sgl-project/sglang · error · ValueError

{name} must not mix nested and flat entries

Error message

{name} must not mix nested and flat entries

What it means

When the first element of video_block_token_counts is a sequence (nested form), every subsequent element must also be a sequence; mixing ints and lists is ambiguous and rejected.

Source

Thrown at python/sglang/multimodal_gen/runtime/pipelines_core/stages/model_specific_stages/minimax_h3/presentation.py:187

def _as_nested_int_list(
    value: Sequence[int] | Sequence[Sequence[int]] | None,
    *,
    name: str,
) -> list[list[int]]:
    if value is None:
        return []
    if not isinstance(value, Sequence) or isinstance(value, (str, bytes)):
        raise ValueError(f"{name} must be a sequence")
    if len(value) == 0:
        return []
    first = value[0]
    if isinstance(first, Sequence) and not isinstance(first, (str, bytes)):
        out: list[list[int]] = []
        for group in value:
            if not isinstance(group, Sequence) or isinstance(group, (str, bytes)):
                raise ValueError(f"{name} must not mix nested and flat entries")
            out.append([int(item) for item in group])
        return out
    return [[int(item) for item in value]]


def _as_nested_float_list(
    value: Sequence[float] | Sequence[Sequence[float]] | None,
    *,
    name: str,
) -> list[list[float]]:
    if value is None:
        return []
    if not isinstance(value, Sequence) or isinstance(value, (str, bytes)):
        raise ValueError(f"{name} must be a sequence")
    if len(value) == 0:
        return []
    first = value[0]
    if isinstance(first, Sequence) and not isinstance(first, (str, bytes)):

View on GitHub (pinned to 0132848349)

Solutions

1. Normalize all entries to the same shape — nested lists for multiple videos ([[196],[128]]) or a flat list for a single video ([196,128])
2. Fix the data assembly to always append lists

Example fix

// before
video_block_token_counts=[[196,196], 128]
// after
video_block_token_counts=[[196,196], [128]]

assert all(isinstance(g, (list, tuple)) for g in video_block_token_counts), "uniform nesting required"

def uniform_nested(v: list) -> bool:
    return all(isinstance(g, (list, tuple)) for g in v) or all(not isinstance(g, (list, tuple)) for g in v)

When it happens

Trigger: Passing video_block_token_counts=[[196,196], 128] — first entry nested, second flat.

Common situations: Concatenating per-video lists with stray scalars; heterogeneous data built from mixed sources; appending a default scalar to a list of lists.

Related errors

• refiner cu_seqlens live text length must be in [1, {int(text
• {context}video block token counts and timestamps must align
• {context}video block token count must be positive
• {name} must be a sequence
• video reference requires block token counts and timestamps

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/5e42a71216061808. Report an issue: GitHub.