the-stack-go / README.md
s0duku's picture
Add files using upload-large-folder tool
083f6a0 verified
|
Raw History Blame Contribute Delete
1.22 kB
# the-stack-go-ge2
This dataset is a merged, multi-range, star-filtered export of The Stack v2. The repository snapshot is split into two coordinated datasets:
- `repo-meta/`: repository and file metadata, grouped into 64 source shards.
- `repo-contents/`: raw file bytes, grouped by the same 64 source shards and further split into Parquet files.
For a file, read `blob_id` from `repo-meta/<shard>/repositories.parquet` and find the matching row in `repo-contents/<shard>/*.parquet`. The join key is **`(shard, blob_id)`**. The shard component is required because the downloader deduplicates blobs within each source shard; a merged export also deduplicates blobs within each shard across source ranges.
The metadata records preserve the repository name, revision, path, language, license and the file's `blob_id`. Content rows contain `blob_id`, raw `content`, actual length, indexed length and length comparison status. `content` is stored as bytes.
The export uses the strict snapshot filter `star_events_count > 1`. This is the Stack v2 index's `star_events_count`, not a live GitHub star count. Source licenses and removal requests remain governed by the original dataset and upstream content providers.