Update app.py

app.py

@@ -38,60 +38,57 @@ Rules:
 """.strip()
 
 
-def build_prompt(message, history):
-  """
-  Gradio ChatInterface (type='messages') passes history as a list of
-  dicts like: {'role': 'user'|'assistant', 'content': '...'}.[web:120]
-  Turn that into a plain-text prompt for DialoGPT.
-  """
-  lines = [f"System: {SHINCHAN_SYSTEM_PROMPT}"]
-  for turn in history:
-    role = turn.get("role")
-    content = turn.get("content", "")
-    if role == "user":
-      lines.append(f"User: {content}")
-    elif role == "assistant":
-      lines.append(f"Shinchan: {content}")
-  lines.append(f"User: {message}")
-  lines.append("Shinchan:")
-  return "\n".join(lines)
-
-
-def respond(message, history):
-  """
-  ChatInterface expects: fn(message:str, history:list[dict]) -> str or dict or list.[web:120]
-  Do NOT return (reply, history) here – that caused your 'tuple has no attribute get' error.
-  """
-  prompt = build_prompt(message, history)
-  inputs = tokenizer(prompt, return_tensors="pt")
-
-  with torch.no_grad():
-    output_ids = model.generate(
-      **inputs,
-      max_new_tokens=80,
-      pad_token_id=tokenizer.eos_token_id,
-      do_sample=True,
-      top_p=0.9,
-      temperature=0.9,
+def respond(message: str, history: list[dict]) -> str:
+    """
+    Gradio ChatInterface(type='messages') calls this as (message, history)
+    and expects JUST a reply string or message dict.[web:120]
+    """
+
+    # Build a simple one-turn prompt: system + latest user message.
+    prompt = (
+        f"System: {SHINCHAN_SYSTEM_PROMPT}\n\n"
+        f"User: {message}\n"
+        f"Shinchan:"
     )
 
-  full_text = tokenizer.decode(output_ids[0], skip_special_tokens=True)
+    inputs = tokenizer(prompt, return_tensors="pt")
 
-  if "Shinchan:" in full_text:
-    reply = full_text.split("Shinchan:")[-1].strip()
-  else:
-    reply = full_text.strip()
+    with torch.no_grad():
+        output_ids = model.generate(
+            **inputs,
+            max_new_tokens=80,
+            pad_token_id=tokenizer.eos_token_id,
+            do_sample=True,
+            top_p=0.9,
+            temperature=0.9,
+        )
 
-  # IMPORTANT: return just the reply string
-  return reply
+    full_text = tokenizer.decode(output_ids[0], skip_special_tokens=True)
+
+    # Extract text after the last "Shinchan:" marker.
+    reply = ""
+    if "Shinchan:" in full_text:
+        reply = full_text.split("Shinchan:")[-1].strip()
+    else:
+        # Fallback: everything after the original prompt length
+        reply = full_text[len(prompt):].strip()
+
+    # If model somehow returns empty or just punctuation, use a safe fallback.
+    if not reply or len(reply.replace(".", "").replace("!", "").strip()) < 3:
+        reply = (
+            "Heeey, it’s Shinchan! 😂 I heard you, even if my brain glitched for a second. "
+            "Tell me more, I’m listening. 🌻"
+        )
+
+    return reply
 
 
 demo = gr.ChatInterface(
-  fn=respond,
-  type="messages",  # explicit; default since Gradio 5 but makes intent clear[web:120]
-  title="Shinchan for Ruru",
-  description="Private Shinchan-style chat for Ruru.",
+    fn=respond,
+    type="messages",  # use new messages format internally[web:120]
+    title="Shinchan for Ruru",
+    description="Private Shinchan-style chat for Ruru.",
 )
 
 if __name__ == "__main__":
-  demo.launch()
+    demo.launch()