app.py

import torch
from fastapi import FastAPI, Request
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from transformers import (
    MBartForConditionalGeneration, MBart50Tokenizer,
    MT5ForConditionalGeneration, T5Tokenizer
)
from peft import PeftModel
import warnings
from dotenv import load_dotenv

# ================== Config ==================
load_dotenv()
warnings.filterwarnings("ignore", category=FutureWarning)

app = FastAPI(title="Khmer Summarization API")

# Allow CORS for JS frontend
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# ================== Models ==================
MODELS = {
    "model1": {
        "name": "Model 1 - Khmer MBart Summarization (LoRA)",
        "repo": "sedtha/mBart-50-large_LoRa_kh_sumerize",
        "type": "mbart_lora",
        "model": None,
        "tokenizer": None
    },
    "model2": {
        "name": "Model 2 - Khmer mT5 Summarization",
        "repo": "angkor96/khmer-mT5-news-summarization",
        "type": "mt5",
        "model": None,
        "tokenizer": None
    }
}

# ================== Load Model ==================
def load_model(model_key):
    model_info = MODELS[model_key]

    if model_info["model"] is None:
        print(f"🔄 Loading {model_info['name']} ...")

        if model_info["type"] == "mbart_lora":
            tokenizer = MBart50Tokenizer.from_pretrained(
                model_info["repo"], src_lang="km_KH", tgt_lang="km_KH"
            )
            base = MBartForConditionalGeneration.from_pretrained("facebook/mbart-large-50").to(device)
            model = PeftModel.from_pretrained(base, model_info["repo"]).merge_and_unload().to(device)

        elif model_info["type"] == "mt5":
            tokenizer = T5Tokenizer.from_pretrained(model_info["repo"])
            model = MT5ForConditionalGeneration.from_pretrained(model_info["repo"]).to(device)

        else:
            raise ValueError("Unknown model type")

        model.eval()
        model_info["tokenizer"] = tokenizer
        model_info["model"] = model

        print(f"✅ Loaded {model_info['name']}")

    return model_info["model"], model_info["tokenizer"]


# ================== Request Model ==================
class SummarizeRequest(BaseModel):
    text: str
    models: list[str] = ["model1"]


# ================== Summarization ==================
def summarize_text(text, model_key):
    model, tokenizer = load_model(model_key)

    inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=1024).to(device)

    with torch.no_grad():
        output_ids = model.generate(
            **inputs,
            max_new_tokens=300,
            num_beams=4,
            early_stopping=True
        )

    return tokenizer.decode(output_ids[0], skip_special_tokens=True).strip()


# ================== Endpoints ==================
@app.get("/")
def root():
    return {"message": "✅ Khmer Summarization API (FastAPI) Running!"}

@app.get("/models")
def list_models():
    return {key: {"name": v["name"]} for key, v in MODELS.items()}

@app.post("/summarize")
async def summarize(req: SummarizeRequest):
    if not req.text.strip():
        return {"error": "⚠️ សូមវាយបញ្ចូលអត្ថបទជាមុន!"}

    results = {}
    for key in req.models:
        if key in MODELS:
            try:
                summary = summarize_text(req.text, key)
            except Exception as e:
                summary = f"Error: {str(e)}"
            results[key] = {
                "name": MODELS[key]["name"],
                "summary": summary
            }

    return {"results": results}