sgl-project/sglang · error · ValueError

Unknown match_type: '{match_type}'. Must be 'BFS' or 'PROB'.

Error message

Unknown match_type: '{match_type}'. Must be 'BFS' or 'PROB'.

What it means

The ngram corpus only accepts match_type of 'BFS' or 'PROB' (mapped via _MATCH_TYPE_MAP). Any other string raises immediately in __init__.

Source

Thrown at python/sglang/kernels/ops/speculative/ngram_corpus.py:58

    @tvm_ffi.register_object("sgl.NgramCorpus")
    class NgramCorpusFFI(tvm_ffi.Object):
        __slots__ = ("__dict__",)

        def __init__(
            self,
            capacity: int,
            max_trie_depth: int,
            min_bfs_breadth: int,
            max_bfs_breadth: int,
            draft_token_num: int,
            match_type: str,
            external_sam_budget: int = 0,
            external_corpus_max_tokens: int = 10000000,
        ) -> None:
            mt = _MATCH_TYPE_MAP.get(match_type)
            if mt is None:
                raise ValueError(
                    f"Unknown match_type: '{match_type}'. Must be 'BFS' or 'PROB'."
                )
            self.__ffi_init__(
                capacity,
                max_trie_depth,
                min_bfs_breadth,
                max_bfs_breadth,
                draft_token_num,
                mt,
                external_sam_budget,
                external_corpus_max_tokens,
            )
            self._draft_token_num = draft_token_num

        def insert(self, batch_tokens: List[List[int]]) -> None:
            tokens_flat, offsets = _to_csr(batch_tokens)
            self.async_insert(tokens_flat, offsets)  # type: ignore

View on GitHub (pinned to 0132848349)

Solutions

1. Use exactly 'BFS' or 'PROB' (case-sensitive).
2. Check _MATCH_TYPE_MAP keys in ngram_corpus.py for the accepted values.
3. Validate/normalize the spec config string before constructing.

Example fix

// before
corpus = NgramCorpus(..., match_type='bfs')
// after
corpus = NgramCorpus(..., match_type='BFS')

assert match_type in ('BFS', 'PROB'), match_type

def is_valid_match_type(s: str) -> bool:
    return s in ('BFS', 'PROB')

When it happens

Trigger: Passing match_type='bfs' (lowercase), 'prob_sample', or a typo like 'PRO ' to the ngram corpus constructor.

Common situations: Config files with lowercase or renamed speculation settings (e.g. porting from vLLM-style 'greedy'/'sample' ngram params to this API).

Understand the failure class

Background: Invalid enum value errors: "Unknown type", "Invalid scope", "must be one of" — when a string is not on the library's allowed list — this error's family across 23 libraries.

Related errors

• This browser cannot encode H.264 MP4
• --speculative-draft-window-size must be positive, got {}.
• DFLASH requires --speculative-dflash-block-size to be positi
• --speculative-draft-window-size must be >= --speculative-num
• DSpark requires --speculative-dspark-block-size to be positi

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/a8ae5c892d2876f6. Report an issue: GitHub.