app-v0.py

import gradio as gr
from huggingface_hub import InferenceClient
from transformers import AutoTokenizer  # Import the 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).  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:
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.
    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 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))  # Add to the beginning
            current_length += turn_tokens
        else:
            break  # Stop adding turns if we exceed the limit

    return truncated_history

def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    """Responds to a user message, maintaining conversation history, using special tokens and message list."""

    formatted_system_message = nvc_prompt_template

    truncated_history = truncate_history(history, formatted_system_message, MAX_CONTEXT_LENGTH - max_tokens - 100) # Reserve space for the new message and some generation

    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": f"<|user|>\n{user_msg}</s>"}) # Format history user message
        if assistant_msg:
            messages.append({"role": "assistant", "content": f"<|assistant|>\n{assistant_msg}</s>"}) # Format history assistant message

    messages.append({"role": "user", "content": f"<|user|>\n{message}</s>"}) # Format current user message


    response = ""
    try:
      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}") # 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(
    respond,
    additional_inputs=[
        gr.Textbox(value=nvc_prompt_template, label="System message", visible=False), # Set the NVC prompt as default and hide the system message box
        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)",
        ),
    ],
)

if __name__ == "__main__":
    demo.launch()