| license: other | |
| task_categories: | |
| - text-generation | |
| language: | |
| - code | |
| size_categories: | |
| - 1M<n<10M | |
| # EB-Sky Python Code (10GB) | |
| ~10GB of filtered Python source code for domain-adaptive pretraining. | |
| ## Source | |
| Filtered subset of | |
| [codeparrot/codeparrot-clean](https://huggingface.co/datasets/codeparrot/codeparrot-clean) | |
| (originally collected from public GitHub repositories). | |
| ## Processing | |
| - Size filter: 200–200,000 characters per file | |
| - Removed auto-generated files (header heuristics) | |
| - Dropped files matching common hardcoded-secret patterns | |
| - Exact deduplication (SHA-1) | |
| - Email addresses redacted | |
| ## Format | |
| Single `text` column. Parquet shards under `data/`. | |
| ## Load it | |
| ```python | |
| from datasets import load_dataset | |
| ds = load_dataset("EB-Sky/python-code-10gb", split="train") | |
| ``` | |
| ## License | |
| Code retains the original licenses of its source repos. | |
| Review before commercial use. | |