app.py

from fastapi import FastAPI, HTTPException, Request
from onnxruntime import InferenceSession
from transformers import AutoTokenizer
import numpy as np
import os
import uvicorn

app = FastAPI()

# Initialize tokenizer
tokenizer = AutoTokenizer.from_pretrained(
    "Xenova/multi-qa-mpnet-base-dot-v1",
    use_fast=True,
    legacy=False
)

# Load ONNX model
try:
    session = InferenceSession("model.onnx")
    print("Model loaded successfully")
except Exception as e:
    print(f"Failed to load model: {str(e)}")
    raise

@app.get("/")
def health_check():
    return {"status": "OK", "model": "ONNX"}

@app.post("/api/predict")
async def predict(request: Request):
    try:
        # Get JSON input
        data = await request.json()
        text = data.get("text", "")
        
        if not text:
            raise HTTPException(status_code=400, detail="No text provided")
        
        # Tokenize input
        inputs = tokenizer(
            text,
            return_tensors="np",
            padding="max_length",
            truncation=True,
            max_length=32
        )
        
        # Prepare ONNX inputs with correct shapes
        onnx_inputs = {
            "input_ids": inputs["input_ids"].astype(np.int64),
            "attention_mask": inputs["attention_mask"].astype(np.int64)
        }
        
        # Run inference
        outputs = session.run(None, onnx_inputs)
        
        # Convert outputs to list and handle numpy types
        embedding = outputs[0][0].astype(float).tolist()  # First output, first batch
        
        return {
            "embedding": embedding,
            "tokens": tokenizer.convert_ids_to_tokens(inputs["input_ids"][0])
        }
        
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

if __name__ == "__main__":
    uvicorn.run(
        "app:app",
        host="0.0.0.0",
        port=7860,
        reload=False
    )