dancinlab/hexa-forge-corpus-stack-v2-sample-v0.1.3

hexa-forge — Stack v2 5% permissive sample (v0.1.3)

Repository: dancinlab/hexa-forge-corpus-stack-v2-sample-v0.1.3 Phase: v0.1.3 G-BASE (RESEARCH_FIRST) Forge commit: 20204bacfe Generated: 2026-05-11

What this is

Deterministic 5% sample of the bigcode/the-stack-v2 permissive subset across the hexa-forge core languages: Python, Rust, TypeScript, Go, C, Zig. Produced by tool/stack_v2_sample.py ; selection is by stable BLAKE2b hash of repo/path (seed 42), so the sample is reproducible from the same source snapshot.

Companion to papers/datasets-source-manifest.md STRUCT row pretrain-bias. Used to validate the ~600B-token estimate by direct measurement on the sampled subset and extrapolation to 100%.

Files

• 22646 payload files
• total size: 259.04 MB
• per-language directory layout: python/, rust/, typescript/, go/, c/, zig/
• licenses.jsonl sidecar per language (repo, path, license, permissive) for provenance
• runs/stack_v2_sample.json (when present) — aggregate token counts

License

Each file inherits its upstream license tag from the-stack-v2's permissive config. The aggregate is permissive-only ; see licenses.jsonl for the per-file tag.

This dataset card is Apache-2.0.

Provenance

No local license audit was found alongside the source path. Run tool/license_clean_scan.py --path /home/summer/runs/corpus/stack-v2-v0.1.3 --report /home/summer/runs/corpus/stack-v2-v0.1.3/license_audit.json before publishing for a permissive-only attestation.

Cross-reference

• papers/plan-runbook-v0.1.3.md §4.1 — generation step
• papers/datasets-source-manifest.md — STRUCT row
• tool/license_clean_scan.py — gate that produced the audit block above

RESEARCH_FIRST notice

This artefact is a v0.1.x research deliverable. The full corpus, the final tokenizer, and any model weights derived from this sample will land at v1.0.0 with RELEASE semantics.