56 Commits

Author	SHA1	Message	Date
gaotingquan	8b218b01ac	refactor amp auto_cast context manager & loss scaler	2023-05-25 11:58:05 +08:00
gaotingquan	f37cb543b1	rm op black list in amp ... the op flatten_contiguous_range and greater_than has supported amp mode since paddle 2.4	2023-03-29 14:57:02 +08:00
Tingquan Gao	f2fc43baeb	Revert "refactor: mv all dataloaders to engine.dataloader_dict" ... This reverts commit 284e2a67564d4b1f5f4a9c04c7c1ab0e8d3ada75.	2023-03-14 16:47:13 +08:00
Tingquan Gao	a1e840e0da	Revert "refactor: iter_per_epoch -> max_iter" ... This reverts commit a38e42f644fcba8b60a9672762211b6f7054b290.	2023-03-14 16:47:13 +08:00
Tingquan Gao	915dde176a	Revert "refactor: rm train and eval from engine" ... This reverts commit 5a6fe171a7cfab842adc6a744f11a2e24deb5384.	2023-03-14 16:47:13 +08:00
Tingquan Gao	aa52682c55	Revert "rm amp code from train and eval & use decorator for amp training" ... This reverts commit d3941dc1e9628fa7cc83de7c3a6da3dfcd03b5de.	2023-03-14 16:47:13 +08:00
Tingquan Gao	85e200edb6	Revert "refactor" ... This reverts commit 32593b63751b922e17b59384ed64654e6fcef42d.	2023-03-14 16:47:13 +08:00
Tingquan Gao	03795249c1	Revert "revert for running" ... This reverts commit d3374e897e162053d93a20c21142135c3e7ee11c.	2023-03-14 16:47:13 +08:00
Tingquan Gao	7865207096	Revert "revert for running" ... This reverts commit 392b75b1acac742b74e808353059d0281df26dcc.	2023-03-14 16:47:13 +08:00
gaotingquan	392b75b1ac	revert for running	2023-03-10 16:56:55 +08:00
gaotingquan	d3374e897e	revert for running	2023-03-10 16:56:55 +08:00
gaotingquan	32593b6375	refactor	2023-03-10 16:56:55 +08:00
gaotingquan	d3941dc1e9	rm amp code from train and eval & use decorator for amp training	2023-03-10 16:56:55 +08:00
gaotingquan	5a6fe171a7	refactor: rm train and eval from engine	2023-03-10 16:56:55 +08:00
gaotingquan	a38e42f644	refactor: iter_per_epoch -> max_iter	2023-03-10 16:56:55 +08:00
gaotingquan	284e2a6756	refactor: mv all dataloaders to engine.dataloader_dict	2023-03-10 16:56:55 +08:00
HydrogenSulfate	7d9f4dcb59	change Tensor.numpy()[0] to float(Tensor) for 0-D tensor case	2022-11-01 14:37:11 +08:00
HydrogenSulfate	61b4153907	add batch Tensor collate to simplify dali code in train/eval/retrival code	2022-10-13 12:05:50 +08:00
cuicheng01	50fc7d0eae	fix bugs to adapt to the new framework	2022-09-20 10:06:20 +00:00
cuicheng01	cc46db1586	fix bugs to adapt to the new framework	2022-09-19 02:01:31 +00:00
HydrogenSulfate	266db4d89c	fix classification bug	2022-07-06 20:38:40 +08:00
gaotingquan	df3e75dde4	fix: warn when topk parameter setting is wrong	2022-06-06 16:47:57 +08:00
zhiboniu	edf1129e5d	match new eval function	2022-05-23 10:27:55 +00:00
zhiboniu	699c10aaeb	Merge remote-tracking branch 'ppcls/develop' into develop	2022-05-23 08:03:46 +00:00
zhiboniu	05ecf1d045	multi-card eval support	2022-05-18 04:54:44 +00:00
zhiboniu	50900443f3	remove strongbaseline_attr, etc...	2022-05-18 04:54:43 +00:00
zhiboniu	26d5b7d1cc	adapted dataset and loss	2022-05-18 04:54:43 +00:00
zhiboniu	0a3ecf60b4	add attribute strongbaseline	2022-05-18 04:54:43 +00:00
cuicheng01	45b1296c25	Add cls_demo_person code	2022-05-14 09:31:52 +00:00
Wei Shengyu	4bc4b7e0e2	Merge pull request #1876 from TingquanGao/dev/fix_amp_eval ... fix: amp eval	2022-05-05 20:39:02 +08:00
littletomatodonkey	bb13f3c4f5	fix single card dist (#1889) ... * fix single card logit * fix distillation yaml files	2022-05-05 09:48:56 +08:00
gaotingquan	275945dff9	fix: compatible with Paddle 2.2, 2.3, and develop.	2022-04-29 10:21:09 +00:00
gaotingquan	59a3dcfc1c	fix: amp eval	2022-04-26 12:26:17 +00:00
cuicheng01	4e6c36e269	Merge pull request #1833 from TingquanGao/dev/fix_dist_loss ... fix calc metric error and calc loss error in distributed.	2022-04-22 22:41:57 +08:00
gaotingquan	83ed5195c3	fix: set use_fp16_test to True when AMP O2 is enabled	2022-04-18 06:14:43 +00:00
gaotingquan	efde56ffc6	fix: only fp16 evaluation is supported when ampO2 is enabled	2022-04-13 12:14:14 +00:00
gaotingquan	474c918b27	fix: fix bug of batch_size statistics error	2022-04-13 09:19:30 +00:00
gaotingquan	c46189bad0	fix: fix bug about calc loss in dist	2022-04-12 06:56:44 +00:00
gaotingquan	b761325faa	fix: fp32 eval by default when enable amp ... If you want to eval by fp16 when enable amp, please set Amp.use_fp16_test=True, False by default.	2022-04-02 19:22:10 +08:00
WangChen0902	7595ba6d70	add AFD (#1683) ... * add AFD	2022-02-28 19:11:50 +08:00
gaotingquan	7040ce8314	refactor: change params to be consistent with amp	2022-01-25 11:58:07 +08:00
zhangbo9674	cd039a7b37	add save_dtype	2022-01-10 18:19:03 +08:00
zhangbo9674	d437bb0a7e	use fp32 to eval	2022-01-10 18:19:03 +08:00
zhangbo9674	bb19c1f7a6	fix eval bug	2022-01-10 18:19:03 +08:00
littletomatodonkey	aea712cc87	add dist of rec model (#1574) ... * add distillation loss func and rec distillation	2022-01-05 19:25:36 +08:00
gaotingquan	7732a69f1b	fix: fix key error in distillation	2021-12-16 18:21:08 +08:00
dongshuilong	f7ccc874e2	fix dali distributed eval bug	2021-11-16 11:09:21 +08:00
dongshuilong	278f6d8050	fig goooglenet distributed eval bug	2021-10-26 11:56:30 +00:00
dongshuilong	fd6f1ad2ca	fix clas distributed eval bug	2021-10-21 03:47:03 +00:00
dongshuilong	c93d638f4c	fix clas distributed eval bug	2021-10-20 11:22:37 +00:00