Add debugging

backend/runner/inference.py

@@ -770,6 +770,16 @@ def process_embedding_batch_streaming(
     
     print(f"🔍 Processing batch of {len(batch)} items...")
     
+    # Debug: show first few items to understand the data structure
+    for i, item in enumerate(batch[:3]):
+        print(f" Item {i}: keys = {list(item.keys())}")
+        if 'clip_embedding' in item:
+            print(f"🔍 Item {i}: clip_embedding shape = {len(item['clip_embedding'])}")
+        if 'paintingclip_embedding' in item:
+            print(f" Item {i}: paintingclip_embedding shape = {len(item['paintingclip_embedding'])}")
+        if 'sentence_id' in item:
+            print(f" Item {i}: sentence_id = {item['sentence_id']}")
+    
     for item in batch:
         try:
             sentence_id = item.get('sentence_id', '')
@@ -780,7 +790,8 @@ def process_embedding_batch_streaming(
             elif model_type == "PaintingCLIP" and 'paintingclip_embedding' in item:
                 embedding = torch.tensor(item['paintingclip_embedding'])
             else:
-                print(f"⚠️  No embedding found for {model_type} in item: {list(item.keys())}")
+                if processed_count < 3:  # Only show first few errors
+                    print(f"⚠️  No embedding found for {model_type} in item: {list(item.keys())}")
                 continue
             
             # Calculate similarity
@@ -807,7 +818,8 @@ def process_embedding_batch_streaming(
             
         except Exception as e:
             error_count += 1
-            print(f"⚠️  Error processing item in streaming batch: {e}")
+            if error_count < 3:  # Only show first few errors
+                print(f"⚠️  Error processing item in streaming batch: {e}")
             continue
     
     print(f"🔍 Batch processing complete: {processed_count} successful, {error_count} errors")