format prompts and add debug log

utils/models.py

@@ -39,22 +39,37 @@ def generate_summaries(example, model_a_name, model_b_name):
     """
     if generation_interrupt.is_set():
         return "", ""
-
+    
     context_text = ""
     context_parts = []
-    if "full_contexts" in example:
-        for ctx in example["full_contexts"]:
+    
+    if "full_contexts" in example and example["full_contexts"]:
+        for i, ctx in enumerate(example["full_contexts"]):
+            content = ""
+            
+            # Extract content from either dict or string
             if isinstance(ctx, dict) and "content" in ctx:
-                context_parts.append(ctx["content"])
-        context_text = "\n---\n".join(context_parts)
+                content = ctx["content"]
+            elif isinstance(ctx, str):
+                content = ctx
+            
+            # Add document number if not already present
+            if not content.strip().startswith("Document"):
+                content = f"Document {i+1}:\n{content}"
+            
+            context_parts.append(content)
+        
+        context_text = "\n\n".join(context_parts)
     else:
-        raise ValueError("No context found in the example.")
-
+        # Provide a graceful fallback instead of raising an error
+        print("Warning: No full context found in the example, using empty context")
+        context_text = ""
+    
     question = example.get("question", "")
-
+    
     if generation_interrupt.is_set():
         return "", ""
-
+    
     # Run model A
     summary_a = run_inference(models[model_a_name], context_text, question)
     
@@ -105,6 +120,8 @@ def run_inference(model_name, context, question):
         )
 
         text_input = format_rag_prompt(question, context, accepts_sys)
+        print(type(text_input))
+        print(text_input)
 
         # Check interrupt before generation
         if generation_interrupt.is_set():