app.py

# app.py
import gradio as gr
from transformers import pipeline

# ---- Load model via pipeline ----
MODEL_NAME = "vicgalle/gpt2-open-instruct-v1"
pipe = pipeline("text-generation", model=MODEL_NAME, device_map="auto")

# ---- Inference function ----
def generate_response(instruction, max_new_tokens=150, temperature=0.7, top_k=50, top_p=0.9, rep_pty=1.2):
    system_prompt = f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.
### Instruction:
{instruction}
### Response:
"""
    output = pipe(
        system_prompt,
        max_new_tokens=max_new_tokens,
        temperature=temperature,
        top_k=top_k,
        top_p=top_p,
        do_sample=True,
        pad_token_id=pipe.tokenizer.eos_token_id,
        repetition_penalty=rep_pty
    )
    # Clean up output text
    text = output[0]["generated_text"]
    return text.split("### Response:")[-1].strip()

# ---- Gradio UI ----
with gr.Blocks() as demo:
    gr.Markdown("# 🛸 GPT-2 Open Instruct Playground\nFine-tuned with Open Instruct v1!")
    with gr.Row():
        with gr.Column(scale=4):
            instruction = gr.Textbox(label="Instruction", value="What is the capital city of France?", lines=6)
            output_box = gr.Textbox(label="Model Output", lines=10)
        with gr.Column(scale=1):
            generate_btn = gr.Button("Generate ✨")
            max_new_tokens = gr.Slider(50, 500, value=150, step=10, label="Max new tokens")
            temperature = gr.Slider(0.1, 1.5, value=0.7, step=0.05, label="Temperature")
            top_k = gr.Slider(10, 100, value=50, step=5, label="Top-K sampling")
            top_p = gr.Slider(0.1, 1.0, value=0.9, step=0.05, label="Top-P (nucleus) sampling")
            rep_pty = gr.Slider(label="Repetition Penalty",info="Token repetition penalty. 1.0 means no penalty.",minimum=1.0, maximum=2.0, step=0.01, value=1.2)
            
    generate_btn.click(generate_response, [instruction, max_new_tokens, temperature, top_k, top_p, rep_pty], output_box)

# ---- Launch ----
if __name__ == "__main__":
    demo.launch()