HPDL-Group / Merak

Apache License 2.0
69 stars 9 forks source link

[fp16 error] In GPT-2 run example #2

Closed lin88lin8850 closed 1 year ago

lin88lin8850 commented 1 year ago

when I set fp16=true during training gpt-2, error happens:

image
TXacs commented 1 year ago

This error come from the gpt2 model of transformers. Because the reorder_and_upcast_attn is True in config.

lin88lin8850 commented 1 year ago

thanks a lot! bug solved, the issue will be closed