kd13/stack-v2-mini

---
license: bigcode-openrail-m
task_categories:
- text-generation
- fill-mask
language:
- en
pretty_name: mini-multicode
size_categories:
- 100K<n<1M
---

# Stack v2 Clean — 200K Multi-Language Code Subset

A cleaned and filtered subset of [bigcode/the-stack-v2](https://huggingface.co/datasets/bigcode/the-stack-v2).
## Dataset Summary

This dataset was assembled to support fine-tuning models such as ModernBERT, Llama, Qwen on programming language data. It provides a balanced, deduplicated, and filtered collection of real-world source files drawn from public repositories indexed by Software Heritage and curated by the BigCode project.

| Language | Files |
|-------------|---------|
| Python | 40,001 |
| JavaScript | 40,001 |
| Java | 40,001 |
| C++ | 40,001 |
| Go | 40,001 |
| **Total** | **200,005** |

## Limitations and Considerations

This dataset is a relatively small sample of the full Stack v2 corpus and is not intended for training large code generation models from scratch. Files retain their original licenses as classified upstream, and users are responsible for verifying license compatibility with their downstream use cases. No personally identifiable information removal pass has been applied beyond what is present in the upstream Stack v2 release; users redistributing derivative artifacts should consider running a PII scrubbing pass such as `bigcode-pii` before publication. Near-duplicate detection (for example, MinHash-based) was not applied and may be beneficial for some training scenarios.

## Source and License

The underlying data originates from `bigcode/the-stack-v2`, which is governed by the BigCode project's terms of use and the original repository licenses of each source file. Users of this derivative dataset must comply with the upstream Stack v2 terms, available on the [original dataset page](https://huggingface.co/datasets/bigcode/the-stack-v2). The `license_type` field is preserved from the upstream dataset to support license-aware filtering.

## Citation

If you use this dataset, please cite the original Stack v2 release:

```
@article{lozhkov2024starcoder,
title={StarCoder 2 and The Stack v2: The Next Generation},
author={Lozhkov, Anton and others},
journal={arXiv preprint arXiv:2402.19173},
year={2024}
}
```