GPT4All API Scaffolding. Matches OpenAI OpenAPI spec for chats and completions (#839)

* GPT4All API Scaffolding. Matches OpenAI OpenAI spec for engines, chats and completions

* Edits for docker building

* FastAPI app builds and pydantic models are accurate

* Added groovy download into dockerfile

* improved dockerfile

* Chat completions endpoint edits

* API uni test sketch

* Working example of groovy inference with open ai api

* Added lines to test

* Set default to mpt

gpt4all-api/.gitignore

@@ -0,0 +1,112 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+app/__pycache__/
+gpt4all_api/__pycache__/
+gpt4all_api/app/api_v1/__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# VS Code
+.vscode/
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+.hypothesis/
+.pytest_cache/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# pyenv
+.python-version
+
+# celery beat schedule file
+celerybeat-schedule
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+
+*.lock
+*.cache

gpt4all-api/LICENSE

@@ -0,0 +1,13 @@
+Copyright 2023 Nomic, Inc.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.

gpt4all-api/README.md

@@ -1,2 +1,47 @@
-# GPT4All API
+# WIP - not yet ready.
-This directory will contain code to build out a RESTful API for GPT4All models. Exact details TBD, but as an MVP, user should be able to send requests to list, download, and generate text with different models.
+
+# GPT4All REST API
+This directory contains the source code to run and build docker images that run a FastAPI app
+for serving inference from GPT4All models. The API matches the OpenAI API spec.
+
+## Tutorial
+
+### Starting the app
+
+First build the FastAPI docker image. You only have to do this on initial build or when you add new dependencies to the requirements.txt file:
+```bash
+DOCKER_BUILDKIT=1 docker build -t gpt4all_api --progress plain -f gpt4all_api/Dockerfile.buildkit .
+```
+
+Then, start the backend with:
+
+```bash
+docker compose up --build
+```
+
+#### Spinning up your app
+Run `docker compose up` to spin up the backend. Monitor the logs for errors in-case you forgot to set an environment variable above.
+
+
+#### Development
+Run
+
+```bash
+docker compose up --build
+```
+and edit files in the `api` directory. The api will hot-reload on changes.
+
+You can run the unit tests with
+
+```bash
+make test
+```
+
+#### Viewing API documentation
+
+Once the FastAPI ap is started you can access its documentation and test the search endpoint by going to:
+```
+localhost:80/docs
+```
+
+This documentation should match the OpenAI OpenAPI spec located at https://github.com/openai/openai-openapi/blob/master/openapi.yaml

gpt4all-api/docker-compose.yaml

@@ -0,0 +1,18 @@
+version: "3.5"
+
+services:
+  gpt4all_api:
+    image: gpt4all_api
+    container_name: gpt4all_api
+    restart: always #restart on error (usually code compilation from save during bad state)
+    ports:
+      - "4891:4891"
+    environment:
+      - APP_ENVIRONMENT=dev
+      - WEB_CONCURRENCY=2
+      - LOGLEVEL=debug
+      - PORT=4891
+      - model=ggml-mpt-7b-chat.bin
+    volumes:
+      - './gpt4all_api/app:/app'
+    command: ["/start-reload.sh"]

gpt4all-api/gpt4all_api/Dockerfile.buildkit

@@ -0,0 +1,23 @@
+# syntax=docker/dockerfile:1.0.0-experimental
+FROM tiangolo/uvicorn-gunicorn:python3.11
+
+ARG MODEL_BIN=ggml-mpt-7b-chat.bin
+
+# Put first so anytime this file changes other cached layers are invalidated.
+COPY gpt4all_api/requirements.txt /requirements.txt
+
+RUN pip install --upgrade pip
+
+# Run various pip install commands with ssh keys from host machine.
+RUN --mount=type=ssh pip install -r /requirements.txt && \
+  rm -Rf /root/.cache && rm -Rf /tmp/pip-install*
+
+# Finally, copy app and client.
+COPY gpt4all_api/app /app
+
+RUN mkdir -p /models
+
+# Include the following line to bake a model into the image and not have to download it on API start.
+RUN wget -q --show-progress=off https://gpt4all.io/models/${MODEL_BIN} -P /models \
+  && md5sum /models/${MODEL_BIN}
+

gpt4all-api/gpt4all_api/README.md

@@ -0,0 +1 @@
+# FastAPI app for serving GPT4All models

gpt4all-api/gpt4all_api/app/api_v1/api.py

@@ -0,0 +1,8 @@
+from api_v1.routes import chat, completions, engines
+from fastapi import APIRouter
+
+router = APIRouter()
+
+router.include_router(chat.router)
+router.include_router(completions.router)
+router.include_router(engines.router)

gpt4all-api/gpt4all_api/app/api_v1/events.py

@@ -0,0 +1,26 @@
+import logging
+from fastapi import HTTPException
+from fastapi.responses import JSONResponse
+from starlette.requests import Request
+from api_v1.settings import settings
+log = logging.getLogger(__name__)
+
+
+startup_msg_fmt = """
+ Starting up GPT4All API
+"""
+
+
+async def on_http_error(request: Request, exc: HTTPException):
+    return JSONResponse({'detail': exc.detail}, status_code=exc.status_code)
+
+
+async def on_startup(app):
+    startup_msg = startup_msg_fmt.format(settings=settings)
+    log.info(startup_msg)
+
+def startup_event_handler(app):
+    async def start_app() -> None:
+        await on_startup(app)
+
+    return start_app

gpt4all-api/gpt4all_api/app/api_v1/routes/chat.py

@@ -0,0 +1,63 @@
+from fastapi import APIRouter, Depends, Response, Security, status
+from pydantic import BaseModel, Field
+from typing import List, Dict
+import logging
+import time
+from api_v1.settings import settings
+
+logger = logging.getLogger(__name__)
+logger.setLevel(logging.DEBUG)
+
+### This should follow https://github.com/openai/openai-openapi/blob/master/openapi.yaml
+
+
+
+class ChatCompletionMessage(BaseModel):
+    role: str
+    content: str
+
+class ChatCompletionRequest(BaseModel):
+    model: str = Field(..., description='The model to generate a completion from.')
+    messages: List[ChatCompletionMessage] = Field(..., description='The model to generate a completion from.')
+
+
+class ChatCompletionChoice(BaseModel):
+    message: ChatCompletionMessage
+    index: int
+    finish_reason: str
+
+class ChatCompletionUsage(BaseModel):
+    prompt_tokens: int
+    completion_tokens: int
+    total_tokens: int
+
+class ChatCompletionResponse(BaseModel):
+    id: str
+    object: str = 'text_completion'
+    created: int
+    model: str
+    choices: List[ChatCompletionChoice]
+    usage: ChatCompletionUsage
+
+
+router = APIRouter(prefix="/chat", tags=["Completions Endpoints"])
+
+@router.post("/completions", response_model=ChatCompletionResponse)
+async def chat_completion(request: ChatCompletionRequest):
+    '''
+    Completes a GPT4All model response.
+    '''
+
+    return ChatCompletionResponse(
+        id='asdf',
+        created=time.time(),
+        model=request.model,
+        choices=[{}],
+        usage={
+            'prompt_tokens': 0,
+            'completion_tokens': 0,
+            'total_tokens': 0
+        }
+    )
+
+

gpt4all-api/gpt4all_api/app/api_v1/routes/completions.py

@@ -0,0 +1,86 @@
+from fastapi import APIRouter, Depends, Response, Security, status
+from pydantic import BaseModel, Field
+from typing import List, Dict
+import logging
+from uuid import uuid4
+from api_v1.settings import settings
+from gpt4all import GPT4All
+import time
+
+logger = logging.getLogger(__name__)
+logger.setLevel(logging.DEBUG)
+
+### This should follow https://github.com/openai/openai-openapi/blob/master/openapi.yaml
+
+class CompletionRequest(BaseModel):
+    model: str = Field(..., description='The model to generate a completion from.')
+    prompt: str = Field(..., description='The prompt to begin completing from.')
+    max_tokens: int = Field(7, description='Max tokens to generate')
+    temperature: float = Field(0, description='Model temperature')
+    top_p: float = Field(1.0, description='top_p')
+    n: int = Field(1, description='')
+    stream: bool = Field(False, description='Stream responses')
+
+
+class CompletionChoice(BaseModel):
+    text: str
+    index: int
+    logprobs: float
+    finish_reason: str
+
+class CompletionUsage(BaseModel):
+    prompt_tokens: int
+    completion_tokens: int
+    total_tokens: int
+class CompletionResponse(BaseModel):
+    id: str
+    object: str = 'text_completion'
+    created: int
+    model: str
+    choices: List[CompletionChoice]
+    usage: CompletionUsage
+
+
+router = APIRouter(prefix="/completions", tags=["Completion Endpoints"])
+
+@router.post("/", response_model=CompletionResponse)
+async def completions(request: CompletionRequest):
+    '''
+    Completes a GPT4All model response.
+    '''
+
+    # global model
+    if request.stream:
+        raise NotImplementedError("Streaming is not yet implements")
+
+    model = GPT4All(model_name=settings.model, model_path=settings.gpt4all_path)
+
+    output = model.generate(prompt=request.prompt,
+                     n_predict = request.max_tokens,
+                     top_k = 20,
+                     top_p = request.top_p,
+                     temp=request.temperature,
+                     n_batch = 1024,
+                     repeat_penalty = 1.2,
+                     repeat_last_n = 10,
+                     context_erase = 0)
+
+
+    return CompletionResponse(
+        id=str(uuid4()),
+        created=time.time(),
+        model=request.model,
+        choices=[dict(CompletionChoice(
+            text=output,
+            index=0,
+            logprobs=-1,
+            finish_reason='stop'
+        ))],
+        usage={
+            'prompt_tokens': 0, #TODO how to compute this?
+            'completion_tokens': 0,
+            'total_tokens': 0
+        }
+    )
+
+

gpt4all-api/gpt4all_api/app/api_v1/routes/engines.py

@@ -0,0 +1,38 @@
+from fastapi import APIRouter, Depends, Response, Security, status
+from pydantic import BaseModel, Field
+from typing import List, Dict
+import logging
+from api_v1.settings import settings
+
+logger = logging.getLogger(__name__)
+logger.setLevel(logging.DEBUG)
+
+### This should follow https://github.com/openai/openai-openapi/blob/master/openapi.yaml
+
+class ListEnginesResponse(BaseModel):
+    data: List[Dict] = Field(..., description="All available models.")
+
+class EngineResponse(BaseModel):
+    data: List[Dict] = Field(..., description="All available models.")
+
+router = APIRouter(prefix="/engines", tags=["Search Endpoints"])
+
+@router.get("/", response_model=ListEnginesResponse)
+async def list_engines():
+    '''
+    List all available GPT4All models from
+    https://raw.githubusercontent.com/nomic-ai/gpt4all/main/gpt4all-chat/metadata/models.json
+    '''
+    raise NotImplementedError()
+    return ListEnginesResponse(data=[])
+
+
+@router.get("/{engine_id}", response_model=EngineResponse)
+async def retrieve_engine(engine_id: str):
+    '''
+
+    '''
+
+    raise NotImplementedError()
+    return EngineResponse()
+

gpt4all-api/gpt4all_api/app/api_v1/routes/health.py

@@ -0,0 +1,12 @@
+import logging
+from fastapi import APIRouter
+from fastapi.responses import JSONResponse
+log = logging.getLogger(__name__)
+
+router = APIRouter(prefix="/health", tags=["Health"])
+
+
+@router.get('/', response_class=JSONResponse)
+async def health_check():
+    """Runs a health check on this instance of the API."""
+    return JSONResponse({'status': 'ok'}, headers={'Access-Control-Allow-Origin': '*'})

gpt4all-api/gpt4all_api/app/api_v1/settings.py

@@ -0,0 +1,10 @@
+from pydantic import BaseSettings
+
+
+class Settings(BaseSettings):
+    app_environment = 'dev'
+    model: str = 'ggml-mpt-7b-chat.bin'
+    gpt4all_path: str = '/models'
+
+
+settings = Settings()

gpt4all-api/gpt4all_api/app/docs.py

@@ -0,0 +1,3 @@
+desc = 'GPT4All API'
+
+endpoint_paths = {'health': '/health'}

gpt4all-api/gpt4all_api/app/main.py

@@ -0,0 +1,61 @@
+import os
+import docs
+import logging
+from fastapi import FastAPI, HTTPException, Request
+from starlette.middleware.cors import CORSMiddleware
+from fastapi.logger import logger as fastapi_logger
+from api_v1.settings import settings
+from api_v1.api import router as v1_router
+from api_v1 import events
+import os
+
+logger = logging.getLogger(__name__)
+
+app = FastAPI(title='GPT4All API', description=docs.desc)
+
+#CORS Configuration (in-case you want to deploy)
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_credentials=True,
+    allow_methods=["GET", "POST", "OPTIONS"],
+    allow_headers=["*"],
+)
+
+logger.info('Adding v1 endpoints..')
+
+# add v1
+app.include_router(v1_router, prefix='/v1')
+app.add_event_handler('startup', events.startup_event_handler(app))
+app.add_exception_handler(HTTPException, events.on_http_error)
+
+@app.on_event("startup")
+async def startup():
+    global model
+    logger.info(f"Downloading/fetching model: {os.path.join(settings.gpt4all_path, settings.model)}")
+    from gpt4all import GPT4All
+    model = GPT4All(model_name=settings.model, model_path=settings.gpt4all_path)
+
+    logger.info("GPT4All API is ready.")
+
+@app.on_event("shutdown")
+async def shutdown():
+    logger.info("Shutting down API")
+
+
+# This is needed to get logs to show up in the app
+if "gunicorn" in os.environ.get("SERVER_SOFTWARE", ""):
+    gunicorn_error_logger = logging.getLogger("gunicorn.error")
+    gunicorn_logger = logging.getLogger("gunicorn")
+
+    root_logger = logging.getLogger()
+    fastapi_logger.setLevel(gunicorn_logger.level)
+    fastapi_logger.handlers = gunicorn_error_logger.handlers
+    root_logger.setLevel(gunicorn_logger.level)
+
+    uvicorn_logger = logging.getLogger("uvicorn.access")
+    uvicorn_logger.handlers = gunicorn_error_logger.handlers
+else:
+    # https://github.com/tiangolo/fastapi/issues/2019
+    LOG_FORMAT2 = "[%(asctime)s %(process)d:%(threadName)s] %(name)s - %(levelname)s - %(message)s | %(filename)s:%(lineno)d"
+    logging.basicConfig(level=logging.INFO, format=LOG_FORMAT2)

gpt4all-api/gpt4all_api/app/tests/test_endpoints.py

@@ -0,0 +1,35 @@
+"""
+Use the OpenAI python API to test gpt4all models.
+"""
+import openai
+openai.api_base = "http://localhost:4891/v1"
+
+openai.api_key = "not needed for a local LLM"
+
+
+def test_completion():
+    model = "gpt4all-j-v1.3-groovy"
+    prompt = "Who is Michael Jordan?"
+    response = openai.Completion.create(
+        model=model,
+        prompt=prompt,
+        max_tokens=50,
+        temperature=0.28,
+        top_p=0.95,
+        n=1,
+        echo=True,
+        stream=False
+    )
+    assert len(response['choices'][0]['text']) > len(prompt)
+    print(response)
+
+# def test_chat_completions():
+#     model = "gpt4all-j-v1.3-groovy"
+#     prompt = "Who is Michael Jordan?"
+#     response = openai.ChatCompletion.create(
+#         model=model,
+#         messages=[]
+#     )
+
+
+

gpt4all-api/gpt4all_api/requirements.txt

@@ -0,0 +1,10 @@
+aiohttp>=3.6.2
+aiofiles
+pydantic>=1.4.0
+requests>=2.24.0
+ujson>=2.0.2
+fastapi>=0.95.0
+Jinja2>=3.0
+gpt4all==0.2.3
+pytest
+openai

gpt4all-api/makefile

@@ -0,0 +1,37 @@
+ROOT_DIR:=$(shell dirname $(realpath $(lastword $(MAKEFILE_LIST))))
+APP_NAME:=gpt4all_api
+PYTHON:=python3.8
+
+all: dependencies
+
+fresh: clean dependencies
+
+testenv: clean_testenv test_build
+	docker compose up --build
+
+testenv_d: clean_testenv test_build
+	docker compose up --build -d
+
+test:
+	docker compose exec gpt4all_api pytest -svv --disable-warnings -p no:cacheprovider /app/tests
+
+test_build:
+    DOCKER_BUILDKIT=1 docker build -t gpt4all_api --progress plain -f gpt4all_api/Dockerfile.buildkit .
+
+clean_testenv:
+	docker compose down -v
+
+fresh_testenv: clean_testenv testenv
+
+venv:
+	if [ ! -d $(ROOT_DIR)/env ]; then $(PYTHON) -m venv $(ROOT_DIR)/env; fi
+
+dependencies: venv
+	source $(ROOT_DIR)/env/bin/activate; yes w | python -m pip install -r $(ROOT_DIR)/atlas_api/requirements.txt
+
+clean: clean_testenv
+	# Remove existing environment
+	rm -rf $(ROOT_DIR)/env;
+	rm -rf $(ROOT_DIR)/$(APP_NAME)/*.pyc;
+
+