llamamodel: fix BERT tokenization after llama.cpp update #2381
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
llama_tokenize now appends EOS and BOS consistently, so the logic that strips EOS needs to change. Signed-off-by: Jared Van Bortel <jared@nomic.ai>
Signed-off-by: Jared Van Bortel <jared@nomic.ai>
manyoso
approved these changes
May 28, 2024
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Outdated suggestions cannot be applied.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.
We have been stripping an extra token from the end of input to BERT models after the llama.cpp update in #2310. This was caught by the assertion, but only in debug builds:
I had changed llama_tokenize upstream to use the same logic as BOS/CLS for the EOS/SEP token appended by the BERT tokenizer, but this code was still relying on the old behavior where EOS was appended unconditionally.