app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM, set_seed
from transformers import pipeline
import os

description = """# SantaCoder Endpoint"""
token = os.environ["HUB_TOKEN"]
device="cuda:0"

tokenizer = AutoTokenizer.from_pretrained("bigcode/christmas-models", use_auth_token=token)
model = AutoModelForCausalLM.from_pretrained("bigcode/christmas-models", trust_remote_code=True, use_auth_token=token)


def code_generation(gen_prompt, max_tokens, temperature=0.6, seed=42):
    set_seed(seed)
    pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
    generated_text = pipe(gen_prompt, do_sample=True, top_p=0.95, temperature=temperature, max_new_tokens=max_tokens)[0]['generated_text']
    return generated_text


demo = gr.Blocks()
with demo:
    with gr.Row():
        gr.Markdown(value=description)
    with gr.Row():
        with gr.Column():
            code = gr.Textbox(lines=10, label="Input code")
            max_tokens= gr.Slider(
                minimum=8,
                maximum=1000,
                step=1,
                label="Number of tokens to generate",
            )
            temperature = gr.Slider(
                minimum=0.1,
                maximum=2.5,
                step=0.1,
                label="Temperature",
            )
            seed = gr.Slider(
                minimum=0,
                maximum=1000,
                step=1,
                label="Random seed to use for the generation"
            )
            run = gr.Button()
        with gr.Column():
            output = gr.Textbox(lines=10, label="Generated code")

    event = run.click(code_generation, [code, max_tokens, temperature, seed], output)
    gr.HTML(label="Contact", value="<img src='https://huggingface.co/datasets/bigcode/admin/resolve/main/bigcode_contact.png' alt='contact' style='display: block; margin: auto; max-width: 800px;'>")

demo.launch()