PR #3639 Add Distributed Checkpointing support

Change naming of moments to Moment_x_<weight_name>

ashbhandare committed 5 years ago

Modify zero test to hit bert like scenario.

ashbhandare committed 5 years ago

Revert "Modify zero test to hit bert like scenario."

ashbhandare committed 5 years ago

Add checkpointing code and zero checkpoint aggregation

ashbhandare committed 5 years ago

Correct aggregation for LAMB, cleanup

ashbhandare committed 5 years ago

Add simple checkpointing test

ashbhandare committed 5 years ago

Add test for zero checkpoint aggregation

ashbhandare committed 5 years ago

Fix tests

ashbhandare committed 5 years ago

fix test

ashbhandare committed 5 years ago

Review changes

ashbhandare committed 5 years ago

Fix test after review comment fix

ashbhandare committed 5 years ago

Fix API, test

ashbhandare committed 5 years ago

Fix test after API change

ashbhandare committed 5 years ago

Decouple save load from ORTTrainer

ashbhandare committed 5 years ago

Add flag to not break checkpointing with ORTModel'

ashbhandare committed 5 years ago

onnxruntime Add Distributed Checkpointing support #3639 Merged