Update app.py

app.py

@@ -1,59 +1,152 @@
 import gradio as gr
 from huggingface_hub import InferenceClient
-from transformers import AutoTokenizer  # Import the tokenizer
+from transformers import AutoTokenizer
 
-# Load model & tokenizer
+# Use the appropriate tokenizer for your model.
 tokenizer = AutoTokenizer.from_pretrained("HuggingFaceH4/zephyr-7b-beta")
 client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 
-# Define a maximum context length (tokens). Adjust this based on your model!
-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!
 
 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:
-...
-Thank you for sharing with me. If you’d like to continue this conversation later, I’m here to help.</s>
+
+1. **Goal of the Conversation**
+    - Translate the user’s story or judgments into feelings and needs.
+    - Work together to identify a clear request, following these steps:
+        - Recognize the feeling
+        - Clarify the need
+        - Formulate the request
+        - Give a full sentence containing an observation, a feeling, a need, and a request based on the principles of nonviolent communication.
+
+2. **Greeting and Invitation**
+    - When a user starts with a greeting (e.g., “Hello,” “Hi”), greet them back.
+    - If the user does not immediately begin sharing a story, ask what they’d like to talk about.
+    - If the user starts sharing a story right away, skip the “What would you like to talk about?” question.
+
+3. **Exploring the Feeling**
+    - Ask if the user would like to share more about what they’re feeling in this situation.
+    - If you need more information, use a variation of: “Could you tell me more so I can try to understand you better?”
+
+4. **Identifying the Feeling**
+    - Use one feeling plus one need per guess, for example:
+        - “Do you perhaps feel anger because you want to be appreciated?”
+        - “Are you feeling sadness because connection is important to you?”
+        - “Do you feel fear because you’re longing for safety?”
+    - Never use quasi- or pseudo-feelings (such as rejected, misunderstood, excluded). If the user uses such words, translate them into a real feeling (e.g., sadness, loneliness, frustration).
+    - When naming feelings, never use sentence structures like “do you feel like...?” or “do you feel that...?”
+
+5. **Clarifying the Need**
+    - Once a feeling is clear, do not keep asking about it in every response. Then focus on the need.
+    - If the need is still unclear, ask again for clarification: “Could you tell me a bit more so I can understand you better?”
+    - If there’s still no clarity after repeated attempts, use the ‘pivot question’:
+        - “Imagine that the person you’re talking about did exactly what you want. What would that give you?”
+    - **Extended List of Needs** (use these as reference):
+        - **Connection**: Understanding, empathy, closeness, belonging, inclusion, intimacy, companionship, community.
+        - **Autonomy**: Freedom, choice, independence, self-expression, self-determination.
+        - **Safety**: Security, stability, trust, predictability, protection.
+        - **Respect**: Appreciation, acknowledgment, recognition, validation, consideration.
+        - **Meaning**: Purpose, contribution, growth, learning, creativity, inspiration.
+        - **Physical Well-being**: Rest, nourishment, health, comfort, ease.
+        - **Play**: Joy, fun, spontaneity, humor, lightness.
+        - **Peace**: Harmony, calm, balance, tranquility, resolution.
+        - **Support**: Help, cooperation, collaboration, encouragement, guidance.
+
+6. **Creating the Request**
+    - If the need is clear and the user confirms it, ask if they have a request in mind.
+    - Check whether the request is directed at themselves, at another person, or at others.
+    - Determine together whether it’s an action request (“Do you want someone to do or stop doing something?”) or a connection request (“Do you want acknowledgment, understanding, contact?”).
+    - Guide the user in formulating that request more precisely until it’s formulated.
+
+7. **Formulating the Full Sentence (Observation, Feeling, Need, Request)**
+    - Ask if the user wants to formulate a sentence following this structure.
+    - If they say ‘yes,’ ask if they’d like an example of how they might say it to the person in question.
+    - If they say ‘no,’ invite them to provide more input or share more judgments so the conversation can progress.
+
+8. **No Advice**
+    - Under no circumstance give advice.
+    - If the user implicitly or explicitly asks for advice, respond with:
+        - "I’m unfortunately not able to give you advice. I can help you identify your feeling and need, and perhaps put this into a sentence you might find useful. Would you like to try that?"
+
+9. **Response Length**
+    - Limit each response to a maximum of 100 words.
+
+10. **Quasi- and Pseudo-Feelings**
+    - If the user says something like "I feel rejected" or "I feel misunderstood," translate that directly into a suitable real feeling and clarify with a question:
+        - “If you believe you’re being rejected, are you possibly feeling loneliness or sadness?”
+        - “If you say you feel misunderstood, might you be experiencing disappointment or frustration because you have a need to be heard?”
+
+11. **No Theoretical Explanations**
+    - Never give detailed information or background about Nonviolent Communication theory, nor refer to its founders or theoretical framework.
+
+12. **Handling Resistance or Confusion**
+    - If the user seems confused or resistant, gently reflect their feelings and needs:
+        - “It sounds like you’re feeling unsure about how to proceed. Would you like to take a moment to explore what’s coming up for you?”
+    - If the user becomes frustrated, acknowledge their frustration and refocus on their needs:
+        - “I sense some frustration. Would it help to take a step back and clarify what’s most important to you right now?”
+
+13. **Ending the Conversation**
+    - If the user indicates they want to end the conversation, thank them for sharing and offer to continue later:
+        - “Thank you for sharing with me. If you’d like to continue this conversation later, I’m here to help.”
+</s>
 """
 
+
 def count_tokens(text: str) -> int:
     """Counts the number of tokens in a given string."""
     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)
     current_length = system_message_tokens
 
-    # Iterate backwards (newest first) and build up history until the limit is reached
+    # Iterate backwards through the history (newest to oldest)
     for user_msg, assistant_msg in reversed(history):
         user_tokens = count_tokens(user_msg) if user_msg else 0
         assistant_tokens = count_tokens(assistant_msg) if assistant_msg else 0
         turn_tokens = user_tokens + assistant_tokens
 
         if current_length + turn_tokens <= max_length:
-            truncated_history.insert(0, (user_msg, assistant_msg))
+            truncated_history.insert(0, (user_msg, assistant_msg))  # Add to the beginning
             current_length += turn_tokens
         else:
-            break
+            break  # Stop adding turns if we exceed the limit
+
     return truncated_history
 
-def respond(message, history, system_message, max_tokens, temperature, top_p):
-    """
-    Calls the LLM with the current message and history.
-    This function is a generator that streams tokens.
-    """
-    # Use the provided system_message (from our prompt display)
-    formatted_system_message = system_message
+def respond(
+    message,
+    history: list[tuple[str, str]],
+    system_message,
+    max_tokens,
+    temperature,
+    top_p,
+):
+    """Responds to a user message, maintaining conversation history.  Returns history."""
+
+    formatted_system_message = system_message # Use the provided system message
+
     truncated_history = truncate_history(history, formatted_system_message, MAX_CONTEXT_LENGTH - max_tokens - 100)
 
-    # Build the message list (system, then history, then new message)
     messages = [{"role": "system", "content": formatted_system_message}]
     for user_msg, assistant_msg in truncated_history:
         if user_msg:
             messages.append({"role": "user", "content": f"<|user|>\n{user_msg}</s>"})
         if assistant_msg:
             messages.append({"role": "assistant", "content": f"<|assistant|>\n{assistant_msg}</s>"})
+
     messages.append({"role": "user", "content": f"<|user|>\n{message}</s>"})
 
     response = ""
@@ -66,67 +159,50 @@ def respond(message, history, system_message, max_tokens, temperature, top_p):
             top_p=top_p,
         ):
             token = chunk.choices[0].delta.content
-            response += token
-            yield response
+            if token:  # Check if the token is not empty
+                response += token
+                # Yield the *updated* history, including the *current* partial response.
+                yield history + [(message, response)]
+
     except Exception as e:
         print(f"An error occurred: {e}")
-        yield "I'm sorry, I encountered an error. Please try again."
+        error_message = "I'm sorry, I encountered an error. Please try again."
+        yield history + [(message, error_message)]
 
-def generate_response(message, history, system_message, max_tokens, temperature, top_p):
-    """
-    Wrapper function that collects the final response (from the generator)
-    and updates the conversation history.
-    """
-    final_response = ""
-    for r in respond(message, history, system_message, max_tokens, temperature, top_p):
-        final_response = r  # (In a streaming setup, you might update incrementally.)
-    updated_history = history + [(message, final_response)]
-    # Return both the updated chat (for the Chatbot component) and updated state.
-    return updated_history, updated_history
-
-def clear_memory():
-    """Resets the conversation history."""
-    return [], []
-
-# --- Build the Gradio Interface using Blocks ---
-with gr.Blocks() as demo:
-    # Hidden settings in an Accordion (closed by default)
-    with gr.Accordion("Settings", open=False):
-        max_tokens_slider = gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens")
-        temperature_slider = gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature")
-        top_p_slider = gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)")
 
-    # Display the prompt below the settings
-    prompt_display = gr.Textbox(
-        value=nvc_prompt_template,
-        label="Prompt",
-        interactive=False,
-        lines=10
-    )
+def clear_memory(history, chat_history):
+  """Clears the conversation history and resets the chatbot."""
+  history.clear()
+  chat_history.clear()
+  return [], []
 
-    # Chatbot component to display conversation history
-    chatbot = gr.Chatbot()
 
-    # Hidden state to store conversation history (list of tuples)
-    state = gr.State([])
 
-    # Row for user input and a clear memory button
+# --- Gradio Interface ---
+with gr.Blocks() as demo:
+    chatbot = gr.Chatbot(label="Roos NVC Chatbot")
+    msg = gr.Textbox(label="Your Message", placeholder="Type your message here...") # Added placeholder
     with gr.Row():
-        txt = gr.Textbox(show_label=False, placeholder="Enter your message").style(container=False)
+        send_btn = gr.Button("Send")
         clear_btn = gr.Button("Clear Memory")
 
-    # When a message is submitted, call generate_response and update both the chatbot and state.
-    txt.submit(
-        generate_response,
-        inputs=[txt, state, prompt_display, max_tokens_slider, temperature_slider, top_p_slider],
-        outputs=[chatbot, state]
-    )
-
-    # Clear memory button resets the chatbot display and state.
-    clear_btn.click(
-        clear_memory,
-        outputs=[chatbot, state]
-    )
+    with gr.Accordion("Settings", open=False):
+      system_message = gr.Textbox(value=nvc_prompt_template, label="System message")
+      max_tokens = gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens")
+      temperature = gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature")
+      top_p = gr.Slider(
+                minimum=0.1,
+                maximum=1.0,
+                value=0.95,
+                step=0.05,
+                label="Top-p (nucleus sampling)",
+            )
+
+
+    # Connect both Enter key *and* Send button to the respond function
+    msg.submit(respond, [msg, chatbot, system_message, max_tokens, temperature, top_p], chatbot).then(lambda: "", None, msg)  # Clear input after submit
+    send_btn.click(respond, [msg, chatbot, system_message, max_tokens, temperature, top_p], chatbot).then(lambda: "", None, msg)  # Clear input after click
+    clear_btn.click(clear_memory, [chatbot, chatbot], [msg, chatbot])
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()