Evaluation of LLMs on Syntax-Aware Code Fill-in-the-Middle Tasks

Linyuan Gong, Sida Wang, Mostafa Elhoushi, Alvin Cheung
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:15907-15928, 2024.

Abstract

We introduce Syntax-Aware Fill-in-the-Middle (SAFIM), a new benchmark for evaluating Large Language Models (LLMs) on the code Fill-in-the-Middle (FIM) task. This benchmark focuses on syntax-aware completions of program structures such as code blocks and conditional expressions, and includes 17,720 examples from multiple programming languages, sourced from recent code submissions after April 2022 to minimize data contamination. SAFIM provides a robust framework with various prompt designs and novel syntax-aware post-processing techniques, facilitating accurate and fair comparisons across LLMs. Our comprehensive evaluation of 15 LLMs shows that FIM pretraining not only enhances FIM proficiency but also improves Left-to-Right (L2R) inference using LLMs. Our findings challenge conventional beliefs and suggest that pretraining methods and data quality have more impact than model size. SAFIM thus serves as a foundational platform for future research in effective pretraining strategies for code LLMs. The evaluation toolkit and dataset are available at https://github.com/gonglinyuan/safim, and the leaderboard is available at https://safimbenchmark.com.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-gong24f, title = {Evaluation of {LLM}s on Syntax-Aware Code Fill-in-the-Middle Tasks}, author = {Gong, Linyuan and Wang, Sida and Elhoushi, Mostafa and Cheung, Alvin}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {15907--15928}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/gong24f/gong24f.pdf}, url = {https://proceedings.mlr.press/v235/gong24f.html}, abstract = {We introduce Syntax-Aware Fill-in-the-Middle (SAFIM), a new benchmark for evaluating Large Language Models (LLMs) on the code Fill-in-the-Middle (FIM) task. This benchmark focuses on syntax-aware completions of program structures such as code blocks and conditional expressions, and includes 17,720 examples from multiple programming languages, sourced from recent code submissions after April 2022 to minimize data contamination. SAFIM provides a robust framework with various prompt designs and novel syntax-aware post-processing techniques, facilitating accurate and fair comparisons across LLMs. Our comprehensive evaluation of 15 LLMs shows that FIM pretraining not only enhances FIM proficiency but also improves Left-to-Right (L2R) inference using LLMs. Our findings challenge conventional beliefs and suggest that pretraining methods and data quality have more impact than model size. SAFIM thus serves as a foundational platform for future research in effective pretraining strategies for code LLMs. The evaluation toolkit and dataset are available at https://github.com/gonglinyuan/safim, and the leaderboard is available at https://safimbenchmark.com.} }
Endnote
%0 Conference Paper %T Evaluation of LLMs on Syntax-Aware Code Fill-in-the-Middle Tasks %A Linyuan Gong %A Sida Wang %A Mostafa Elhoushi %A Alvin Cheung %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-gong24f %I PMLR %P 15907--15928 %U https://proceedings.mlr.press/v235/gong24f.html %V 235 %X We introduce Syntax-Aware Fill-in-the-Middle (SAFIM), a new benchmark for evaluating Large Language Models (LLMs) on the code Fill-in-the-Middle (FIM) task. This benchmark focuses on syntax-aware completions of program structures such as code blocks and conditional expressions, and includes 17,720 examples from multiple programming languages, sourced from recent code submissions after April 2022 to minimize data contamination. SAFIM provides a robust framework with various prompt designs and novel syntax-aware post-processing techniques, facilitating accurate and fair comparisons across LLMs. Our comprehensive evaluation of 15 LLMs shows that FIM pretraining not only enhances FIM proficiency but also improves Left-to-Right (L2R) inference using LLMs. Our findings challenge conventional beliefs and suggest that pretraining methods and data quality have more impact than model size. SAFIM thus serves as a foundational platform for future research in effective pretraining strategies for code LLMs. The evaluation toolkit and dataset are available at https://github.com/gonglinyuan/safim, and the leaderboard is available at https://safimbenchmark.com.
APA
Gong, L., Wang, S., Elhoushi, M. & Cheung, A.. (2024). Evaluation of LLMs on Syntax-Aware Code Fill-in-the-Middle Tasks. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:15907-15928 Available from https://proceedings.mlr.press/v235/gong24f.html.

Related Material