from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer
import os

app = FastAPI()

# Set the cache directory for Hugging Face
os.environ['TRANSFORMERS_CACHE'] = os.getenv('TRANSFORMERS_CACHE', '/app/cache')

# Load model and tokenizer
model_name = "Bijoy09/MObilebert"
try:
    model = AutoModelForSequenceClassification.from_pretrained(model_name)
    tokenizer = AutoTokenizer.from_pretrained(model_name)
except Exception as e:
    raise RuntimeError(f"Failed to load model or tokenizer: {e}")

class TextRequest(BaseModel):
    text: str

class BatchTextRequest(BaseModel):
    texts: list[str]


@app.post("/predict")
async def predict(request: TextRequest):
    try:
        model.eval()
        inputs = tokenizer.encode_plus(
            request.text,
            add_special_tokens=True,
            max_length=64,
            truncation=True,
            padding='max_length',
            return_attention_mask=True,
            return_tensors='pt'
        )
        with torch.no_grad():
            logits = model(inputs['input_ids'], attention_mask=inputs['attention_mask']).logits
            prediction = torch.argmax(logits, dim=1).item()
        return {"prediction": "Spam" if prediction == 1 else "Ham"}
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Prediction failed: {e}")


@app.post("/batch_predict")
async def batch_predict(request: BatchTextRequest):
    try:
        model.eval()
        results = []
        for idx, text in enumerate(request.texts):
            inputs = tokenizer.encode_plus(
                text,
                add_special_tokens=True,
                max_length=64,
                truncation=True,
                padding='max_length',
                return_attention_mask=True,
                return_tensors='pt'
            )
            with torch.no_grad():
                logits = model(inputs['input_ids'], attention_mask=inputs['attention_mask']).logits
                prediction = torch.argmax(logits, dim=1).item()
                results.append({"id": idx + 1, "text": text, "prediction": "Spam" if prediction == 1 else "Ham"})
        return {"results": results}
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Batch prediction failed: {e}")
# @app.post("/batch_predict")
# async def batch_predict(request: BatchTextRequest):
#     try:
#         model.eval()
        
#         # Batch encode all texts in the request at once
#         inputs = tokenizer(
#             request.texts,
#             add_special_tokens=True,
#             max_length=64,
#             truncation=True,
#             padding='max_length',
#             return_attention_mask=True,
#             return_tensors='pt'
#         )

#         # Run batch inference
#         with torch.no_grad():
#             logits = model(inputs['input_ids'], attention_mask=inputs['attention_mask']).logits
#             predictions = torch.argmax(logits, dim=1).tolist()
        
#         # Format results
#         results = [
#             {"id": idx + 1, "text": text, "prediction": "Spam" if pred == 1 else "Ham"}
#             for idx, (text, pred) in enumerate(zip(request.texts, predictions))
#         ]
        
#         return {"results": results}

#     except Exception as e:
#         logging.error(f"Batch prediction failed: {e}")
#         raise HTTPException(status_code=500, detail="Batch prediction failed. Please try again.")


@app.get("/")
async def root():
    return {"message": "Welcome to the MobileBERT API"}