Improve summary chunking #397
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
Signed-off-by: Michael Clifford <mcliffor@redhat.com>
MichaelClifford
requested review from
rhatdan,
sallyom,
lmilbaum,
cgwalters and
Gregory-Pereira
as code owners
sallyom
pushed a commit
to sallyom/locallm
that referenced
this pull request
May 1, 2024
Signed-off-by: Michael Clifford <mcliffor@redhat.com>
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Outdated suggestions cannot be applied.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.
Document chunking in
summarizer.pywas still failing with some models on some texts. (There were some model vocabularies that caused the model server to fail to return the detokenized data if specific token sequences were split into separate chunks by the chunking process).To address this issue, we've implemented a new approach to chunking. Instead of chunking the raw tokens, we will chunk the text, like in our earlier implementation, and rely on
extras/tokenize/countto confirm that the text chunk is within the token limit, if not, we split the text chunk in two.I've manually evaluated this with the problematic and non-problemtatic models. It appears to work as well as before, and avoids any of the detokenization crashes we were experiencing.