v0.4.10

• All UTF-8 codepoints can now be used in json schema properties.

v0.4.9

• Use newer kbnf version to make Formatter's copy and deepcopy actually work

v0.4.8

• Custom vocabulary processors by @lukaszkolodziejczyk
• Properties keyword is now optional for type: object json schema

v0.4.7

• Support substring_of as a constraint metadata in typing.Annotated and substringOf as a custom constraint in json schema on the generated string.
• Support FormatterBuilder.regex_complement to allow constraining the output to be a string that does not match the string anywhere.

v0.4.6

• Support more non-type level constraints(see #12)
• Update docs to reflect newest features and potential issues
• Bug fix for FormatterBuilder.str()
• Make accept bytes behaviour consistent with accept token

v0.4.5

• Fix quality degradation when duplicate tokens(tokens that correspond to the same bytes but different token ids) are present in vocabulary.

v0.4.4

Bug fix: typing.Literal["xxx"] will now actually work correctly with JsonExtractor

v0.4.3

• Bug fix: typing.Literal["xxx"] will now work correctly with JsonExtractor
• Top level array is now supported in json schema with JsonExtractor

v0.4.2

• logits processors now support is_completed()
• Formatters sequence can now be sparce; in other words, it is possible now to constrain some sequences in one batch but leave others unconstrained.

v0.4.1

• Significantly improved vllm integrations' speed
• Update kbnf dependency to v0.3.10