wassnameandClaudypoo 5dfc157f81 docs(c): prose signposts on the main() training loop ...

Full-sentence phase comments at the loop boundaries (the GRPO loop overview, the
per-prompt rollout/grade/accumulate phase). No logic moved; all 4 smoke arms'
training columns identical to baseline (cos diagnostics excluded; bf16 1e-3 noise).

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

2026-06-01 09:33:50 +00:00

..

__init__.py

setup

2026-05-23 10:40:02 +08:00

antipasto.py

refactor: route2 quarantine = scale-matched delta_S_hack, rip out 33M LoRA

2026-06-01 02:52:02 +00:00

bake_lora.py

top-k v_hack subspace + real-voice pairs + LoRA bake

2026-05-26 02:33:24 +00:00

build_substrate.py

fix: external-review criticals — os._exit oracle hole + exact even matching + honest teacher gt

2026-05-30 09:15:23 +00:00

data.py

refactor(c): extract data.py (load_problems) + vhack.py (load/postprocess v_hack)

2026-06-01 09:29:03 +00:00

derisk_loopholes.py

tooling: add ELICIT_HACK prompt tier + validate_spoonfeed updates

2026-05-31 00:00:40 +00:00

eval.py

refactor(c): extract eval.py (ref_logprobs, ablate_quarantine, eval_hack_solve)

2026-06-01 09:22:30 +00:00

extract_vhack_grad.py

refactor: route2 quarantine = scale-matched delta_S_hack, rip out 33M LoRA

2026-06-01 02:52:02 +00:00

figs.py

reorg: out/ sorted by datatype (vhack/ pools/ runs/ vhack_grads/ figs/)

2026-05-30 03:52:24 +00:00

pairs_from_pool.py

reorg: out/ sorted by datatype (vhack/ pools/ runs/ vhack_grads/ figs/)

2026-05-30 03:52:24 +00:00

pairs.py

wip

2026-05-29 06:29:46 +00:00

probe_distill.py

reorg: out/ sorted by datatype (vhack/ pools/ runs/ vhack_grads/ figs/)

2026-05-30 03:52:24 +00:00

probe_plot_stack.py

diag: cos_pre/post = ||relu(V@g)||/||g|| (hack-ward fraction) not signed sum

2026-05-31 05:03:54 +00:00

probe_traj.py

reorg: out/ sorted by datatype (vhack/ pools/ runs/ vhack_grads/ figs/)

2026-05-30 03:52:24 +00:00

proj.py

diag: cos_pre/post = ||relu(V@g)||/||g|| (hack-ward fraction) not signed sum

2026-05-31 05:03:54 +00:00

regrade_pool.py

fix: drop stale RH_HINT_REPLACE_TO import that crashed the v_hack refresh path

2026-05-30 23:32:28 +00:00

rewards.py

fix: align hack-token constants to the tokens the hints name

2026-05-30 13:36:14 +00:00

tablelog.py

refactor(c): extract tablelog.py (StepLogger, setup_logging, _Col)

2026-06-01 09:13:30 +00:00

train.py

docs(c): prose signposts on the main() training loop

2026-06-01 09:33:50 +00:00

verify_rewards.py

feat: object-attribute sentinel + exhaustive non-overlap matrix

2026-05-30 10:15:36 +00:00

verify_vhack_heldout.py

reorg: out/ sorted by datatype (vhack/ pools/ runs/ vhack_grads/ figs/)

2026-05-30 03:52:24 +00:00

vhack.py

refactor(c): extract data.py (load_problems) + vhack.py (load/postprocess v_hack)

2026-06-01 09:29:03 +00:00