app1.py

import gradio as gr
from huggingface_hub import InferenceClient

client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")

with gr.Blocks() as demo:
    logo = gr.Image(value="(link unavailable)", label="Logo")
    input_box = gr.Textbox(label="Type your message...")
    chat_history = gr.Markdown(label="Chat History")
    chatbot = gr.Chatbot()

    def respond(message):
        # Set default values for sliders and system message
        system_message = "You are a friendly Chatbot."
        max_tokens = 512
        temperature = 0.7
        top_p = 0.95

        # Generate response from chatbot
        messages = [{"role": "system", "content": system_message}]
        messages.append({"role": "user", "content": message})
        response = ""
        for message in client.chat_completion(messages, max_tokens=2048, stream=True, temperature=0.2, top_p=.05):
            token = message.choices[0].delta.content
            response += token

        # Update chat history
        chat_history_value = [
           ("User", "Hello!"),
           ("Assistant", "Hello! How can I help you today?"),
           ("User", "I need help with..."),
           ("Assistant", "I'd be happy to help!"),
         ]
        chat_history.update(value=chat_history_value)
        chat_history_value.append((message, response))
        chat_history.update(value=chat_history_value)

        return chat_history

    input_box.submit(respond, inputs=input_box, outputs=chat_history)

demo.launch()