StarCoder2
Family overview — summarizes releases; licenses and availability belong to each release.
Maintained by BigCode project (open scientific collaboration)24, Hugging Face (BigCode co-steward)23, ServiceNow (BigCode co-steward)23
StarCoder2 is a family of code generation models from the BigCode project, released on February 28, 2024 in 3B, 7B, and 15B sizes and trained on The Stack v2. Per Hugging Face's announcement, ServiceNow trained the 3B model, Hugging Face the 7B model, and NVIDIA the 15B model. The bigcode organization on Hugging Face also hosts an instruction-tuned variant, StarCoder2-15B-Instruct-v0.1.315
- Model hub: BigCode on Hugging Face (external site: huggingface.co)
- Repository: StarCoder2 repository (external site: github.com)
- Paper: StarCoder 2 and The Stack v2: The Next Generation (arXiv 2402.19173) (external site: arxiv.org)
- Website: BigCode project website (external site: bigcode-project.org)
Releases assessed
| Release | Weights | Tier (USASI rubric v0.1) | License | Released |
|---|---|---|---|---|
| StarCoder2-15B | Weights: Public | Model-disclosure tier (USASI rubric v0.1): Open-weight | BigCode OpenRAIL-M v1 License Agreement, Apache-2.0 | Feb 28, 2024 |
| StarCoder2-3B | Weights: Public | Model-disclosure tier (USASI rubric v0.1): Open-weight | BigCode OpenRAIL-M v1 License Agreement, Apache-2.0 | Feb 28, 2024 |
What it is useful for
Organization context
Provenance and derivatives
StarCoder2 was trained by the BigCode collaboration on The Stack v2, which was built in partnership with Software Heritage from its public source-code archive, plus other sources such as GitHub issues and pull requests, Kaggle notebooks, Wikipedia, and arXiv.1367
- Derived from: The Stack v2 (external site: huggingface.co) — Pretraining data built from the Software Heritage archive; no separate catalog record.
U.S. eligibility
Eligible · basis: U.S.-governed project
Multi-organization assessment. The StarCoder2 report says the BigCode project is stewarded by ServiceNow and Hugging Face, and Hugging Face's announcement describes it as led jointly by the two. ServiceNow, Inc. is headquartered in Santa Clara, California (2025 Form 10-K). Hugging Face's privacy policy names Hugging Face, Inc. as the provider of its services and says the company and its servers are located in the United States, with Hugging Face SAS in Paris as its EU main establishment (see the Hugging Face record for the full dual-country assessment). Contributors to BigCode come from many institutions, including outside the United States; eligibility rests on the two documented stewards, not on contributors. Each release is assessed on its own record.2389
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.