S2SGreedySearcher : Do not continue decoding when EOS token was generated for all samples from a batch#1899
Merged
Conversation
Collaborator
|
@Adel-Moumen what do you think? It could be added to the minor? |
Adel-Moumen
approved these changes
Mar 24, 2023
Adel-Moumen
left a comment
Collaborator
There was a problem hiding this comment.
LGTM! Thanks for spotting/solving the bug!
Collaborator
|
This PR will be shipped in the new release today, this was perfect timing ahah. |
Collaborator
|
Did you test that everything is running and providing the same performance
as before (at least in one case like librispeech)?
…On Fri, Mar 24, 2023, 1:21 PM Parcollet Titouan ***@***.***> wrote:
This PR will be shipped in the new release today, this was perfect timing
ahah.
—
Reply to this email directly, view it on GitHub
<#1899 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AEA2ZVUZKROXUMZP6B54Q5DW5XJY5ANCNFSM6AAAAAAWGZSIGQ>
.
You are receiving this because you are subscribed to this thread.Message
ID: ***@***.***>
|
Collaborator
|
Yes @mravanelli, I conducted the experiment using Whisper on LibriSpeech. I was previously aware of this bug and had attempted a somewhat similar approach that partially addressed the issue. The current fix fully resolves the problem by monitoring the batches that are reaching or have reached the eos. |
Collaborator
|
Perfect,
we should inherit this fix into the beamsearch refactor for the major
version.
…On Fri, Mar 24, 2023, 2:13 PM Adel Moumen ***@***.***> wrote:
Yes @mravanelli <https://github.com/mravanelli>, I conducted the
experiment using Whisper on LibriSpeech.
I was previously aware of this bug and had attempted a somewhat similar
approach that partially addressed the issue. The current fix fully resolves
the problem by monitoring the batches that are reaching or have reached the
eos.
—
Reply to this email directly, view it on GitHub
<#1899 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AEA2ZVQ4TJ36VJAVPUO3GVLW5XP4ZANCNFSM6AAAAAAWGZSIGQ>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
|
Collaborator
|
Yep. Just so everyone knows about this change: We went from 41 mins and 4 seconds with Whisper Tiny on LibriSpeech test-clean to 2 mins and 36 seconds. That's like a ~20x speed boost. Both of them led to the same WER of 7.50. |
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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 can speed up things when validating a seq2seq model