For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Python class

GrammarMatcher

`GrammarMatcher`

class max.pipelines.context.GrammarMatcher(*args, **kwargs)

source

Bases: Protocol

Per-request grammar matcher stepped each decode step.

Backend-agnostic interface; method names mirror llguidance’s LLMatcher so a context can hold any backend’s matcher (llguidance, xgrammar) without branching. The llguidance LLMatcher satisfies this protocol natively.

`deep_copy()`

deep_copy()

source

Independent copy for speculative walks (never mutates the original).

Return type:: GrammarMatcher

`get_error()`

get_error()

source

Error message for the last rejection, if any (diagnostics).

Return type:: str | None

`get_grammar_warnings()`

get_grammar_warnings()

source

Grammar compilation warnings, if any (diagnostics).

Return type:: Any

`is_accepting()`

is_accepting()

source

Whether the matcher is at an accepting (stoppable) state.

Return type:: bool

`is_stopped()`

is_stopped()

source

Whether the matcher has reached a terminal state.

Return type:: bool

`try_consume_tokens()`

try_consume_tokens(tokens)

source

Advance the matcher; returns the number of tokens consumed.

Parameters:: tokens (list[int])
Return type:: int

GrammarMatcher​

deep_copy()​

get_error()​

get_grammar_warnings()​

is_accepting()​

is_stopped()​

try_consume_tokens()​