Update app.py

app.py

@@ -348,26 +348,43 @@ def generate_chat(message, history, system_prompt, max_tokens, temperature, top_
         yield f"🚨 MODEL FAILED TO LOAD 🚨\n\nError Details:\n{LOAD_ERROR}"
         return
 
-    if not message.strip():
+    # Helper function to guarantee we always get a string out of Gradio's weird formats
+    def force_str(item):
+        if item is None:
+            return ""
+        if isinstance(item, str):
+            return item
+        if isinstance(item, dict) and "text" in item:
+            return str(item["text"])
+        if isinstance(item, (list, tuple)):
+            if len(item) > 0 and isinstance(item[0], str):
+                return item[0]
+            return " ".join(str(x) for x in item)
+        return str(item)
+
+    safe_message = force_str(message)
+    if not safe_message.strip():
         yield "Please provide a valid text string."
         return
 
     # --- BUILD CHATML PROMPT FORMAT USING THE TOKENIZER ---
-    messages = [{"role": "system", "content": system_prompt}]
+    messages = [{"role": "system", "content": force_str(system_prompt)}]
     
     for msg in history:
-        # Robust history parsing for older Gradio versions (list of lists) and newer versions (list of dicts)
+        # Robust history parsing for all Gradio versions
         if isinstance(msg, dict):
-            messages.append({"role": msg.get("role", "user"), "content": msg.get("content", "")})
+            messages.append({"role": force_str(msg.get("role", "user")), "content": force_str(msg.get("content", ""))})
         elif hasattr(msg, "role"):
-            messages.append({"role": msg.role, "content": msg.content})
-        elif isinstance(msg, (list, tuple)) and len(msg) == 2:
-            messages.append({"role": "user", "content": msg[0]})
-            messages.append({"role": "assistant", "content": msg[1]})
-
-    messages.append({"role": "user", "content": message})
+            messages.append({"role": force_str(msg.role), "content": force_str(msg.content)})
+        elif isinstance(msg, (list, tuple)) and len(msg) >= 2:
+            if msg[0] is not None:
+                messages.append({"role": "user", "content": force_str(msg[0])})
+            if msg[1] is not None:
+                messages.append({"role": "assistant", "content": force_str(msg[1])})
+
+    messages.append({"role": "user", "content": safe_message})
     
-    # Use tokenizer's built in template to guarantee perfect tokenization and trailing space handling
+    # Use tokenizer's built in template to guarantee perfect tokenization
     prompt = TOKENIZER.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
     input_ids = TOKENIZER(prompt, return_tensors="pt")["input_ids"].to(DEVICE)
     
@@ -411,6 +428,7 @@ def generate_chat(message, history, system_prompt, max_tokens, temperature, top_
         next_token_logits = outputs["logits"][0, -1, :].clone()
         past_key_values = outputs["past_key_values"]
 
+
 with gr.Blocks(title="🤖 OPTIMUS Web Inference Hub") as demo:
     gr.Markdown(
         """