main.py

from fastapi import FastAPI
from pydantic import BaseModel
from huggingface_hub import InferenceClient
import uvicorn
import prompt_style
import os


model_id = "failspy/Meta-Llama-3-8B-Instruct-abliterated-v3"
client = InferenceClient(token=os.getenv('HF_TOKEN'), model=model_id)

class Item(BaseModel):
    prompt: str
    history: list
    system_prompt: str
    token:str
    temperature: float = 0.6
    max_new_tokens: int = 1024
    top_p: float = 0.95
    repetition_penalty: float = 1.0
    seed : int = 42
    
app = FastAPI()

def format_prompt(item: Item):
    messages = [
        {"role": "system", "content": prompt_style.data},
    ]
    for it in item.history:
        messages.append[{"role" : "user", "content": it[0]}]
        messages.append[{"role" : "assistant", "content": it[1]}]
    return messages

def generate(item: Item):
    temperature = float(item.temperature)
    if temperature < 1e-2:
        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=item.seed,
    )

    formatted_prompt = format_prompt(item)
    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
    output = ""

    for response in stream:
        output += response.token.text
    return output

@app.post("/generate/")
async def generate_text(item: Item):
    ans = generate(item)
    return {"response": ans}


@app.get("/")
def read_root():
    return {"Hello": "World!"}