Datasets:
|
Download README.md from placeholderlabs/exp-pool-commit-code-dolma2-tokenized: direct link, hf CLI and curl.
- Browser
- Download file 932 Bytes
-
https://huggingface.co/datasets/placeholderlabs/exp-pool-commit-code-dolma2-tokenized/resolve/main/README.md
- Command line
-
hf download hf://datasets/placeholderlabs/exp-pool-commit-code-dolma2-tokenized/README.md
-
curl -L -o README.md https://huggingface.co/datasets/placeholderlabs/exp-pool-commit-code-dolma2-tokenized/resolve/main/README.md
932 Bytes
metadata
license: other
task_categories:
- text-generation
language:
- en
pretty_name: Locus EXP Commit Code (Dolma 2 tokenized)
Locus EXP Commit Code - Dolma 2 tokenized
Pretokenized experiment pool for reproducible proxy-training runs.
MANIFEST.jsonis the authoritative schema, provenance, checksums, and train/holdout assignment.shards/shard-NNNNN/tokens.binstores little-endian int32 token IDs.offsets.binstores little-endian int64 document boundaries.index.parquetstores document IDs, offsets, and compact filter fields.metadata.parquetstores complete source metadata and is downloaded only for explicit cold-metadata filters.
Each document already ends in exactly one EOD token. Training must not add another document-level EOD.
Commit examples use target_start_token to mask the commit-message prompt and train only on the unified-diff target. Each commit document must fit one sequence.