Fixes a couple of issues to add fp16 training support #476
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Moves the explicit `sigmoid` calculation inside the `dsl_assigner` so that `binary_cross_entropy_with_logits` can be used. This allows for the use of `auto_cast` to support training with `fp16` precision. If this is not done `torch` will complain that using `binary_cross_entropy` with `fp16` is unstable and as such refuses to train the model in `fp16` precision.
Allows for setting the model precision during training using the config system.
RangiLyu
approved these changes
Jan 20, 2023
RangiLyu
added a commit
that referenced
this pull request
Jan 20, 2023
* Fixes a couple of issues to add fp16 training support (#476) * Add half precision support to `nanodet_plus` head Moves the explicit `sigmoid` calculation inside the `dsl_assigner` so that `binary_cross_entropy_with_logits` can be used. This allows for the use of `auto_cast` to support training with `fp16` precision. If this is not done `torch` will complain that using `binary_cross_entropy` with `fp16` is unstable and as such refuses to train the model in `fp16` precision. * Add model precision settings to config Allows for setting the model precision during training using the config system. Co-authored-by: RangiLyu <[email protected]> * fix lint * lightning version Co-authored-by: Bjarne <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
There were a couple of issues when trying to use
fp16training. For one was that it was not exposed through the configuration system. The other was that theDynamicSoftLabelAssignerusedbinary_cross_entropyinstead ofbinary_cross_entropy_with_logits. This changes wheresigmoidis called on the predictions so that the more stablebinary_cross_entropy_with_logitscan be used and theTrainercan be configured to usefp16precision.