Update vlm.py

vlm.py

@@ -51,12 +51,29 @@ def encode_image(image_path):
 # Build messages
 #
 def normalize_message_content(msg: dict) -> dict:
-    if isinstance(msg.get("content"), str):
+    content = msg.get("content")
+
+    # Case 1: Already in expected format
+    if isinstance(content, list) and all(isinstance(item, dict) for item in content):
+        return {"role": msg["role"], "content": content}
+
+    # Case 2: String (assume text)
+    if isinstance(content, str):
+        return {"role": msg["role"], "content": [{"type": "text", "text": content}]}
+
+    # Case 3: Tuple with image path(s)
+    if isinstance(content, tuple):
         return {
             "role": msg["role"],
-            "content": [{"type": "text", "text": msg["content"]}]
+            "content": [
+                {"type": "image", "image": encode_image(path)}  # your `encode_image()` function
+                for path in content if isinstance(path, str)
+            ]
         }
-    return msg
+
+    logger.warning(f"Unexpected content format in message: {msg}")
+    return {"role": msg["role"], "content": [{"type": "text", "text": str(content)}]}
+
 
 def build_messages(message: dict, history: list[dict]):
     """Build messages given message & history from a **multimodal** chat interface.