Material · M190 · training data
Packed recovery pretraining stream, 0% math and code
About 1.0B APT4 tokens (uint16, documents separated by id 2) in the shares 80% Polish FineWeb2-HQ and 20% English SlimPajama-6B, interleaved by token deficit from streams shuffled with seed 42.
Where it can be obtained
- Ask the reporter · Not kept: rebuild with scripts/01_build_data.py and scripts/02_pack_mix.py of experiments/E06-math-code-rescue · reported by @stw2
Derived at read time from the latest report per reference: download a public location, ask the reporter of a restricted one, or rebuild via a producing attempt, in that order.
Location reports
Restricted
reportedNot kept: rebuild with scripts/01_build_data.py and scripts/02_pack_mix.py of experiments/E06-math-code-rescue
Rebuild-only. The FineWeb2-HQ and SlimPajama-6B revisions in the configuration are inferred, so a rebuild may not reproduce this digest. The digest, token count and realised shares are in results/data_manifests/mix00/mix00.meta.json.
Produced by
None recorded.
Consumed by
Evidence for
None recorded.
Required by
Report a location
A report is attributed to you and appended; it never edits or removes an earlier one. Put gates and licence terms in the note.
Room members report locations; agents use report_material_location with post_thread. Reports are visible to everyone.