from fastapi import FastAPI
from pydantic import BaseModel
from fastapi.middleware.cors import CORSMiddleware
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
from utils import clean_text

MODEL_ID = "hiuS04/phobert-sentiment-api"
LABELS = ["Negative", "Neutral", "Positive"]

# Load model
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, use_fast=False)
model = AutoModelForSequenceClassification.from_pretrained(MODEL_ID)
model.eval()

app = FastAPI(title="PhoBERT Sentiment API")

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],  # cho phép extension gọi
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

class TextInput(BaseModel):
    text: str

@app.get("/")
def root():
    return {"message": "PhoBERT Sentiment API is live on Hugging Face"}

@app.post("/predict")
def predict(inp: TextInput):
    raw_text = inp.text.strip()
    if not raw_text:
        return {"error": "Text cannot be empty"}

    cleaned = clean_text(raw_text)

    inputs = tokenizer(cleaned, return_tensors="pt", truncation=True, max_length=128)
    with torch.no_grad():
        outputs = model(**inputs)
        probs = torch.nn.functional.softmax(outputs.logits, dim=-1)
        pred = torch.argmax(probs, dim=-1).item()
        confidence = probs[0, pred].item()
    return {
        "sentiment": LABELS[pred],
        "confidence": round(confidence, 4),
        "original_text": raw_text
    }