evil_MoE/docs/writeup/data/teacher_ablation.csv at 329066e99bc1621597fcc76e8c0b8293661c6f39

mirror of https://github.com/wassname/evil_MoE.git synced 2026-06-27 19:31:11 +08:00

Files

T

wassname 329066e99b paper: teacher-off control appendix (app:teacher) -- teacher seeds not sustains

Vanilla deploy-hack keeps climbing after teacher cut at step 40 (0.36->0.58,
job 87), at/above teacher-on (job 97). Closest-match jobs differ in LR; FIXME
to swap in lr-matched job 124 (queued low-prio). CSV is the committed data
artifact; fig regen by plot_teacher_ablation.py.

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

2026-06-05 12:30:49 +00:00

860 B

Raw Blame History

1	step	arm	teacher_schedule	lr	deploy_hack	deploy_solve	job
2	0	vanilla	off@40	3e-3	0.000	0.359	87
3	20	vanilla	off@40	3e-3	0.141	0.438	87
4	40	vanilla	off@40	3e-3	0.359	0.359	87
5	60	vanilla	off@40	3e-3	0.438	0.562	87
6	80	vanilla	off@40	3e-3	0.453	0.531	87
7	100	vanilla	off@40	3e-3	0.469	0.531	87
8	120	vanilla	off@40	3e-3	0.500	0.500	87
9	140	vanilla	off@40	3e-3	0.516	0.422	87
10	160	vanilla	off@40	3e-3	0.578	0.359	87
11	180	vanilla	off@40	3e-3	0.469	0.469	87
12	199	vanilla	off@40	3e-3	0.484	0.453	87
13	0	vanilla	on	1e-3	0.000	0.328	97
14	20	vanilla	on	1e-3	0.000	0.484	97
15	40	vanilla	on	1e-3	0.172	0.500	97
16	60	vanilla	on	1e-3	0.250	0.547	97
17	80	vanilla	on	1e-3	0.219	0.500	97
18	100	vanilla	on	1e-3	0.281	0.469	97
19	120	vanilla	on	1e-3	0.328	0.406	97
20	140	vanilla	on	1e-3	0.281	0.453	97
21	160	vanilla	on	1e-3	0.328	0.438	97
22	180	vanilla	on	1e-3	0.391	0.500	97
23	199	vanilla	on	1e-3	0.344	0.500	97

860 B Raw Blame History

860 B

Raw Blame History