Magicoder: Empowering Code Generation with OSS-Instruct

TL;DR

Magicoder introduces open-source large language models for code, trained on synthetic instruction data enhanced with open-source code snippets, achieving competitive performance and surpassing some state-of-the-art models on coding benchmarks.

Contribution

The paper presents OSS-Instruct, a novel method for generating diverse instruction data using open-source code, significantly improving code LLM performance with models under 7B parameters.

Findings

Magicoder models outperform similar-sized state-of-the-art code models.

MagicoderS-CL-7B surpasses ChatGPT on HumanEval+ benchmark.

Open-source data enhances the realism and controllability of synthetic instructions.

Abstract

We introduce Magicoder, a series of fully open-source (code, weights, and data) Large Language Models (LLMs) for code that significantly closes the gap with top code models while having no more than 7B parameters. Magicoder models are trained on 75K synthetic instruction data using OSS-Instruct, a novel approach to enlightening LLMs with open-source code snippets to generate diverse instruction data for code. Our main motivation is to mitigate the inherent bias of the synthetic data generated by LLMs through the wealth of open-source references for the production of more realistic and controllable data. The orthogonality of OSS-Instruct and other data generation methods like Evol-Instruct further enables us to build an enhanced MagicoderS. Both Magicoder and MagicoderS substantially outperform state-of-the-art code models with similar or even larger sizes on a wide range of coding benchmarks. Notably, MagicoderS-CL-7B based on CodeLlama even surpasses the prominent ChatGPT on HumanEval+ (66.5 vs. 65.9 in pass@1 ). Overall, OSS-Instruct opens a new direction for crafting diverse synthetic instruction data for code using abundant open-source references.