StarCoder: may the source be with you!
By Raymond Li, Loubna Ben Allal, Yangtian Zi, Niklas Muennighoff, Denis Kocetkov, Chenghao Mou, Marc Marone, Christopher Akiki, Jia Li, Jenny Chim and 57 more. Trans. Mach. Learn. Res., 2023
1,263 citations recorded by the source corpus.
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.
Subfields: Artificial Intelligence, Information Systems, Large Language Models, Code Generation & Program Synthesis
View Paper: https://arxiv.org/abs/2305.06161
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Jianfeng Gao
- Mohit Bansal
- Percy Liang
- Luke S. Zettlemoyer
- Yoshua Bengio
- Quoc V. Le
- Graham Neubig
- Yejin Choi
In Information Systems
In Large Language Models
In Code Generation & Program Synthesis
- Ge Li
- Zhi Jin
- Michael Pradel
- Pushmeet Kohli
- Armando Solar-Lezama
- Brendan Dolan-Gavitt
- Benjamin Tan
- Chun Xia
Citations in the corpus
- Gorilla: Large Language Model Connected with Massive APIs (2023, neutral): “Besides prompting, there have also been efforts to pretrain language models specifically for code generation [25, 21, 26].”
- RepoBench: Benchmarking Repository-Level Code Auto-Completion Systems (2023, methodology): “Given that the code in this dataset has been widely utilized for training various models [23, 27], we primarily use this dataset for constructing our training data.”
- TACO: Topics in Algorithmic COde generation dataset (2023, background): “…have made significant advancements in applications involving generating code from a natural language description, establishing themselves as the foundational backbone for a myriad of code-related downstream tasks [Chen et al. , 2021; Nijkamp et al. , 2022; Li et al. , 2023; Rozi`ere et al. , 2023].”
- CodeS: Towards Building Open-source Language Models for Text-to-SQL (2024, neutral): “List bank names whose proportion of estimated liabilities in their total liabilities exceeds the industry average, and whose net increase in borrowing funds from other financial institutions exceeds 3 billion Yuan.”
- AWQ: Activation-aware Weight Quantization for On-Device LLM Compression and Acceleration (2023, positive): “In contrast, our TinyChat supports a wide range of applications, including StarCoder (Li et al., 2023c), StableCode (GPT-NeoX) (Black et al., 2022), Mistral (Jiang et al., 2023), and Falcon (Penedo et al., 2023) while consistently delivering significant speedup over AutoGPTQ.”
- Reflexion: language agents with verbal reinforcement learning (2023, methodology): “Approach Pass@1 accuracy (avg over 8 trials) Pass@1 accuracy (std) Baseline 0.26 0.00481 Reflexion 0.26 0.00305 Table 4: Pass@1 accuracy on HumanEval Python using starchat-beta [13].”
- NoFunEval: Funny How Code LMs Falter on Requirements Beyond Functional Correctness (2024, background): “For instance, compare the scores of StarCoder-15.”
- Demystifying GPT Self-Repair for Code Generation (2023, background): “The use of large language models for program synthesis has been studied extensively in the literature [2, 5, 9, 11, 18, 20, 21, 25, 32].”
- MetaMath: Bootstrap Your Own Mathematical Questions for Large Language Models (2023, background): “…the rapid development of large language models (LLMs) which emerge as the favored approach for various applications and demonstrate multi-dimensional abilities, including instruction following [7, 38, 55, 67], coding assistance [8, 36, 44, 51], and mathematical problem-solving [14, 28, 43, 79].”
- A Comprehensive Overview of Large Language Models (2023, neutral): “The StarCoder trains an encoder to filter names, emails, and other personal data from the training data.”
- Sheared LLaMA: Accelerating Language Model Pre-training via Structured Pruning (2023, background): “7B also surpasses OpenLLaMA-3B-v2, which was trained on 1T tokens from a mixture of RedPajama, RefinedWeb, and StarCoder.”
- Structured Chain-of-Thought Prompting for Code Generation (2023, positive): “, CodeParrot [29], CodeGen [17], CodeGeeX [40], InCoder [8], StarCoder [15] and CodeT5+ [33].”