Learning and Evaluating Contextual Embedding of Source Code

TL;DR

This paper introduces CuBERT, a pre-trained contextual embedding model for source code, trained on a large Python corpus, and evaluates it across multiple program-understanding tasks, demonstrating superior performance over existing models.

Contribution

The paper develops CuBERT, a high-quality pre-trained code embedding, and provides a comprehensive benchmark for evaluating source code understanding models.

Findings

CuBERT outperforms existing models on multiple tasks.

Pre-training on large code corpus improves understanding accuracy.

Fewer labeled examples needed for fine-tuning CuBERT.

Abstract

Recent research has achieved impressive results on understanding and improving source code by building up on machine-learning techniques developed for natural languages. A significant advancement in natural-language understanding has come with the development of pre-trained contextual embeddings, such as BERT, which can be fine-tuned for downstream tasks with less labeled data and training budget, while achieving better accuracies. However, there is no attempt yet to obtain a high-quality contextual embedding of source code, and to evaluate it on multiple program-understanding tasks simultaneously; that is the gap that this paper aims to mitigate. Specifically, first, we curate a massive, deduplicated corpus of 7.4M Python files from GitHub, which we use to pre-train CuBERT, an open-sourced code-understanding BERT model; and, second, we create an open-sourced benchmark that comprises five classification tasks and one program-repair task, akin to code-understanding tasks proposed in the literature before. We fine-tune CuBERT on our benchmark tasks, and compare the resulting models to different variants of Word2Vec token embeddings, BiLSTM and Transformer models, as well as published state-of-the-art models, showing that CuBERT outperforms them all, even with shorter training, and with fewer labeled examples. Future work on source-code embedding can benefit from reusing our benchmark, and from comparing against CuBERT models as a strong baseline.