Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
In AMP fp16 mode, overflow issues were observed in several places in the network. To address this, those regions are now converted back to fp32. This change does not introduce significant performance cost which offsets the benefits of mixed precision training. We also disallow the use of memory efficient kernel (use_memory_efficient_kernel=false) in AMP fp16 mode explicitly.
It worth mentioning the debugging tool developed at huggingface for detecting overflow issue: https://huggingface.co/docs/transformers/v4.14.1/en/debugging