main.py

from fastapi import FastAPI
from pydantic import BaseModel
from huggingface_hub import InferenceClient
import uvicorn
import os
from transformers import (
    AutoModelForCausalLM,
    AutoTokenizer,
    TextIteratorStreamer,
)

app = FastAPI()

HF_TOKEN = os.environ["HF_TOKEN"]

model = AutoModelForCausalLM.from_pretrained("microsoft/Phi-3-mini-128k-instruct",
                                            device_map="auto",
                                            token=HF_TOKEN
                                            )
tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-128k-instruct", trust_remote_code=True, token=HF_TOKEN)

class Item(BaseModel):
    prompt: str
    system_prompt: str
    temperature: float = 0.2
    max_new_tokens: int = 1048
    top_p: float = 0.15
    repetition_penalty: float = 1.0

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

    device = "cpu"
    
    prompt = item.prompt

    encodeds = tokenizer(prompt, return_tensors="pt", add_special_tokens=True)

    model_inputs = encodeds.to(device)


    generated_ids = model.generate(**model_inputs, max_new_tokens=item.max_new_tokens, do_sample=True, pad_token_id=tokenizer.eos_token_id, temperature=item.temperature)
    decoded = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
    return decoded


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