| --- |
| license: other |
| task_categories: |
| - text-generation |
| language: |
| - code |
| - en |
| size_categories: |
| - 1M<n<10M |
| pretty_name: EB-Sky Python Domain 10GB |
| --- |
| |
| # EB-Sky Python Domain (10GB) |
|
|
| ~10GB of raw Python domain knowledge for continued pretraining: |
| ~8GB of source code plus Stack Overflow Python Q&A. |
| (2.49GB as stored — zstd-compressed parquet shards.) |
|
|
| ## Sources |
| - **Code:** [codeparrot/codeparrot-clean](https://huggingface.co/datasets/codeparrot/codeparrot-clean) |
| - **Q&A:** [koutch/stackoverflow_python](https://huggingface.co/datasets/koutch/stackoverflow_python) |
|
|
| ## Processing |
| - Size filter (200–200,000 chars per document) |
| - Auto-generated file removal (header heuristics) |
| - Hardcoded-secret regex scan, email redaction |
| - Exact deduplication (SHA-1) |
| - Q&A: HTML stripped, minimum score 1 for question and answer |
|
|
| ## Format |
| | Column | Description | |
| |---|---| |
| | `text` | the document content | |
| | `source` | `github-code` or `stackoverflow-qa` | |
|
|
| ## Load it |
| ```python |
| from datasets import load_dataset |
| ds = load_dataset("EB-Sky/python-domain-10gb", split="train") |
