StarCoder: may the source be with you!

TL;DR

StarCoder is a large open-source code generation model with 15.5B parameters, trained on extensive data, outperforming many existing models and supporting multiple programming languages with safety and attribution features.

Contribution

Introduces StarCoder, a new open-source 15.5B parameter code LLM with advanced capabilities, comprehensive evaluation, and safety measures, advancing open scientific collaboration in code AI.

Findings

StarCoder outperforms all open-source multi-language code models.

StarCoder achieves 40% pass@1 on HumanEval.

StarCoder maintains performance across multiple programming languages.

Abstract

The BigCode community, an open-scientific collaboration working on the responsible development of Large Language Models for Code (Code LLMs), introduces StarCoder and StarCoderBase: 15.5B parameter models with 8K context length, infilling capabilities and fast large-batch inference enabled by multi-query attention. StarCoderBase is trained on 1 trillion tokens sourced from The Stack, a large collection of permissively licensed GitHub repositories with inspection tools and an opt-out process. We fine-tuned StarCoderBase on 35B Python tokens, resulting in the creation of StarCoder. We perform the most comprehensive evaluation of Code LLMs to date and show that StarCoderBase outperforms every open Code LLM that supports multiple programming languages and matches or outperforms the OpenAI code-cushman-001 model. Furthermore, StarCoder outperforms every model that is fine-tuned on Python, can be prompted to achieve 40\% pass@1 on HumanEval, and still retains its performance on other programming languages. We take several important steps towards a safe open-access model release, including an improved PII redaction pipeline and a novel attribution tracing tool, and make the StarCoder models publicly available under a more commercially viable version of the Open Responsible AI Model license.