I am totally not 'PauliePocket' trust

app.py

@@ -86,15 +86,38 @@ def load_model(model_id: str, device: str):
     return tokenizer, model
 
 
+def _extract_text(content) -> str:
+    """Gradio 6's messages format stores `content` as either a plain string
+    or a list of content blocks (e.g. [{"type": "text", "text": "..."}]).
+    Normalize either shape down to plain text."""
+    if content is None:
+        return ""
+    if isinstance(content, str):
+        return content
+    if isinstance(content, list):
+        parts = []
+        for block in content:
+            if isinstance(block, str):
+                parts.append(block)
+            elif isinstance(block, dict):
+                if "text" in block:
+                    parts.append(block.get("text") or "")
+        return "".join(parts)
+    if isinstance(content, dict):
+        return content.get("text", "") or ""
+    return str(content)
+
+
 def build_messages(history, user_message: str):
-    """history is already a list of {'role': ..., 'content': ...} dicts
-    (Gradio Chatbot 'messages' format)."""
+    """history is a list of {'role': ..., 'content': ...} dicts
+    (Gradio Chatbot 'messages' format). content may be a string or a
+    list of content blocks."""
     messages = []
     for msg in history or []:
         role = msg.get("role")
-        content = msg.get("content")
-        if role in ("user", "assistant") and content:
-            messages.append({"role": role, "content": content})
+        text = _extract_text(msg.get("content"))
+        if role in ("user", "assistant") and text:
+            messages.append({"role": role, "content": text})
 
     messages.append({"role": "user", "content": user_message})
     return messages
@@ -104,13 +127,13 @@ def build_fallback_prompt(history, user_message: str) -> str:
     parts = []
     for msg in history or []:
         role = msg.get("role")
-        content = msg.get("content")
-        if not content:
+        text = _extract_text(msg.get("content"))
+        if not text:
             continue
         if role == "user":
-            parts.append(f"User: {content}")
+            parts.append(f"User: {text}")
         elif role == "assistant":
-            parts.append(f"Assistant: {content}")
+            parts.append(f"Assistant: {text}")
 
     parts.append(f"User: {user_message}")
     parts.append("Assistant:")
@@ -137,8 +160,10 @@ def _generate(message, history, model_label, max_new_tokens, device_name):
         input_ids = tokenizer(prompt, return_tensors="pt").input_ids
 
     input_ids = input_ids.to(_model_device(model))
+    attention_mask = torch.ones_like(input_ids)
 
     generation_kwargs = {
+        "attention_mask": attention_mask,
         "max_new_tokens": max_new_tokens,
         "do_sample": True,
         "temperature": 0.7,
@@ -221,7 +246,7 @@ with gr.Blocks(title="SmolLM2 Chat") as demo:
                 label="Conversation",
             )
             message = gr.Textbox(
-                placeholder="I like dih...",
+                placeholder="PauliePocket is just better than you...",
                 label="Message",
                 lines=3,
             )
@@ -233,7 +258,7 @@ with gr.Blocks(title="SmolLM2 Chat") as demo:
                     ["Generate a song about a girl falling in love."],
                 ],
                 inputs=message,
-                label="Try these prompts",
+                label="Try these stupid prompts",
             )
 
             with gr.Row():