app.py

import gradio as gr
from huggingface_hub import InferenceClient
import re

client = InferenceClient("Pinkstack/Superthoughts-lite-v1")

def process_think(text):
    """
    Searches for text enclosed in <think>...</think> and replaces it with a
    collapsible HTML details block.
    """
    pattern = re.compile(r'<think>(.*?)</think>', re.DOTALL)
    
    def replacer(match):
        content = match.group(1).strip()
        # You can adjust the inline styles or classes as needed.
        return (
            '<details class="think-details">'
            '<summary class="think-summary">Show thoughts</summary>'
            f'<div class="think-content">{content}</div>'
            '</details>'
        )
    
    return pattern.sub(replacer, text)

def respond(message, history: list[tuple[str, str]], system_message, max_tokens, temperature, top_p):
    # Build the conversation history with the system prompt first.
    messages = [{"role": "system", "content": system_message}]
    for user_text, bot_text in history:
        if user_text:
            messages.append({"role": "user", "content": user_text})
        if bot_text:
            messages.append({"role": "assistant", "content": bot_text})
    messages.append({"role": "user", "content": message})
    
    response = ""
    # Stream the response from the client.
    for chunk in client.chat_completion(
        messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
    ):
        token = chunk.choices[0].delta.content
        response += token
        # Process the response to swap <think> sections with collapsible blocks.
        yield process_think(response)

# Custom CSS to style the collapsible block
css = """
.think-details {
    border: 1px solid #ccc;
    border-radius: 5px;
    margin: 10px 0;
    padding: 5px;
}
.think-summary {
    cursor: pointer;
    font-weight: bold;
    background-color: #f1f1f1;
    padding: 5px;
    border-radius: 3px;
    user-select: none;
}
"""

# Note: The parameter for allowing HTML rendering may vary.
# In many cases, the chat output component will render HTML if you pass allow_html=True.
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You must always include <think> ... </think> <output> </output> tokens.", label="System message"),
        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)"),
    ],
    css=css,
    allow_html=True  # Ensures the HTML in the output is rendered.
)

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