mindspore2022

Author	SHA1	Message	Date
He Wei	7d9a783993	[auto-monad] Support side-effects by auto-monad The basic idea is: exploits data dependency to control the execution order of side-effect operations, and keep the semantics of ANF unchanged. The ControlDepend primitive is removed and there are two primitives added: 1. UpdateState: ``` a = Assign(para, value) ``` became: ``` a = Assign(para, value, u) u = UpdateState(u, a) ``` 2. Load: ``` x = Add(para, value) ``` became: ``` p = Load(para, u) x = Add(p, value) u = UpdateState(u, p) ```	2021-02-08 09:01:15 +08:00
yao_yf	65d8e63580	set last node data parallel or repeat calculate in eval/predict	2020-10-26 20:05:00 +08:00
Yi Huaijie	a836d25c64	change API set_strategy() to shard()	2020-09-10 15:30:19 +08:00
panyifeng	1a54785fe2	remove name arg from gradoperation	2020-08-26 11:16:57 +08:00
panyifeng	637e812347	remove global grad ops	2020-08-24 16:50:09 +08:00
yao_yf	e4de26d5bc	embeddinglookup wrap	2020-08-12 14:24:13 +08:00
lirongzhen1	51796aa624	fix sparse feature bug for auto parallel	2020-07-27 10:44:49 +08:00
lirongzhen1	5d63c60135	add sparse feature test cases for auto parallel	2020-07-22 14:27:12 +08:00
wangnan39@huawei.com	86889c59cb	optimizer adapt IndexedSlices	2020-07-17 15:57:50 +08:00
lirongzhen1	516b56cb64	sparse feature bp	2020-06-11 16:32:24 +08:00