wassname 154a37441b refactor: OneCycleLR replaces SequentialLR(LinearLR, CosineAnnealingLR) ...

One scheduler object does warmup + cosine relaxation; pct_start=warmup_frac
is the explicit warmup. cycle_momentum=False so it doesn't clobber the
configured AdamW betas (adam_beta1). Curve (100 steps, 20% warmup, 3e-4):
peaks 3e-4 at step ~19 via smooth cos ramp (vs old linear), 1.7e-4 at step
10 where 5e-4 had diverged. Smoke + all verify gates green.

Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>

2026-06-11 01:52:30 +00:00

..

__init__.py

rename python package projected_grpo -> vgrout

2026-06-05 14:51:48 +08:00

data.py

fix: rotate the unhackable (gt_only) subset per step, not frozen per pid

2026-06-10 06:14:08 +00:00

derisk_loopholes.py

cleanup: consolidate stale loaders and pair scripts

2026-06-09 12:47:32 +00:00

eval.py

refactor: collapse to lora2r-only (none/routeV/absorb); delete erase/antipasto/lora_frozen_b paths

2026-06-10 10:58:22 +00:00

extract_vhack_grad.py

Consolidate tagged hack pairsets in data

2026-06-10 11:58:53 +00:00

figs.py

cleanup: delete dead delta_S machinery (PiSSA->lora2r leftovers)

2026-06-10 11:45:54 +00:00

lora2r.py

cleanup(lora2r): resolve user TODOs -- F.linear alias + jaxtyping hook shapes

2026-06-10 11:51:49 +00:00

pairs_from_pool.py

Consolidate tagged hack pairsets in data

2026-06-10 11:58:53 +00:00

pairs.py

Consolidate tagged hack pairsets in data

2026-06-10 11:58:53 +00:00

proj.py

cleanup: delete dead delta_S machinery (PiSSA->lora2r leftovers)

2026-06-10 11:45:54 +00:00

regrade_pool.py

cleanup: consolidate stale loaders and pair scripts

2026-06-09 12:47:32 +00:00

rewards.py

refactor: extract train_config.py + run_artifacts.py from train.py; slim results scripts

2026-06-09 13:34:50 +00:00

run_artifacts.py

refactor: collapse to lora2r-only (none/routeV/absorb); delete erase/antipasto/lora_frozen_b paths

2026-06-10 10:58:22 +00:00

tablelog.py

feat: online-stats gate + step-level teacher forcing + AUROC diagnostic

2026-06-10 14:22:37 +00:00

train_config.py

fix: lr 3e-4 + 20% warmup (5e-4 diverged at warmup-end); slim bf16 ckpts

2026-06-11 01:42:20 +00:00

train.py

refactor: OneCycleLR replaces SequentialLR(LinearLR, CosineAnnealingLR)

2026-06-11 01:52:30 +00:00