OSSDatasets

0.3.7 verified Sat May 09 auth: no python

OSSDatasets provides scalable access to Software Engineering datasets such as SWE-bench and GitHub issue data. It downloads and caches datasets locally for reproducible ML experiments. Current version 0.3.7, release cadence irregular.

pip install ossdata

Common errors

error ModuleNotFoundError: No module named 'ossdata' ↓

cause Package not installed or installed in wrong environment.

fix

Run pip install ossdata in your active Python environment.

error ValueError: Dataset 'swe_bench' not found. Available datasets: ['swe-bench', 'swe-bench-lite', 'github-issues'] ↓

cause Dataset name typo (underscore instead of hyphen).

fix

Use exact dataset name with hyphens: load_dataset('swe-bench', ...).

Warnings

gotcha The dataset name must be exact; allowed names: 'swe-bench', 'swe-bench-lite', 'github-issues' (case-sensitive). ↓

fix Check the exact dataset name in the docs.

gotcha The returned data format is a list of dicts, not a Hugging Face Dataset or pandas DataFrame. You may need to convert manually. ↓

fix Wrap with `pd.DataFrame(dataset)` or use dict-comprehensions.

breaking Version 0.2.x and earlier used a different API: `import ossdata` then `ossdata.get_dataset(...)`. The old API is removed in 0.3.0. ↓

fix Use `from ossdata import load_dataset` and call `load_dataset(...)`.

Imports

load_dataset
```
from ossdata import load_dataset
```

Quickstart

Load a dataset by name (e.g., 'swe-bench') with automatic caching. Returns a list of dicts.

import os
from ossdata import load_dataset

# Load SWE-bench dataset (cached automatically)
dataset = load_dataset('swe-bench', split='train')
print(dataset[:2])