PR #12363 Refine gradient accumulation (on device training)

a

pengwa committed 3 years ago

optimize inplace accumulator a bit

pengwa committed 3 years ago

fix inputs

pengwa committed 3 years ago

revert logging

pengwa committed 3 years ago

minor fix

pengwa committed 3 years ago

tune perf and resolve comments

pengwa committed 3 years ago

Merge branch 'master' of https://github.com/microsoft/onnxruntime into pengwa/inplace_acc

pengwa committed 3 years ago

typo

pengwa committed 3 years ago

fix

pengwa committed 3 years ago

fix tests

pengwa committed 3 years ago

move threshold to constexpr.

pengwa committed 3 years ago

onnxruntime Refine gradient accumulation (on device training) #12363 Merged