app.py

import os
os.system('pip install minijinja')
import gradio as gr
from huggingface_hub import InferenceClient
import torch
import spaces

# Initialize the client with your model
client = InferenceClient("karpathy/gpt2_1558M_final2_hf")

default_system = 'You are a helpful assistant'

@spaces.GPU
def respond(message, history, system_message, max_tokens, temperature, top_p):
    # Combine system message, history, and new message
    full_prompt = f"{system_message}\n\n"
    for user, assistant in history:
        full_prompt += f"Human: {user}\nAssistant: {assistant}\n"
    full_prompt += f"Human: {message}\nAssistant:"

    response = ""
    for chunk in client.text_generation(
        full_prompt,
        max_new_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
    ):
        if chunk.token.text:
            response += chunk.token.text
            yield history + [(message, response)]

    # If the response is empty, yield a default message
    if not response:
        yield history + [(message, "I apologize, but I couldn't generate a response.")]

def clear_session():
    return "", []

def modify_system_session(system):
    if not system:
        system = default_system
    return system, system, []

def use_example_and_respond(example, history, system_message, max_tokens, temperature, top_p):
    return example, respond(example, history, system_message, max_tokens, temperature, top_p)

# Define example prompts
unicorn_example = "In a shocking finding, scientist discovered a herd of unicorns living in a remote, previously unexplored valley, in the Andes Mountains. Even more surprising to the researchers was the fact that the unicorns spoke perfect English."
time_travel_example = "Explain the grandfather paradox in time travel and propose a potential resolution."

with gr.Blocks() as demo:
    gr.Markdown("<h1 style='text-align: center;'>GPT-2 1.5B Chat Demo</h1>")
    
    with gr.Row():
        with gr.Column(scale=3):
            system_input = gr.Textbox(value=default_system, lines=1, label='System Prompt')
        with gr.Column(scale=1):
            modify_system = gr.Button("🛠️ Set system prompt and clear history")
    
    system_state = gr.Textbox(value=default_system, visible=False)
    chatbot = gr.Chatbot(label='GPT-2 1.5B Chat')
    message = gr.Textbox(lines=1, label='Your message')
    
    with gr.Row():
        clear_history = gr.Button("🧹 Clear history")
        submit = gr.Button("🚀 Send")
    
    # New section for example prompts
    gr.Markdown("### Example prompts")
    with gr.Row():
        example1 = gr.Button("🦄 Unicorn Discovery")
        example2 = gr.Button("⏳ Time Travel Paradox")

    with gr.Accordion("Advanced Settings", open=False):
        max_tokens = gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max New Tokens")
        temperature = gr.Slider(minimum=0.1, maximum=2.0, 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)")

    # Set up event handlers
    message.submit(respond, inputs=[message, chatbot, system_state, max_tokens, temperature, top_p], outputs=[chatbot])
    submit.click(respond, inputs=[message, chatbot, system_state, max_tokens, temperature, top_p], outputs=[chatbot])
    clear_history.click(fn=clear_session, inputs=[], outputs=[message, chatbot])
    modify_system.click(fn=modify_system_session, inputs=[system_input], outputs=[system_state, system_input, chatbot])
    
    # New event handlers for example prompts
    example1.click(
        fn=use_example_and_respond, 
        inputs=[lambda: unicorn_example, chatbot, system_state, max_tokens, temperature, top_p], 
        outputs=[message, chatbot]
    )
    example2.click(
        fn=use_example_and_respond, 
        inputs=[lambda: time_travel_example, chatbot, system_state, max_tokens, temperature, top_p], 
        outputs=[message, chatbot]
    )

    gr.Markdown(
        """
        ## About GPT-2 1.5B
        This is a large language model trained by OpenAI and fine-tuned by Andrej Karpathy. It's capable of generating human-like text based on the input it receives.
        """
    )

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