This model is an attempt to simplify code understanding by generating line by line explanation of a source code. This model was fine-tuned using the Salesforce/codet5-large model. Currently it is trained on a small subset of Python snippets.

Model Usage

from transformers import (
    AutoModelForSeq2SeqLM,
    AutoTokenizer,
    AutoConfig,
    pipeline,
)

model_name = "sagard21/python-code-explainer"

tokenizer = AutoTokenizer.from_pretrained(model_name, padding=True)

model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

config = AutoConfig.from_pretrained(model_name)

model.eval()

pipe = pipeline("summarization", model=model_name, config=config, tokenizer=tokenizer)

raw_code = """
def preprocess(text: str) -> str:
    text = str(text)
    text = text.replace("\n", " ")
    tokenized_text = text.split(" ")
    preprocessed_text = " ".join([token for token in tokenized_text if token])

    return preprocessed_text
"""

print(pipe(raw_code)[0]["summary_text"])

Validation Metrics

Loss: 2.156
Rouge1: 29.375
Rouge2: 18.128
RougeL: 25.445
RougeLsum: 28.084
Gen Len: 19.000