handler.py

import time
import torch
from transformers import CLIPProcessor, CLIPModel
from PIL import Image
import requests
import io

class EndpointHandler:
    def __init__(self, model_dir=None):
        print("🔄 Loading model...")
        self.model = CLIPModel.from_pretrained("dazpye/clip-image")
        self.processor = CLIPProcessor.from_pretrained("dazpye/clip-image")

    def _load_image(self, image_url):
        """Fetches an image from a URL."""
        try:
            print(f"🌐 Fetching image from: {image_url}")
            response = requests.get(image_url, timeout=5)
            response.raise_for_status()
            return Image.open(io.BytesIO(response.content)).convert("RGB")
        except Exception as e:
            print(f"❌ Image loading failed: {e}")
        return None

    def __call__(self, data):
        """Processes input and runs inference."""
        start_time = time.time()  # Start timer

        print("📥 Processing input...")

        if "inputs" in data:
            data = data["inputs"]

        text = data.get("text", ["default text"])
        image_urls = data.get("images", [])

        images = [self._load_image(url) for url in image_urls if url]
        images = [img for img in images if img]  # Remove failed images

        if not images:
            return {"error": "No valid images provided."}

        # Enable padding & truncation to fix tensor error
        inputs = self.processor(
            text=text, 
            images=images, 
            return_tensors="pt", 
            padding=True, 
            truncation=True
        )

        print("🖥️ Running inference...")
        with torch.no_grad():
            outputs = self.model(**inputs)

        # Get scores & find best matches
        logits_per_image = outputs.logits_per_image
        probabilities = logits_per_image.softmax(dim=1)

        # Get top categories per image
        predictions = []
        for i, probs in enumerate(probabilities):
            sorted_indices = torch.argsort(probs, descending=True)
            best_matches = [(text[idx], probs[idx].item()) for idx in sorted_indices[:3]]  # Get top 3 matches
            predictions.append({"image_index": i, "top_matches": best_matches})

        total_time = time.time() - start_time  # Calculate time taken

        return {
            "predictions": predictions,
            "processing_time_seconds": round(total_time, 4)
        }