mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-09-19 14:04:19 +00:00
Code Issues Packages Projects Releases Wiki Activity
730 Commits 5 Branches 3 Tags
Commit Graph

730 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Daniel Povey
245d39b1bb Still debugging but close to done 2022-07-11 00:33:37 -07:00
Daniel Povey
27da50a1f6 Committing partial work.. 2022-07-10 15:46:32 -07:00
Daniel Povey
d25df4af5e Slight refactoring, preparing for batching. 2022-07-09 22:24:36 -07:00
Daniel Povey
d9a6180ae0 Bug fix 2022-07-10 10:20:39 +08:00
Daniel Povey
b7035844a2 Introduce scalar_max, stop eps getting large or small 2022-07-10 10:13:55 +08:00
Daniel Povey
2f73434541 Reduce debug frequency 2022-07-10 06:44:50 +08:00
Daniel Povey
b3bb2dac6f Iterative, more principled way of estimating param_cov 2022-07-10 06:28:01 +08:00
Daniel Povey
d139c18f22 Max eig of Q limited to 5 times the mean 2022-07-09 14:30:03 +08:00
Daniel Povey
ffeef4ede4 Remove rank-1 dims, meaning where size==numel(), from processing. 2022-07-09 13:36:48 +08:00
Daniel Povey
2fc9eb9789 Respect param_pow 2022-07-09 12:49:04 +08:00
Daniel Povey
209acaf6e4 Increase lr_update_period to 200. The update takes about 2 minutes, fore entire model. 2022-07-09 11:36:54 +08:00
Daniel Povey
61cab3ab65 introduce grad_cov_period 2022-07-09 10:29:23 +08:00
Daniel Povey
35a51bc153 Reduce debug probs 2022-07-09 10:22:19 +08:00
Daniel Povey
65bc964854 Fix bug for scalar update 2022-07-09 10:14:20 +08:00
Daniel Povey
aa2237a793 Bug fix 2022-07-09 10:11:54 +08:00
Daniel Povey
50ee414486 Fix train.py for new optimizer 2022-07-09 10:09:53 +08:00
Daniel Povey
6810849058 Implement new version of learning method. Does more complete diagonalization of grads than the previous methods. 2022-07-09 10:02:17 +08:00
Daniel Povey
a9edecd32c Conformed that symmetrizing helps because of interaction with regular update; still meta_lr_scale=0 best :-( 2022-07-09 05:20:04 +08:00
Daniel Povey
52bfb2b018 This works better for reasons I dont understand. transpose is enough, same as symmetrizing. 2022-07-08 11:53:59 +08:00
Daniel Povey
e9ab1ddd39 Inconseqeuential config change 2022-07-08 11:03:16 +08:00
Daniel Povey
be6680e3ba Couple configuration changes, comment simplification 2022-07-08 09:46:42 +08:00
Daniel Povey
75e872ea57 Fix bug in getting denom in proj update 2022-07-08 09:13:54 +08:00
Daniel Povey
914ac1e621 Works better with meta_lr_scale=0, must be bug. 2022-07-08 09:07:06 +08:00
Daniel Povey
923468b8af Deal with SVD failure better. 2022-07-08 09:00:12 +08:00
Daniel Povey
97feb8a3ec Reduce meta_lr_scale, reduces loss @140 from 1.4 to 0.39 2022-07-08 06:33:07 +08:00
Daniel Povey
b6199a71e9 Introduce delta_scale to slow down changes on M; significantly better. 2022-07-08 06:05:31 +08:00
Daniel Povey
ceb9815f2b Increase lr_est_period 2022-07-08 05:51:18 +08:00
Daniel Povey
fb36712e6b Another bug fix, regarding Q being transposed. 2022-07-08 05:22:24 +08:00
Daniel Povey
ad2e698fc3 Cleanups 2022-07-08 04:44:21 +08:00
Daniel Povey
04d2e10b4f Version that runs 2022-07-08 04:37:46 +08:00
Daniel Povey
e6d00ee3e4 More drafts of new method, not tested. 2022-07-06 23:05:06 -07:00
Daniel Povey
26815d177f Draft of the new method.. 2022-07-06 22:59:36 -07:00
Daniel Povey
e9e2a85c95 In the middle of reworking for new idea 2022-07-06 13:35:19 -07:00
Daniel Povey
41368f6b63 Change comment 2022-07-05 17:11:45 -07:00
Daniel Povey
2692d5f903 Closer to finished 2022-06-30 23:54:26 -07:00
Daniel Povey
d64cb1cb48 draft, not working, will edit locally 2022-06-30 15:35:26 +08:00
Daniel Povey
0b811546f3 Apply reverse_cutoff with param_pow, to make it not too strong. 2022-06-26 11:07:08 +08:00
Daniel Povey
0aa5a334d6 Fix regarding reverse_cutoff formula 2022-06-25 18:24:05 +08:00
Daniel Povey
8a0277d493 Increase param_reverse_cutoff from 4 to 16 2022-06-25 18:02:00 +08:00
Daniel Povey
2ccc7ccbeb Make the application of param_reverse_cutoff conditional, not applied when rank is a problem 2022-06-25 17:56:04 +08:00
Daniel Povey
146d7c5a93 Bug fix 2022-06-24 19:49:13 +08:00
Daniel Povey
117d348f70 This time get it right, last time was starting from wrong base. 2022-06-24 19:00:58 +08:00
Daniel Povey
6002e19836 Change max_fullcov_size back to 1023.. doing it this way because I forgot to commit the changes on this branch 2022-06-24 18:59:34 +08:00
Daniel Povey
c2ec901e73 Change param_pow from 0.75 to 0.5 2022-06-24 17:14:41 +08:00
Daniel Povey
f327407308 Avoid error if svd fails 2022-06-24 13:20:20 +08:00
Daniel Povey
2233c852fd Increase max_fullcov_size from 1023 to 2048 2022-06-24 13:02:17 +08:00
Daniel Povey
683b8e1504 Some code reworking and fixes, rationalizing how speedup is done and fix an issue affecting learning rate. 2022-06-23 15:40:51 +08:00
Daniel Povey
c34344e98f Change param_pow,grad_pow to 0.75,0.95 2022-06-23 14:04:11 +08:00
Daniel Povey
3767f3d616 Reducd param_pow from 0.8 to 0.7 2022-06-23 11:12:12 +08:00
Daniel Povey
ca17f23da3 Reduce powers to 0.8,0.85 2022-06-23 11:09:27 +08:00