main.py

from fastapi import FastAPI
from pydantic import BaseModel
from huggingface_hub import InferenceClient
from fastapi.responses import StreamingResponse

app = FastAPI()

client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")

class Item(BaseModel):
    prompt: str
    history: list
    system_prompt: str
    temperature: float = 0.0
    max_new_tokens: int = 1048
    top_p: float = 0.15
    repetition_penalty: float = 1.0

def format_prompt(message, history):
    prompt = "<s>"
    for user_prompt, bot_response in history:
        prompt += f"[INST] {user_prompt} [/INST]"
        prompt += f" {bot_response}</s> "
    prompt += f"[INST] {message} [/INST]"
    return prompt

async def generate_stream(item: Item):
    try:
        temperature = max(float(item.temperature), 1e-2)
        top_p = float(item.top_p)

        generate_kwargs = dict(
            temperature=temperature,
            max_new_tokens=item.max_new_tokens,
            top_p=top_p,
            repetition_penalty=item.repetition_penalty,
            do_sample=True,
            seed=42,
        )

        formatted_prompt = format_prompt(f"{item.system_prompt}, {item.prompt}", item.history)
        stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)

        for response in stream:
            yield response.token.text
    except Exception as e:
        print(f"Error in generate_stream: {e}")
    finally:
        if 'stream' in locals():
            stream.close()

@app.post("/generate/")
async def generate_text(item: Item):
    try:
        return StreamingResponse(generate_stream(item), media_type="text/plain")
    except Exception as e:
        print(f"Error in generate_text: {e}")
        return {"error": str(e)}