Skip to content

add momentum optimizer and test FP32 AMP PureFP16#4

Open
pangyoki wants to merge 1 commit intowawltor:masterfrom
pangyoki:add_pure_fp16_for_bert_finetune
Open

add momentum optimizer and test FP32 AMP PureFP16#4
pangyoki wants to merge 1 commit intowawltor:masterfrom
pangyoki:add_pure_fp16_for_bert_finetune

Conversation

@pangyoki
Copy link

@pangyoki pangyoki commented Nov 26, 2020

In Bert SST-2 finetune task:

  • change AdamW optimizer to Momentum optimizer ( for using multi_precision in Pure Fp16 training)

  • use run_glue_fp32.sh, run_glue_amp.sh, run_glue_pure_fp16.sh to run FP32, AMP or Pure FP16 version.

  • you may need to use cast if it occurs LayerNorm or matmul dtype ERROR

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant