Bugfix/accuracy issue #52
Merged
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
- subsequent mask of espnet returns boolean value, which is not suitable for espnet_onnx
- if there is a given cache, then the first LayerNorm fails to calculate properly.
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Outdated suggestions cannot be applied.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.
This PR fixes some bugs that cause an accuracy drop.
fixed problems
Noneas the first cache for theforward_one_stepfunction in the decoder class. We can give bothNoneortorch.Tensorand the TransformerDecoder would change its computation depending on the input.However, in the onnx, we need to give one tensor as a dummy cache for
cache=Nonecase to avoid splitting the onnx model. This dummy cache caused an issue while computing the firstLayerNormin the second or later layers. So I removed the first sequence (dummy cache) if we need to compute with cache and the decoder layer is not the first layer.