Skip to content
This repository was archived by the owner on Aug 3, 2026. It is now read-only.

fix: B6 run uses single S3 rung + per-task cap - #36

Merged
bitzic merged 1 commit into
mainfrom
fix/b6-single-s3-rung
Jun 19, 2026
Merged

fix: B6 run uses single S3 rung + per-task cap#36
bitzic merged 1 commit into
mainfrom
fix/b6-single-s3-rung

Conversation

@bitzic

@bitzic bitzic commented Jun 19, 2026

Copy link
Copy Markdown
Contributor
  • _B6_STANDARD_RUNGS -> one S3 rung, n_examples_per_task=100
  • matches the validated rho=0.614 run (run_b6_s3.sh patched the same)
  • drops 3x redundant eval: run_ladder_eval re-runs the same checkpoint per rung (dim/n_layers are advisory) and only the S3 cell feeds rho
  • caps per-task examples: mode-0 = ALL now spans ~20k-row tasks (bigbench_qa_wikidata) post the core22 fix -> would blow the time cap
  • aligns the canonical orchestrator before the 2026-06-29 pre-reg tag
  • refresh stale 3-rung docstrings + test fixture comment

- _B6_STANDARD_RUNGS -> one S3 rung, n_examples_per_task=100
- matches the validated rho=0.614 run (run_b6_s3.sh patched the same)
- drops 3x redundant eval: run_ladder_eval re-runs the same checkpoint
  per rung (dim/n_layers are advisory) and only the S3 cell feeds rho
- caps per-task examples: mode-0 = ALL now spans ~20k-row tasks
  (bigbench_qa_wikidata) post the core22 fix -> would blow the time cap
- aligns the canonical orchestrator before the 2026-06-29 pre-reg tag
- refresh stale 3-rung docstrings + test fixture comment
@bitzic
bitzic merged commit 2fe3f64 into main Jun 19, 2026
4 checks passed
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant