main.py

from fastapi import FastAPI, Request, Body
from huggingface_hub import InferenceClient
import random

API_URL = "https://api-inference.huggingface.co/models/"

client = InferenceClient(
    "mistralai/Mistral-7B-Instruct-v0.1"
)

app = FastAPI()

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


@app.post("/api/v1/generate_text")
def generate_text(request: Request, prompt: str = Body()):
    history = []  # You might need to handle this based on your actual usage
    print(f"request + {request}")
    temperature = request.headers.get("temperature", 0.5)
    # print(f"temperature + {temperature}")
    top_p = request.headers.get("top_p", 0.95)
    # print(f"top_p + {top_p}")
    repetition_penalty = request.headers.get("repetition_penalty", 1.0)
    # print(f"repetition_penalty + {repetition_penalty}")

    formatted_prompt = format_prompt(prompt, history)
    print(f"formatted_prompt + {formatted_prompt}")
    stream = client.text_generation(
        formatted_prompt,
        temperature=temperature,
        max_new_tokens=512,
        top_p=top_p,
        repetition_penalty=repetition_penalty,
        do_sample=True,
        seed=random.randint(0, 10**7),
        stream=False,
        details=True, 
        return_full_text=True
    )
    # output = ""

    # for response in stream:
    #     output += response.token.text
    #     yield output
        
    # return output[len(output) - 1]

    return stream