app.py

from fastapi import FastAPI
from pydantic import BaseModel
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

app = FastAPI()

MODEL_ID = "nairanu6115/phi3-mini"

print("Loading tokenizer...")
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)

print("Loading model...")
model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID,
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
    device_map="auto"
)
model.eval()
print("Model ready.")


class Request(BaseModel):
    prompt: str
    max_tokens: int = 300


@app.get("/")
def health():
    return {"status": "ok", "model": MODEL_ID}


@app.post("/generate")
def generate(req: Request):
    # Phi-3 chat template
    formatted = f"<|user|>\n{req.prompt}<|end|>\n<|assistant|>\n"

    inputs = tokenizer(formatted, return_tensors="pt").to(model.device)

    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=req.max_tokens,
            temperature=0.7,
            do_sample=True,
            pad_token_id=tokenizer.eos_token_id
        )

    full_text = tokenizer.decode(outputs[0], skip_special_tokens=True)

    # Extract only the assistant's reply
    response = full_text.split("<|assistant|>")[-1].strip() if "<|assistant|>" in full_text else full_text
    response = response.replace("<|end|>", "").replace("<|user|>", "").strip()

    return {"response": response}