onnxruntime
Keep loss_scale and Whole Loss Subgraph in FP32 during Mixed Precision Training
#4268

Merged

Keep loss_scale and Whole Loss Subgraph in FP32 during Mixed Precision Training #4268

Lafi7e merged 9 commits into master from weicwang/lossscale

Keep loss subgraph as FP32 when mixed-p training.

24d22a27

Merge branch 'master' into weicwang/lossscale

9127101c

Fix case where there is no white-list loss op.

851220e5

Lafi7e added training

Lafi7e requested a review from

SherlockNoMad 5 years ago

Lafi7e requested a review from

pengwa 5 years ago

Lafi7e requested a review from

souptc 5 years ago

Lafi7e requested a review 5 years ago

weixingzhang commented on 2020-06-18

Get nodes from loss_scale instead of whitelist.

a21c36c5

Lafi7e requested a review from

xzhu1900 5 years ago

xzhu1900 commented on 2020-06-22

rename const variables.

132c488f

Merge branch 'master' into weicwang/lossscale

1f2d4623

xzhu1900 dismissed these changes on 2020-06-24

Merge branch 'master' into weicwang/lossscale

e40ab65e

merge from master

948480e5

Lafi7e dismissed their stale review via 948480e5 5 years ago

Merge branch 'master' into weicwang/lossscale

287a7a6f

zhijiang-xu approved these changes on 2020-07-02

xzhu1900 approved these changes on 2020-07-02

Lafi7e merged 28e4c0ed into master 5 years ago

Lafi7e deleted the weicwang/lossscale branch 5 years ago

Reviewers

xzhu1900

zhijiang-xu

weixingzhang

SherlockNoMad

pengwa

souptc

Assignees

No one assigned

Labels

training

Milestone

No milestone

onnxruntime Keep loss_scale and Whole Loss Subgraph in FP32 during Mixed Precision Training #4268 Merged

Keep loss_scale and Whole Loss Subgraph in FP32 during Mixed Precision Training #4268

onnxruntime
Keep loss_scale and Whole Loss Subgraph in FP32 during Mixed Precision Training
#4268

Merged