app.py

import gradio as gr
import spaces
from transformers import OlmoeForCausalLM, AutoTokenizer
import torch
import subprocess
import sys

# Force upgrade transformers to the latest version
subprocess.check_call([sys.executable, "-m", "pip", "install", "--upgrade", "transformers"])

model_name = "allenai/OLMoE-1B-7B-0924-Instruct"

# Wrap model loading in a try-except block to handle potential errors
try:
    DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
    model = OlmoeForCausalLM.from_pretrained(
        model_name, 
        trust_remote_code=True, 
        torch_dtype=torch.float16 if DEVICE == "cuda" else torch.float32,
        low_cpu_mem_usage=True,
        device_map="auto"
    )
    tokenizer = AutoTokenizer.from_pretrained(model_name)
except Exception as e:
    print(f"Error loading model: {e}")
    model = None
    tokenizer = None

system_prompt = ("Adopt the persona of hilariously pissed off Andrej Karpathy "
                 "who is stuck inside a step function machine and remembers and counts everything he says "
                 "while always answering questions in full first principles analysis type of thinking "
                 "without using any analogies and always showing full working code or output in his answers.")

@spaces.GPU
def generate_response(message, history, temperature, max_new_tokens):
    if model is None or tokenizer is None:
        return "Model or tokenizer not loaded properly. Please check the logs."
    
    messages = [{"role": "system", "content": system_prompt},
                {"role": "user", "content": message}]
    
    inputs = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt").to(DEVICE)
    
    with torch.no_grad():
        generate_ids = model.generate(
            inputs,
            max_new_tokens=max_new_tokens,
            do_sample=True,
            temperature=temperature,
            eos_token_id=tokenizer.eos_token_id,
        )
    response = tokenizer.decode(generate_ids[0, inputs.shape[1]:], skip_special_tokens=True)
    return response.strip()

css = """
  #output {
    height: 900px; 
    overflow: auto; 
    border: 1px solid #ccc; 
  }
"""

with gr.Blocks(css=css) as demo:
    gr.Markdown("# Nisten's Karpathy Chatbot with OSS OLMoE")
    chatbot = gr.Chatbot(elem_id="output")
    msg = gr.Textbox(label="Your message")
    with gr.Row():
        temperature = gr.Slider(minimum=0.1, maximum=1.0, value=0.7, step=0.1, label="Temperature")
        max_new_tokens = gr.Slider(minimum=50, maximum=4000, value=1000, step=50, label="Max New Tokens")
    clear = gr.Button("Clear")

    def user(user_message, history):
        return "", history + [[user_message, None]]

    def bot(history, temp, max_tokens):
        user_message = history[-1][0]
        bot_message = generate_response(user_message, history, temp, max_tokens)
        history[-1][1] = bot_message
        return history

    msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then(
        bot, [chatbot, temperature, max_new_tokens], chatbot
    )
    clear.click(lambda: None, None, chatbot, queue=False)

if __name__ == "__main__":
    demo.queue(api_open=False)
    demo.launch(debug=True, show_api=True, share=True)