api.py

import fastapi
import json
import uvicorn
from fastapi import HTTPException
from fastapi.responses import HTMLResponse
from fastapi.middleware.cors import CORSMiddleware
from sse_starlette.sse import EventSourceResponse
from starlette.responses import StreamingResponse
from ctransformers import AutoModelForCausalLM
from pydantic import BaseModel
from typing import List, Dict, Any, Generator

llm = AutoModelForCausalLM.from_pretrained("TheBloke/falcon-40b-instruct-GGML", model_file="falcon40b-instruct.ggmlv3.q2_K.bin",
                                           model_type="falcon", threads=8)
app = fastapi.FastAPI(title="🦅Falcon 40B GGML (ggmlv3.q2_K)🦅")
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

class ChatCompletionRequestV0(BaseModel):
    prompt: str

class Message(BaseModel):
    role: str
    content: str

class ChatCompletionRequest(BaseModel):
    messages: List[Message]
    max_tokens: int = 250

@app.post("/v1/completions")
async def completion(request: ChatCompletionRequestV0, response_mode=None):
    response = llm(request.prompt)
    return response

@app.post("/v1/chat/completions")
async def chat(request: ChatCompletionRequest):
    combined_messages = ' '.join([message.content for message in request.messages])
    tokens = llm.tokenize(combined_messages)
    
    try:
        chat_chunks = llm.generate(tokens)
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

    async def format_response(chat_chunks: Generator) -> Any:
        for chat_chunk in chat_chunks:
            response = {
                'choices': [
                    {
                        'message': {
                            'role': 'system',
                            'content': llm.detokenize(chat_chunk)
                        },
                        'finish_reason': 'stop' if llm.detokenize(chat_chunk) == "[DONE]" else 'unknown'
                    }
                ]
            }
            yield f"data: {json.dumps(response)}\n\n"
        yield "event: done\ndata: {}\n\n"

    return StreamingResponse(format_response(chat_chunks), media_type="text/event-stream")

@app.post("/v0/chat/completions")
async def chat(request: ChatCompletionRequestV0, response_mode=None):
    tokens = llm.tokenize(request.prompt)
    async def server_sent_events(chat_chunks, llm):
        for chat_chunk in llm.generate(chat_chunks):
            yield dict(data=json.dumps(llm.detokenize(chat_chunk)))
        yield dict(data="[DONE]")

    return EventSourceResponse(server_sent_events(tokens, llm))

if __name__ == "__main__":
  uvicorn.run(app, host="0.0.0.0", port=7860)