app.py

import os
import gradio as gr
import torch
import spaces
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

# Load model and tokenizer if a GPU is available
if torch.cuda.is_available():
    model_id = "allenai/OLMo-7B-hf"
    adapters_name = "yilunzhao/olmo-finetuned"
    model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", trust_remote_code=True)
    model = PeftModel.from_pretrained(model, adapters_name)
    tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
else:
    raise EnvironmentError("CUDA device not available. Please run on a GPU-enabled environment.")

# Basic function to generate response based on passage and question
@spaces.GPU
def generate_response(passage: str, question: str) -> str:
    # Prepare the input text by combining the passage and question
    chat = [{"role": "user", "content": f"Passage: {passage}\nQuestion: {question}"}]

    prompt = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
    inputs = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt")
    response = model.generate(input_ids=inputs.to(model.device), max_new_tokens=100)
    
    response = tokenizer.batch_decode(response, skip_special_tokens=True)[0].split("<|assistant|>")[-1].strip()
    
    
    return response


# Gradio Interface
with gr.Blocks() as demo:
    gr.Markdown("# Passage and Question Response Generator")
    
    passage_input = gr.Textbox(label="Passage", placeholder="Enter the passage here", lines=5)
    question_input = gr.Textbox(label="Question", placeholder="Enter the question here", lines=2)
    
    output_box = gr.Textbox(label="Response", placeholder="Model's response will appear here")
    
    submit_button = gr.Button("Generate Response")
    submit_button.click(fn=generate_response, inputs=[passage_input, question_input], outputs=output_box)

# Run the app
if __name__ == "__main__":
    demo.launch()