Resync llama_grammar with llama.cpp implementation and use curly braces quantities instead of repetitions #1721
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
…es quantifiers instead of repetitions
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Outdated suggestions cannot be applied.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 resync recent changes from llama.cpp's json_schema_to_grammar.py
Initially I came to this patch because I noticed that generated json strings sometimes contained ascii control characters, hence causing json.loads to fail.
Then I noticed that sometimes unterminated json was generated because llama.cpp was producing very long and wrong replies. I believe this is caused by a bug in grammar management, however using curly braces quantities instead of code-generated repetitions alleviates this problem a lot, and it gets totally fixed by providing max_length of string fields in the json schema.
Something else I noticed is that without this change the generated json could contain newlines and tabs/spaces as separators between json elements, whereas the grammar would impose a sigle whitespace: that should be another sign of a bug in grammar management. Perhaps teh grammar was silently ignored and a json was generated anyway because the json schema was part of the prompt as well.