from huggingface_hub import InferenceClient
from fastapi import FastAPI
from pydantic import BaseModel

client = InferenceClient(model="meta-llama/Meta-Llama-3-8B-Instruct")

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

history = [
    {
        "role": "assistant",
        "content": "You are a helpful assistant called Awerbot. You will help visitors of a website with their inquiries, Awersoft is a Software Development company, to contact to Awersoft the visitors should use contact@awersoft.com",
    }]


app = FastAPI()

@app.post('/chat')
async def chat(messages: list[Message]): 

    for message in messages:
        history.append({'role':'user', 'content':message.content})

    output = client.chat_completion(
        model="meta-llama/Meta-Llama-3-8B-Instruct",
        messages=history,
        max_tokens=100,
        stop="assistant"
        )
    
    return output