main.py

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

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

class Item(BaseModel):
    prompt: str
    history: list = []
    idioma: str = "en"
    temperature: float = 0.7
    max_new_tokens: int = 900
    top_p: float = 0.8
    repetition_penalty: float = 1.2

def generate(item: Item, max_tokens: int, file_name: str, temperature: float, top_p: float, repetition_penalty: float):
    print("request")
    if file_name == "Scripter":
        if item.idioma == "br":
            with open("Scripter2.txt", 'r') as file:
                system_prompt = file.read()
        else:
            with open("Scripter.txt", 'r') as file:
                system_prompt = file.read()
    else:
        with open(file_name, 'r') as file:
            system_prompt = file.read()
        
    # Garantir que todos os itens em item.history sejam strings
    history_str = "\n".join(str(entry) for entry in item.history)
    full_prompt = f"<s>[SYSTEM]{system_prompt}{history_str}\n{item.prompt} [MIA]"
    
    stream = client.text_generation(full_prompt,
        temperature=temperature,
        max_new_tokens=max_tokens,
        top_p=top_p,
        repetition_penalty=repetition_penalty,
        do_sample=True,
        seed=random.randint(1, 1000),
        stream=True,
        details=True,
        return_full_text=False
    )
    for response in stream:
        yield response.token.text.encode('utf-8')

# Endpoint para 'init_en/' com parâmetros ajustados
@app.post("/initen/")
async def generate1(item: Item):
    # Valores específicos para o endpoint 'initen'
    temperature = 0.5
    max_tokens = 170
    top_p = 0.9
    repetition_penalty = 1.2
    return StreamingResponse(generate(item, max_tokens=max_tokens, file_name='init_en.txt', 
                                      temperature=temperature, top_p=top_p, repetition_penalty=repetition_penalty),
                            media_type="application/octet-stream")

# Endpoint para 'initbr/' com parâmetros ajustados
@app.post("/initbr/")
async def generate1(item: Item):
    # Valores específicos para o endpoint 'initbr'
    temperature = 0.5
    max_tokens = 170
    top_p = 0.9
    repetition_penalty = 1.2
    return StreamingResponse(generate(item, max_tokens=max_tokens, file_name='init_br.txt', 
                                      temperature=temperature, top_p=top_p, repetition_penalty=repetition_penalty),
                            media_type="application/octet-stream")




@app.post("/generate1/")
async def generate1(item: Item):
    return StreamingResponse(generate(item, max_tokens=350, file_name='Mia.txt', 
                                      temperature=item.temperature, top_p=item.top_p, 
                                      repetition_penalty=item.repetition_penalty),
                            media_type="application/octet-stream")

@app.post("/generate2/")
async def generate2(item: Item):
    return StreamingResponse(generate(item, max_tokens=350, file_name='Mia.txt', 
                                      temperature=item.temperature, top_p=item.top_p, 
                                      repetition_penalty=item.repetition_penalty),
                            media_type="application/octet-stream")

@app.post("/generate3/")
async def generate3_br(item: Item):
    return StreamingResponse(generate(item, max_tokens=600, file_name='Scripter', 
                                      temperature=item.temperature, top_p=item.top_p, 
                                      repetition_penalty=item.repetition_penalty),
                            media_type="application/octet-stream")

if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=8000)