PacmanAI-2 / main.py
Marroco93's picture
json import
0f34bf3
raw
history blame
2.06 kB
from fastapi import FastAPI
from fastapi.responses import StreamingResponse
from pydantic import BaseModel
from huggingface_hub import InferenceClient
import uvicorn
from typing import Generator
import json # Asegúrate de que esta línea esté al principio del archivo
app = FastAPI()
# Initialize the InferenceClient with your model
client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
class Item(BaseModel):
prompt: str
history: list
system_prompt: str
temperature: float = 0.8
max_new_tokens: int = 9000
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
def generate_stream(item: Item) -> Generator[bytes, None, None]:
formatted_prompt = format_prompt(f"{item.system_prompt}, {item.prompt}", item.history)
generate_kwargs = {
"temperature": item.temperature,
"max_new_tokens": item.max_new_tokens,
"top_p": item.top_p,
"repetition_penalty": item.repetition_penalty,
"do_sample": True,
"seed": 42, # Adjust or omit the seed as needed
}
# Stream the response from the InferenceClient
for response in client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True):
# This assumes 'details=True' gives you a structure where you can access the text like this
chunk = {
"text": response.token.text,
"complete": response.generated_text is not None # Adjust based on how you detect completion
}
yield json.dumps(chunk).encode("utf-8") + b"\n"
@app.post("/generate/")
async def generate_text(item: Item):
# Stream response back to the client
return StreamingResponse(generate_stream(item), media_type="application/x-ndjson")
if __name__ == "__main__":
uvicorn.run(app, host="0.0.0.0", port=8000)