app.py

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


title = "Code Explainer"
description = "This is a space to convert Python code into english text explaining what it does using [codeparrot-small-code-to-text](https://huggingface.co./codeparrot/codeparrot-small-code-to-text),\
            a code generation model for Python finetuned on [github-jupyter-code-to-text](https://huggingface.co./datasets/codeparrot/github-jupyter-code-to-text) a dataset of Python code followed by a docstring explaining it, the data was originally extracted from Jupyter notebooks."

EXAMPLE_1 = "def bubblesort(elements):\n    n = len(arr)\n# loop through elements\n    swapped = False\n    for n in range(len(elements)-1, 0, -1):\n        for i in range(n):\n            if elements[i] > elements[i + 1]:\n                swapped = True\n                elements[i], elements[i + 1] = elements[i + 1], elements[i]\n            if not swapped:\n                return"
EXAMPLE_2 =  "from sklearn import model_selection\nX_train, X_test, Y_train, Y_test = model_selection.train_test_split(X, Y, test_size=0.2)"
EXAMPLE_3 = "def load_text(file)\n    with open(filename, 'r') as f:\n        text = f.read()\n    return text"
example = [
    [EXAMPLE_1, 32, 0.6, 42],
    [EXAMPLE_2, 16, 0.6, 42],
    [EXAMPLE_3, 11, 0.2, 42],
    ]

# change model to the finetuned one
tokenizer = AutoTokenizer.from_pretrained("codeparrot/codeparrot-small-code-to-text")
model = AutoModelForCausalLM.from_pretrained("codeparrot/codeparrot-small-code-to-text")

def make_doctring(gen_prompt):
    return gen_prompt + f"\n\n\"\"\"\nExplanation:"

def code_generation(gen_prompt, max_tokens, temperature=0.6, seed=42):
    set_seed(seed)
    pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
    prompt = make_doctring(gen_prompt)
    generated_text = pipe(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="Python code"),
        gr.inputs.Slider(
            minimum=8,
            maximum=256,
            step=1,
            default=8,
            label="Number of tokens to generate",
        ),
        gr.inputs.Slider(
            minimum=0,
            maximum=2.5,
            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 explanation", lines=10),
    examples=example,
    layout="horizontal",
    theme="peach",
    description=description,
    title=title
)
iface.launch()