Create app.py

app.py

@@ -0,0 +1,78 @@
+"""
+CLIP Image Embedding Generator
+
+A simple Gradio-based application for generating CLIP embeddings from uploaded images.
+Uses OpenAI's CLIP model with proper preprocessing.
+"""
+
+import gradio as gr
+from transformers import CLIPProcessor, CLIPModel
+from PIL import Image
+import torch
+import numpy as np
+from typing import Tuple
+import spaces
+
+# Load model/processor
+model: CLIPModel = CLIPModel.from_pretrained("openai/clip-vit-large-patch14")
+processor: CLIPProcessor = CLIPProcessor.from_pretrained("openai/clip-vit-large-patch14")
+model.eval()
+
+@spaces.GPU
+def get_embedding(image: Image.Image) -> Tuple[str, str]:
+    """
+    Generate CLIP embedding for an image.
+    
+    Args:
+        image (Image.Image): PIL Image object to process
+    
+    Returns:
+        Tuple[str, str]: A tuple containing (embedding_info, embedding_values)
+    """
+    device: str = "cuda" if torch.cuda.is_available() else "cpu"
+    
+    # Use CLIP's built-in preprocessing
+    inputs = processor(images=image, return_tensors="pt").to(device)
+    model_device = model.to(device)
+    
+    with torch.no_grad():
+        emb: torch.Tensor = model_device.get_image_features(**inputs)
+    
+    # L2 normalize the embeddings
+    emb = emb / emb.norm(p=2, dim=-1, keepdim=True)
+    
+    # Convert to numpy for easier handling
+    emb_numpy = emb.cpu().numpy().squeeze()
+    
+    # Create formatted output
+    embedding_info = f"Embedding Shape: {emb_numpy.shape}\nDevice Used: {device}\nNormalized: Yes (L2)"
+    
+    # Format embedding values (show first 10 and last 10 values for readability)
+    if len(emb_numpy) > 20:
+        embedding_preview = (
+            f"First 10 values: {emb_numpy[:10].tolist()}\n"
+            f"...\n"
+            f"Last 10 values: {emb_numpy[-10:].tolist()}\n\n"
+            f"Full embedding array:\n{emb_numpy.tolist()}"
+        )
+    else:
+        embedding_preview = f"Full embedding array:\n{emb_numpy.tolist()}"
+    
+    return embedding_info, embedding_preview
+
+# Create Gradio interface
+demo: gr.Interface = gr.Interface(
+    fn=get_embedding,
+    inputs=gr.Image(type="pil", label="Upload Image"),
+    outputs=[
+        gr.Textbox(label="Embedding Info", lines=3),
+        gr.Textbox(label="Embedding Values", lines=20, max_lines=30)
+    ],
+    allow_flagging="never",
+    title="CLIP Image Embedding Generator",
+    description="Upload an image to generate its CLIP embedding vector. The embedding is L2-normalized and ready for similarity computations.",
+    theme=gr.themes.Soft()
+)
+
+if __name__ == "__main__":
+    demo.launch(mcp_server=True)