Two changes to reduce the sequence length passed to genienlp #692
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
gcampax
added
bug
If an API returns a very long array, we need to trim it before passing it to the model, or we'll exceed the maximum sequence length.
s-jse
approved these changes
Jul 15, 2021
gcampax
force-pushed
the
wip/reduce-sequence-length
branch
from
1723c64
to
b1b2b41
Compare
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Outdated suggestions cannot be applied.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.
The recently approved movie skill returns the full list of actors to all movies it returns. This can be quite a long list, and it's too long to encode to BART, even if every actor is a single
GENERIC_ENTITY_org.themoviedb:actor_<n>token (which is split intoENTITY actor <n>before BART tokenization).To avoid that, we reduce the number of history items encoded in the context (which also helps the model by giving a stronger recency bias) and we cut the length of array values in the result.
This was caught in the staging environment which keeps going down (stanford-oval/genienlp#174)