188 Commits

Author	SHA1	Message	Date
Daniel Povey	7c46c3b0d4	Remove dropout in output layer	2022-03-30 11:20:04 +08:00
Daniel Povey	21a099b110	Fix padding bug	2022-03-30 11:18:04 +08:00
Daniel Povey	ca6337b78a	Add another convolutional layer	2022-03-30 11:12:35 +08:00
Daniel Povey	1b8d7defd0	Reduce 1st conv channels from 64 to 32	2022-03-30 00:44:18 +08:00
Daniel Povey	4e453a4bf9	Rework conformer, remove some code.	2022-03-29 23:41:13 +08:00
Daniel Povey	11124b03ea	Refactoring and simplifying conformer and frontend	2022-03-29 20:32:14 +08:00
Daniel Povey	2cde99509f	Change max-keep-prob to 0.95	2022-03-27 23:21:42 +08:00
Daniel Povey	262388134d	Increase model_warm_step to 4k	2022-03-27 11:18:16 +08:00
Daniel Povey	8a8134b9e5	Change power of lr-schedule from -0.5 to -0.333	2022-03-27 00:31:08 +08:00
Daniel Povey	953aecf5e3	Reduce layer-drop prob after warmup to 1 in 100	2022-03-27 00:25:32 +08:00
Daniel Povey	b43468bb67	Reduce layer-drop prob	2022-03-26 19:36:33 +08:00
Daniel Povey	0e694739f2	Fix test mode with random layer dropout	2022-03-25 23:28:52 +08:00
Daniel Povey	d2ed3dfc90	Fix bug	2022-03-25 20:35:11 +08:00
Daniel Povey	4b650e9f01	Make warmup work by scaling layer contributions; leave residual layer-drop	2022-03-25 20:34:33 +08:00
Daniel Povey	1f548548d2	Simplify the warmup code; max_abs 10->6	2022-03-24 15:06:11 +08:00
Daniel Povey	aab72bc2a5	Add changes from master to decode.py, train.py	2022-03-24 13:10:54 +08:00
Daniel Povey	9a8aa1f54a	Change how warmup works.	2022-03-22 15:36:20 +08:00
Daniel Povey	cef6348703	Change max-abs from 6 to 10	2022-03-22 13:50:54 +08:00
Daniel Povey	4004ca81d8	Increase warm_step (and valid_interval)	2022-03-22 13:32:24 +08:00
Daniel Povey	b82a505dfc	Reduce initial pruned_loss scale from 0.01 to 0.0	2022-03-22 12:30:48 +08:00
Daniel Povey	11a04c50ae	Change 0.025,0.05 to 0.01 in initializations	2022-03-21 21:29:24 +08:00
Daniel Povey	05e30d0c46	Add max-abs=6, debugged version	2022-03-21 21:15:00 +08:00
Daniel Povey	ccbf8ba086	Incorporate changes from master into pruned_transducer_stateless2.	2022-03-21 21:12:43 +08:00
Daniel Povey	05b5e78d8f	Add norm+balancer to VggSubsampling	2022-03-21 15:55:11 +08:00
Daniel Povey	0ee2404ff0	Remove logging code that broke with newer Lhotse; fix bug with pruned_loss	2022-03-19 14:01:45 +08:00
Daniel Povey	8cff994cd7	Set also scale for embedding to 0.025.	2022-03-18 21:30:05 +08:00
Daniel Povey	188eada7ac	Change initial std from 0.05 to 0.025.	2022-03-18 21:28:34 +08:00
Daniel Povey	c9f1aeb7d1	Fix bug with import	2022-03-18 16:40:24 +08:00
Daniel Povey	2dfcd8f117	Double warm_step	2022-03-18 16:38:36 +08:00
Daniel Povey	ba3611cefd	Cosmetic changes to swish	2022-03-18 16:35:48 +08:00
Daniel Povey	6769087d70	Remove scale_speed, make swish deriv more efficient.	2022-03-18 16:31:25 +08:00
Daniel Povey	cbe6b175d1	Reduce warmup scale on pruned loss form 0.1 to 0.01.	2022-03-17 16:46:59 +08:00
Daniel Povey	acc0eda5b0	Scale down pruned loss in warmup mode	2022-03-17 16:09:35 +08:00
Daniel Povey	13db33ffa2	Fix diagnostics-getting code	2022-03-17 15:53:53 +08:00
Daniel Povey	11bea4513e	Add remaining files in pruned_transducer_stateless2	2022-03-17 11:17:52 +08:00
Daniel Povey	e3ad8f63e7	update decode.py file type	2022-03-16 22:22:10 +08:00
Daniel Povey	cc8e4412f7	Add more files..	2022-03-16 22:16:40 +08:00
Daniel Povey	1f3a15f3c4	Start adding some files..	2022-03-16 22:14:30 +08:00