Update app.py

app.py

@@ -2,39 +2,58 @@ import gradio as gr
 import torch
 from transformers import pipeline
 
-print("Loading Dawn-superlite")
-# We add model_kwargs to silence the "tie_word_embeddings" warnings
+print("Loading Dawn-superlite...")
+
+# The 'trust_remote_code=True' argument is required for this specific model 
+# because it uses a custom architecture defined in the HF repository.
 pipe = pipeline(
     "text-generation", 
-    model="Dawn-AI/Dawn-superlite", # Make sure this is your actual repo name
+    model="Dawn-AI/Dawn-superlite",
     device_map="auto",
-    torch_dtype=torch.bfloat16, # Use bfloat16 to save some memory if possible
+    torch_dtype=torch.bfloat16,
+    trust_remote_code=True, 
     model_kwargs={
         "tie_word_embeddings": False
     }
 )
+
 print("Model loaded successfully!")
 
 def chat_with_dawn(message, history, system_prompt):
     messages = []
     
+    # Add system prompt if provided
     if system_prompt.strip():
         messages.append({"role": "system", "content": system_prompt})
         
+    # Build history for the chat template
     for user_msg, assistant_msg in history:
         messages.append({"role": "user", "content": user_msg})
         messages.append({"role": "assistant", "content": assistant_msg})
         
+    # Append the new user message
     messages.append({"role": "user", "content": message})
     
-    response = pipe(messages, max_new_tokens=512, truncation=True)
+    # Generate response
+    # Using a temperature > 0 and top_p for more natural reasoning responses
+    response = pipe(
+        messages, 
+        max_new_tokens=512, 
+        truncation=True,
+        do_sample=True,
+        temperature=0.7,
+        top_p=0.9
+    )
+    
+    # Extract the text from the returned message object
     generated_text = response[0]['generated_text']
     
+    # The pipeline usually returns the full list of messages including the new one
     if isinstance(generated_text, list):
         return generated_text[-1]['content']
-    else:
-        return generated_text
+    return str(generated_text)
 
+# Define the UI components
 system_textbox = gr.Textbox(
     value="You are Dawn, a brilliant reasoning AI.",
     label="System Prompt",
@@ -42,13 +61,13 @@ system_textbox = gr.Textbox(
     interactive=True
 )
 
-# NOTICE: The theme="soft" line has been removed to prevent the crash!
+# Launch the Gradio Interface
 demo = gr.ChatInterface(
     fn=chat_with_dawn,
     additional_inputs=[system_textbox],
     title="🌅 Dawn Superlite",
-    description="Running Dawn. Warning: May be slow on CPU.",
-    examples=["Who are you?", "Write a Python script to reverse a string."]
+    description="Running Dawn-superlite. Note: Performance depends on your hardware (CPU vs GPU).",
+    examples=["Explain quantum entanglement like I'm five.", "Write a Python script to reverse a string."]
 )
 
 if __name__ == "__main__":