Xiaoda Zhang
bb5d4212f7
enable All2All in infering redistribution ops
4 years ago
lichenever
3c7cfb7c08
auto_parallel_support_control_flow
4 years ago
i-robot
a7d40fc220
!20520 [AutoParallel]Add op AllToAllv
Merge pull request !20520 from lichen/add_op_AllToAllv
4 years ago
lichenever
8c1998fd6b
add_op_AllToAllv
4 years ago
i-robot
c9d3c1d346
!20411 enable optimizer parallel for inference
Merge pull request !20411 from gziyan/enable_opt_shard_predict
4 years ago
yangzhenzhang
b31cd27a08
update check strategy for conv2d
4 years ago
Ziyan
1c9166e0a6
remove restriction for opt shard in inference
4 years ago
Xiaoda Zhang
04381273b3
Add the sharding propagation function:
1) users configure sharding strategies for operators;
2) framework will propagate the strategies from configured-ops to
non-configured ops using BFS;
3) the propagation goal is to minimize redistribution communication
cost;
4 years ago
chenhaozhe
086a871975
Change Loss to LossBase
4 years ago
lichenever
db8850a4a3
pipeline_support_predict_master
4 years ago
Ziyan
be1f5a43d7
opt shard fit micro batch
4 years ago
yangzhenzhang
69acf757d0
add parallel op for conv2d backprop input
4 years ago
yangzhenzhang
24370b5613
add parallel op for maxpool
4 years ago
yangzhenzhang
af0d28de48
add parallel op for batchnorm
4 years ago
i-robot
85d860e6a2
!16457 [AutoParallel]pipeline_split_adapt_master
Merge pull request !16457 from lichen/pipeline_split_adapt_master
4 years ago
lichenever
db5d508356
pipeline_split_adapt_master
4 years ago
yangzhenzhang
7a40741048
add parallel operator for conv2d
4 years ago
Ziyan
95ac0f6d58
fix optimizer weight shard config
4 years ago
chenhaozhe
9da8534396
change _Loss to Loss
4 years ago
mindspore-ci-bot
1c8fda25ef
!16478 handle load op in step parallel
From: @gong_zi_yan
Reviewed-by: @yangzhenzhang,@stsuteng
Signed-off-by: @stsuteng
4 years ago
mindspore-ci-bot
b45b63fc58
!17239 add parallel gathernd test case
From: @hanyang001
Reviewed-by: @stsuteng,@yangzhenzhang
Signed-off-by: @stsuteng
4 years ago
Wan Hanyang
c51dff2634
add parallel gathernd test case
4 years ago
Wan Hanyang
3ce521d78f
add parallel layernorm test case
4 years ago
Ziyan
4b17493e52
handle load in step parallel
4 years ago
yangzhenzhang
d711d98f07
clean duplicate code
4 years ago
yao_yf
732d13ccff
parallel dropout support repeated compute
4 years ago
yangzhenzhang
6aa3859131
modify check strategy for scatter update
4 years ago
Ziyan
2a752f24bf
enable not fully use opt shard
5 years ago
yao_yf
e967f1939b
parallel envs variable check
4 years ago
mindspore-ci-bot
78fcdbc7c9
!15790 modify scatter update op
From: @yangzhenzhang
Reviewed-by: @kisnwang,@stsuteng
Signed-off-by: @stsuteng
4 years ago
yangzhenzhang
075f680a42
modify scatter update op
4 years ago
Xiaoda Zhang
aa52399200
Making the Tile operator to have more parallel strategies
5 years ago
yao_yf
093ef784de
dont insert virtualoutput for scalar
4 years ago
mindspore-ci-bot
3cfd58e8e0
!15643 insert virtual div only for first input of dropout do mask
From: @yangzhenzhang
Reviewed-by: @stsuteng,@kisnwang
Signed-off-by: @stsuteng
4 years ago
mindspore-ci-bot
49d6c029a6
!15542 split axis and batch for gather
From: @yangzhenzhang
Reviewed-by: @kisnwang,@stsuteng,@stsuteng
Signed-off-by: @stsuteng,@stsuteng
4 years ago
yangzhenzhang
5828973978
fix bug for dropout do mask
4 years ago
yao_yf
21276408b8
parallel virtual_out_ops
5 years ago
yangzhenzhang
213922574e
split axis and batch for gatherv2
4 years ago
yangzhenzhang
c2ca2232c5
add select op
5 years ago
mindspore-ci-bot
1c9d3c0aa0
!15353 add parallel operator for scatter update
From: @yangzhenzhang
Reviewed-by: @kisnwang,@stsuteng
Signed-off-by: @stsuteng
4 years ago
mindspore-ci-bot
0fd1726e79
!15172 Clean GraphKernel's codes from frontend
From: @dayschan
Reviewed-by: @gaoxiong1,@dylangeng,@gaoxiong1
Signed-off-by: @dylangeng,@dylangeng
4 years ago
yangzhenzhang
9cdd70433f
add scatterupdate op
5 years ago
yangzhenzhang
d070af122f
add topk op
5 years ago
dayschan
771e3f61f3
Clean GraphKernel's codes from frontend
1. set class GraphKernel as deprecated, and treat it as Cell
2. set class InplaceAssign as deprecated, suggested using Assign instead.
3. set op_selector as deprecated, removed the _selected_ops and _selected_grad_ops, replaced with real operations
4. removed the two passes of GraphKernel from frontend
5. removed the GraphKernel's codes from other modules
5 years ago
yangzhenzhang
f9f5df368e
add gathernd op
4 years ago
yangzhenzhang
bcd2ecc403
check layouts for shared parameter
4 years ago
yao_yf
a83fb3316b
fix parallel timeout
5 years ago
yao_yf
4d0635eabe
set parallel communication init flag in parallel ut
5 years ago
dingpeifei
87e41aaeee
IR operators of GPU and CPU are unified as batchnorm
5 years ago
mindspore-ci-bot
7454ac8ecd
!13382 [PipelineSplit]change pipeline key word
From: @lichen666
Reviewed-by: @kisnwang,@zhunaipan
Signed-off-by: @zhunaipan
5 years ago