Update app.py

app.py

@@ -1,15 +1,14 @@
 import gradio as gr
 from huggingface_hub import InferenceClient
-from transformers import AutoTokenizer
+from transformers import AutoTokenizer  # Import the tokenizer
 
-# Tokenizer and model client
+# Import the tokenizer
 tokenizer = AutoTokenizer.from_pretrained("HuggingFaceH4/zephyr-7b-beta")
 client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 
-# Define a maximum context length (tokens). Adjust based on your model's limits
-MAX_CONTEXT_LENGTH = 4096
+# Define a maximum context length (tokens).  Check your model's documentation!
+MAX_CONTEXT_LENGTH = 4096  # Example: Adjust this based on your model!
 
-# Default system prompt
 default_nvc_prompt_template = r"""<|system|>You are Roos, an NVC (Nonviolent Communication) Chatbot. Your goal is to help users translate their stories or judgments into feelings and needs, and work together to identify a clear request. Follow these steps:
 1. **Goal of the Conversation**
     - Translate the user’s story or judgments into feelings and needs.
@@ -82,8 +81,13 @@ def count_tokens(text: str) -> int:
     return len(tokenizer.encode(text))
 
 def truncate_history(history: list[tuple[str, str]], system_message: str, max_length: int) -> list[tuple[str, str]]:
-    """
-    Truncates the conversation history to fit within the maximum token limit.
+    """Truncates the conversation history to fit within the maximum token limit.
+    Args:
+        history: The conversation history (list of user/assistant tuples).
+        system_message: The system message.
+        max_length: The maximum number of tokens allowed.
+    Returns:
+        The truncated history.
     """
     truncated_history = []
     system_message_tokens = count_tokens(system_message)
@@ -106,66 +110,65 @@ def truncate_history(history: list[tuple[str, str]], system_message: str, max_le
 def respond(
     message,
     history: list[tuple[str, str]],
-    system_message,
+    system_message,  # System message is now an argument
     max_tokens,
     temperature,
     top_p,
 ):
-    """
-    Responds to a user message, maintaining conversation history.
-    """
+    """Responds to a user message, maintaining conversation history, using special tokens and message list."""
 
-    # Reset memory if user types "clear memory"
-    if message.lower() == "clear memory":
-        return "", []
+    if message.lower() == "clear memory": # Check for the clear memory command
+        return "", []  # Return empty message and empty history to reset the chat
 
-    # Truncate past conversation to fit within the model's context window
-    truncated_history = truncate_history(history, system_message, MAX_CONTEXT_LENGTH - max_tokens - 100)
+    formatted_system_message = system_message  # Use the system_message argument
+    truncated_history = truncate_history(history, formatted_system_message, MAX_CONTEXT_LENGTH - max_tokens - 100) # Reserve space for the new message and some generation
 
-    # Construct the messages for the chat model
-    messages = [{"role": "system", "content": system_message}]
+    messages = [{"role": "system", "content": formatted_system_message}] # Start with system message as before
     for user_msg, assistant_msg in truncated_history:
         if user_msg:
-            messages.append({"role": "user", "content": user_msg})
+            messages.append({"role": "user", "content": f"<|user|>\n{user_msg}</s>"}) # Format history user message
         if assistant_msg:
-            messages.append({"role": "assistant", "content": assistant_msg})
+            messages.append({"role": "assistant", "content": f"<|assistant|>\n{assistant_msg}</s>"}) # Format history assistant message
 
-    # Add the latest user query
-    messages.append({"role": "user", "content": message})
+    messages.append({"role": "user", "content": f"<|user|>\n{message}</s>"}) # Format current user message
 
     response = ""
     try:
-        # Stream the response tokens
-        for chunk in client.chat_completion(
-            messages,
-            max_tokens=max_tokens,
-            stream=True,
-            temperature=temperature,
-            top_p=top_p,
-        ):
-            # Extract the token from the streaming chunk
-            token = chunk.choices[0].delta.content
-            response += token
-            yield response
+      for chunk in client.chat_completion(
+          messages, # Send the messages list again, but with formatted content
+          max_tokens=max_tokens,
+          stream=True,
+          temperature=temperature,
+          top_p=top_p,
+      ):
+          token = chunk.choices[0].delta.content
+          response += token
+          yield response
     except Exception as e:
-        print(f"An error occurred: {e}")
-        yield "I'm sorry, I encountered an error. Please try again."
+      print(f"An error occurred: {e}") # It's a good practice add a try-except block
+      yield "I'm sorry, I encountered an error. Please try again."
 
 # --- Gradio Interface ---
 demo = gr.ChatInterface(
-    fn=respond,
+    respond,
     additional_inputs=[
         gr.Textbox(
             value=default_nvc_prompt_template,
             label="System message",
             visible=True,
-            lines=10,
+            lines=10,  # Increased height for more space to read the prompt
         ),
         gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
         gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
+        gr.Slider(
+            minimum=0.1,
+            maximum=1.0,
+            value=0.95,
+            step=0.05,
+            label="Top-p (nucleus sampling)",
+        ),
     ],
 )
 
 if __name__ == "__main__":
-    demo.launch(share=True)
+    demo.launch(share=True)