sgl-project/sglang · critical · ValueError

--enable-strict-thinking requires a grammar backend that sup

Error message

--enable-strict-thinking requires a grammar backend that supports token filtering, but grammar_backend='none' was specified. Use --grammar-backend xgrammar or another backend that supports token filtering.

What it means

--grammar-backend none explicitly disables grammar/constrained decoding, but --enable-strict-thinking requires token filtering, which the 'none' backend cannot provide, so create_grammar_backend raises this configuration error at startup.

Source

Thrown at python/sglang/srt/constrained/base_grammar_backend.py:417

                f"Grammar backend disabled because tokenizer is not supported by XGrammar: {e}. "
                "Falling back to grammar_backend='none'. "
                "Structured outputs (JSON schema, regex, EBNF) will not be available."
            )
            get_context().override("grammar.import_fallback", grammar_backend="none")
            return None
    elif name == "llguidance":
        from sglang.srt.constrained.llguidance_backend import GuidanceBackend

        grammar_backend = GuidanceBackend(
            tokenizer=tokenizer,
            any_whitespace=not get_serving().constrained_json_disable_any_whitespace,
            whitespace_pattern=get_serving().constrained_json_whitespace_pattern,
            n_vocab=vocab_size,
            eos_token_ids=eos_token_ids,
        )
    elif name == "none":
        if get_serving().enable_strict_thinking:
            raise ValueError(
                "--enable-strict-thinking requires a grammar backend that supports "
                "token filtering, but grammar_backend='none' was specified. Use "
                "--grammar-backend xgrammar or another backend that supports token "
                "filtering."
            )
        return None
    else:
        raise ValueError(f"Invalid grammar backend: {name}")

    if get_serving().reasoning_parser and think_end_ids:
        from sglang.srt.constrained.reasoner_grammar_backend import (
            ReasonerGrammarBackend,
        )

        reasoning_parser = ReasoningParser(
            model_type=get_serving().reasoning_parser,
            stream_reasoning=False,
            tokenizer=tokenizer,

View on GitHub (pinned to 0132848349)

Solutions

1. Remove --enable-strict-thinking if you want grammar_backend='none'
2. Change to --grammar-backend xgrammar (or another token-filtering backend) to keep strict thinking
3. Audit scripts/launch configs for conflicting flags

Example fix

# before
--grammar-backend none --enable-strict-thinking
# after
--grammar-backend xgrammar --enable-strict-thinking

if server_args.grammar_backend == 'none':
    assert not server_args.enable_strict_thinking, 'strict thinking requires a token-filtering grammar backend'

When it happens

Trigger: Passing both --grammar-backend none and --enable-strict-thinking on the server command line or ServerArgs.

Common situations: Users disabling the grammar backend to save startup time or avoid xgrammar issues, while separately enabling strict thinking; contradictory flags copied from different runbooks.

Related errors

• --enable-strict-thinking requires a grammar backend with tok
• Invalid grammar backend: {name}
• world_size ({world_size}) is less than tensor_parallel_degre
• think_end_token '{reasoning_parser.detector.think_end_token}
• Strict reasoning format requested but the grammar backend do

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/ef58922c29b30d81. Report an issue: GitHub.