Closed mmoya01 closed 3 years ago
Hi @mmoya01
This error happens if torch.utils.checkpoint
is not imported. This is fixed on master now, see #9626
@patil-suraj thank you, that work
I still get the same error (when training DeBERTa-V3-base) on a colab GPU with Trainsformers==4.12
I using
model.gradient_checkpointing_enable() # to decrease memory usage
Before doing normal training via the HF trainer.
(It's fixed if I run this:)
from torch.utils.checkpoint import checkpoint
I get the same error when running training on DebertaForSequenceClassification
using the Trainer API with gradient_checkpointing
set to True.
@MoritzLaurer 's solution works for this also
thanks, it worked!
hello, I fine tuned my own LED model by following this notebook and I saved it using
however, whenever I try testing that model using something like this
I get the following error
I don't run into this error if I try loading the
patrickvonplaten/led-large-16384-pubmed
model. Not sure if I saved the model incorrectly, @patrickvonplaten or the rest of the community, I'd greatly appreciate any help with this