app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, AutoConfig

# use the official bitnet package to supply the missing code
from bitnet.configuration_bitnet import BitNetConfig
from bitnet.modeling_bitnet import BitNetForCausalLM
from bitnet.tokenization_bitnet import BitNetTokenizer

# Singleton for model and tokenizer
_model = None
_tokenizer = None

def load_model():
    global _model, _tokenizer
    if _model is None or _tokenizer is None:
        model_id = "microsoft/bitnet-b1.58-2B-4T"
        # load tokenizer, config, and model from the bitnet pip package
        _tokenizer = BitNetTokenizer.from_pretrained(model_id)
        config     = BitNetConfig.from_pretrained(model_id)
        _model     = BitNetForCausalLM.from_pretrained(
            model_id,
            config=config,
            torch_dtype=torch.bfloat16
        )
    return _model, _tokenizer

def manage_history(history):
    # Limit to 3 turns (each turn is user + assistant = 2 messages)
    max_messages = 6  # 3 turns * 2 messages per turn
    if len(history) > max_messages:
        history = history[-max_messages:]
    
    # Limit total character count to 300
    total_chars = sum(len(msg["content"]) for msg in history)
    while total_chars > 300 and history:
        history.pop(0)  # Remove oldest message
        total_chars = sum(len(msg["content"]) for msg in history)
    
    return history

def generate_response(user_input, system_prompt, max_new_tokens, temperature, top_p, top_k, history):
    model, tokenizer = load_model()
    
    messages = [
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": user_input},
    ]
    
    prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    chat_input = tokenizer(prompt, return_tensors="pt").to(model.device)
    
    # Generate response
    chat_outputs = model.generate(
        **chat_input,
        max_new_tokens=max_new_tokens,
        temperature=temperature,
        top_p=top_p,
        top_k=top_k,
        do_sample=True
    )
    
    # Decode response
    response = tokenizer.decode(chat_outputs[0][chat_input['input_ids'].shape[-1]:], skip_special_tokens=True)
    
    # Update history
    history.append({"role": "user", "content": user_input})
    history.append({"role": "assistant", "content": response})
    
    # Manage history limits
    history = manage_history(history)
    
    return history, history

# Gradio interface
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown("# BitNet b1.58 2B4T Demo")
    
    with gr.Row():
        with gr.Column():
            gr.Markdown("""
            ## About BitNet b1.58 2B4T
            BitNet b1.58 2B4T is the first open-source, native 1-bit Large Language Model with 2 billion parameters, developed by Microsoft Research. Trained on 4 trillion tokens, it matches the performance of full-precision models while offering significant efficiency gains in memory, energy, and latency. Features include:
            - Transformer-based architecture with BitLinear layers
            - Native 1.58-bit weights and 8-bit activations
            - Maximum context length of 4096 tokens
            - Optimized for efficient inference with bitnet.cpp
            """)
        
        with gr.Column():
            gr.Markdown("""
            ## About Tonic AI
            Tonic AI is a vibrant community of AI enthusiasts and developers always building cool demos and pushing the boundaries of what's possible with AI. We're passionate about creating innovative, accessible, and engaging AI experiences for everyone. Join us in exploring the future of AI!
            """)
    
    with gr.Row():
        with gr.Column():
            user_input = gr.Textbox(label="Your Message", placeholder="Type your message here...")
            system_prompt = gr.Textbox(
                label="System Prompt",
                value="You are a helpful AI assistant.",
                placeholder="Enter system prompt..."
            )
            
            with gr.Accordion("Advanced Options", open=False):
                max_new_tokens = gr.Slider(
                    minimum=10,
                    maximum=500,
                    value=50,
                    step=10,
                    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=0.9,
                    step=0.05,
                    label="Top P"
                )
                top_k = gr.Slider(
                    minimum=1,
                    maximum=100,
                    value=50,
                    step=1,
                    label="Top K"
                )
            
            submit_btn = gr.Button("Send")
        
        with gr.Column():
            chatbot = gr.Chatbot(label="Conversation", type="messages")
    
    chat_history = gr.State([])
    
    submit_btn.click(
        fn=generate_response,
        inputs=[
            user_input,
            system_prompt,
            max_new_tokens,
            temperature,
            top_p,
            top_k,
            chat_history
        ],
        outputs=[chatbot, chat_history]
    )

if __name__ == "__main__":
    # Preload model to avoid threading issues
    load_model()
    demo.launch(ssr_mode=False, share=True)