app.py

import gradio as gr
from transformers import AutoModel, AutoProcessor
import torch
import requests
from PIL import Image
from io import BytesIO

fashion_items = ['top', 'trousers', 'jumper']

# Load model and processor with proper meta tensor handling
model_name = 'Marqo/marqo-fashionSigLIP'

# Force CPU usage to avoid device mapping issues
device = torch.device('cpu')

# Handle meta tensor initialization properly
try:
    # Load model with empty weights initialization to avoid meta tensor issues
    model = AutoModel.from_pretrained(
        model_name,
        trust_remote_code=True,
        torch_dtype=torch.float32
    )
    
    # Check if model has the to_empty method and use it for meta tensor initialization
    if hasattr(model, 'model') and hasattr(model.model, 'to_empty'):
        model.model.to_empty(device=device)
    elif hasattr(model, 'to_empty'):
        model.to_empty(device=device)
    else:
        # Fallback to regular to() method
        model = model.to(device)
        
except Exception as e:
    print(f"Primary loading method failed: {e}")
    # Fallback method - load with minimal configuration
    try:
        model = AutoModel.from_pretrained(
            model_name,
            trust_remote_code=True
        )
        # Move to CPU after loading
        model = model.to(device)
    except Exception as e2:
        print(f"Fallback method also failed: {e2}")
        # Last resort - try loading with low CPU memory usage
        model = AutoModel.from_pretrained(
            model_name,
            trust_remote_code=True,
            low_cpu_mem_usage=False  # Disable to avoid accelerate issues
        )
        model = model.to(device)

processor = AutoProcessor.from_pretrained(model_name, trust_remote_code=True)

# Preprocess and normalize text data
with torch.no_grad():
    # Ensure truncation and padding are activated
    processed_texts = processor(
        text=fashion_items,
        return_tensors="pt",
        truncation=True,  # Ensure text is truncated to fit model input size
        padding=True      # Pad shorter sequences so that all are the same length
    )['input_ids']
    
    text_features = model.get_text_features(processed_texts)
    text_features = text_features / text_features.norm(dim=-1, keepdim=True)

# Prediction function
def predict_from_url(url):
    # Check if the URL is empty
    if not url:
        return {"Error": "Please input a URL"}
    
    try:
        image = Image.open(BytesIO(requests.get(url).content))
    except Exception as e:
        return {"Error": f"Failed to load image: {str(e)}"}
    
    processed_image = processor(images=image, return_tensors="pt")['pixel_values']
    
    with torch.no_grad():
        image_features = model.get_image_features(processed_image)
        image_features = image_features / image_features.norm(dim=-1, keepdim=True)
        text_probs = (100 * image_features @ text_features.T).softmax(dim=-1)
        
    return {fashion_items[i]: float(text_probs[0, i]) for i in range(len(fashion_items))}

# Gradio interface
demo = gr.Interface(
    fn=predict_from_url, 
    inputs=gr.Textbox(label="Enter Image URL"),
    outputs=gr.Label(label="Classification Results"), 
    title="Fashion Item Classifier",
    allow_flagging="never"
)

# Launch the interface
demo.launch()