app.py

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

title = "SantaCoder 🎅 Swift 🍏 Completion"
description = "This is a subspace to make code generation with [SantaCoder fine-tuned on The Stack Swift](https://huggingface.co/mrm8488/santacoder-finetuned-the-stack-swift)"
EXAMPLE_0 = """// Swift "Hello, World!" Program"""


CKPT = "mrm8488/santacoder-finetuned-the-stack-swift"

examples = [[EXAMPLE_0, 8, 0.6, 42]]
tokenizer = AutoTokenizer.from_pretrained(CKPT)
model = AutoModelForCausalLM.from_pretrained(CKPT, trust_remote_code=True).to("cuda")


def code_generation(gen_prompt, max_tokens, temperature=0.6, seed=42):
    set_seed(seed)
    pipe = pipeline("text-generation", model=model, tokenizer=tokenizer, device=0)
    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


iface = gr.Interface(
    fn=code_generation, 
    inputs=[
        gr.Textbox(lines=10, label="Input code"),
        gr.inputs.Slider(
            minimum=8,
            maximum=256,
            step=1,
            default=8,
            label="Number of tokens to generate",
        ),
        gr.inputs.Slider(
            minimum=0,
            maximum=2,
            step=0.1,
            default=0.6,
            label="Temperature",
        ),
        gr.inputs.Slider(
            minimum=0,
            maximum=1000,
            step=1,
            default=42,
            label="Random seed to use for the generation"
        )
    ],
    outputs=gr.Textbox(label="Predicted code", lines=10),
    examples=examples,
    layout="horizontal",
    theme="peach",
    description=description,
    title=title
)
iface.launch()