_output = 256
+max_chunk_overlap = 20
+
+prompt_helper = PromptHelper(max_input_size, num_output, max_chunk_overlap)
+
+# Load documents from the 'data' directory
+service_context = ServiceContext.from_defaults(llm_predictor=llm_predictor, prompt_helper=prompt_helper)
+
+# rebuild storage context
+storage_context = StorageContext.from_defaults(persist_dir='./storage')
+
+# load index
+index = load_index_from_storage(storage_context,     service_context=service_context,    )
+documents = SimpleDirectoryReader('data').load_data()
+index.refresh(documents)
+index.storage_context.persist(persist_dir="./storage")

examples/rwkv/.gitignore

@@ -0,0 +1,2 @@
+models/rwkv
+models/rwkv.tokenizer.json

examples/rwkv/Dockerfile.build

@@ -0,0 +1,12 @@
+FROM python
+
+RUN apt-get update && apt-get -y install cmake
+
+# convert the model (one-off)
+RUN pip3 install torch numpy
+
+WORKDIR /build
+COPY ./scripts/ .
+
+RUN git clone --recurse-submodules https://github.com/saharNooby/rwkv.cpp && cd rwkv.cpp && cmake . && cmake --build . --config Release
+ENTRYPOINT [ "/build/build.sh" ]

examples/rwkv/README.md

@@ -0,0 +1,59 @@
+# rwkv
+
+Example of how to run rwkv models.
+
+## Run models
+
+Setup:
+
+```bash
+# Clone LocalAI
+git clone https://github.com/go-skynet/LocalAI
+
+cd LocalAI/examples/rwkv
+
+# (optional) Checkout a specific LocalAI tag
+# git checkout -b build <TAG>
+
+# build the tooling image to convert an rwkv model locally:
+docker build -t rwkv-converter -f Dockerfile.build .
+
+# download and convert a model (one-off) - it's going to be fast on CPU too!
+docker run -ti --name converter -v $PWD:/data rwkv-converter https://huggingface.co/BlinkDL/rwkv-4-raven/resolve/main/RWKV-4-Raven-1B5-v11-Eng99%25-Other1%25-20230425-ctx4096.pth /data/models/rwkv
+
+# Get the tokenizer
+wget https://raw.githubusercontent.com/saharNooby/rwkv.cpp/5eb8f09c146ea8124633ab041d9ea0b1f1db4459/rwkv/20B_tokenizer.json -O models/rwkv.tokenizer.json
+
+# start with docker-compose
+docker-compose up -d --build
+```
+
+Test it out:
+
+```bash
+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 away",
+    "max_tokens": 100,
+    "temperature": 0.9, "top_p": 0.8, "top_k": 80
+  }'
+
+# {"object":"text_completion","model":"gpt-3.5-turbo","choices":[{"text":", there was a small group of five friends: Annie, Bryan, Charlie, Emily, and Jesse."}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}
+
+curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
+     "model": "gpt-3.5-turbo",            
+     "messages": [{"role": "user", "content": "How are you?"}],
+     "temperature": 0.9, "top_p": 0.8, "top_k": 80
+   }'
+
+# {"object":"chat.completion","model":"gpt-3.5-turbo","choices":[{"message":{"role":"assistant","content":" Good, thanks. I am about to go to bed. I' ll talk to you later.Bye."}}],"usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}
+```
+
+### Fine tuning
+
+See [RWKV-LM](https://github.com/BlinkDL/RWKV-LM#training--fine-tuning). There is also a Google [colab](https://colab.research.google.com/github/resloved/RWKV-notebooks/blob/master/RWKV_v4_RNN_Pile_Fine_Tuning.ipynb).
+
+## See also
+
+- [RWKV-LM](https://github.com/BlinkDL/RWKV-LM)
+- [rwkv.cpp](https://github.com/saharNooby/rwkv.cpp)

examples/rwkv/docker-compose.yaml

@@ -0,0 +1,16 @@
+version: '3.6'
+
+services:
+  api:
+    image: quay.io/go-skynet/local-ai:latest
+    build:
+      context: ../../
+      dockerfile: Dockerfile.dev
+    ports:
+      - 8080:8080
+    environment:
+      - DEBUG=true
+      - MODELS_PATH=/models
+    volumes:
+      - ./models:/models:cached
+    command: ["/usr/bin/local-ai" ]

examples/rwkv/models/gpt-3.5-turbo.yaml

@@ -0,0 +1,18 @@
+name: gpt-3.5-turbo
+parameters:
+  model: rwkv
+  top_k: 80
+  temperature: 0.9
+  max_tokens: 100
+  top_p: 0.8
+context_size: 1024
+backend: "rwkv"
+cutwords:
+- "Bob:.*"
+roles:
+  user: "Bob:"
+  system: "Alice:"
+  assistant: "Alice:"
+template:
+  completion: rwkv_completion
+  chat: rwkv_chat

examples/rwkv/models/rwkv_chat.tmpl

@@ -0,0 +1,13 @@
+The following is a verbose detailed conversation between Bob and a woman, Alice. Alice is intelligent, friendly and likeable. Alice is likely to agree with Bob.
+
+Bob: Hello Alice, how are you doing?
+
+Alice: Hi Bob! Thanks, I'm fine. What about you?
+
+Bob: I am very good! It's nice to see you. Would you mind me chatting with you for a while?
+
+Alice: Not at all! I'm listening.
+
+{{.Input}}
+
+Alice: 

examples/rwkv/models/rwkv_completion.tmpl

@@ -0,0 +1 @@
+Complete the following sentence: {{.Input}}