app.py

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

logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

app = FastAPI()

model_name = "google/gemma-2-2b-it"
tokenizer = None
model = None

try:
    logger.info(f"Loading model: {model_name}")
    tokenizer = AutoTokenizer.from_pretrained(model_name, token=os.getenv("HF_TOKEN"))
    model = AutoModelForCausalLM.from_pretrained(
        model_name,
        torch_dtype=torch.float16,  # メモリ削減
        device_map="cpu",  # GPU利用不可
        token=os.getenv("HF_TOKEN"),
        low_cpu_mem_usage=True
    )
    logger.info("Model loaded successfully")
except Exception as e:
    logger.error(f"Model load error: {e}")
    raise

class TextInput(BaseModel):
    text: str
    max_length: int = 50

@app.post("/generate")
async def generate_text(input: TextInput):
    try:
        logger.info(f"Generating text for input: {input.text}")
        inputs = tokenizer(input.text, return_tensors="pt", max_length=512, truncation=True).to("cpu")
        outputs = model.generate(**inputs, max_length=input.max_length)
        result = tokenizer.decode(outputs[0], skip_special_tokens=True)
        logger.info(f"Generated text: {result}")
        return {"generated_text": result}
    except Exception as e:
        logger.error(f"Generation error: {e}")
        return {"error": str(e)}