Add more debugging to track llama.cpp streaming response

coderai

@@ -1048,6 +1048,7 @@ class VulkanBackend(ModelBackend):
         total_content = ""
         chunk_count = 0
         try:
+            print(f"DEBUG: generate_chat_stream: Calling create_chat_completion with tools={tools}")
             stream = self.model.create_chat_completion(
                 messages=messages,
                 max_tokens=max_tokens,
@@ -1057,8 +1058,10 @@ class VulkanBackend(ModelBackend):
                 tools=tools,
                 stream=True,
             )
+            print(f"DEBUG: generate_chat_stream: Got stream object: {type(stream)}")
             for chunk in stream:
                 chunk_count += 1
+                print(f"DEBUG: generate_chat_stream: Raw chunk {chunk_count}: {repr(chunk)}")
                 delta = chunk["choices"][0].get("delta", {})
                 content = delta.get("content", "")
                 if content:
@@ -1069,6 +1072,9 @@ class VulkanBackend(ModelBackend):
                 print(f"DEBUG: Empty stream from create_chat_completion, using fallback")
                 raise Exception("Empty stream response")
         except Exception as e:
+            print(f"DEBUG: generate_chat_stream exception: {type(e).__name__}: {e}")
+            import traceback
+            traceback.print_exc()
             if chunk_count == 0:
                 print(f"Warning: create_chat_completion stream failed ({e}), falling back to text generation")
                 # Fallback: format messages manually and use text generation