What causes "{name} values must be positive, got {list(value)!r}" in sgl-project/sglang?

Fix the upstream producer so latent/patch dims are >= 1 (e.g. skip empty video segments).

sgl-project/sglang · error · ValueError

{name} values must be positive, got {list(value)!r}

Error message

{name} values must be positive, got {list(value)!r}

What it means

After checking length, _int_tuple converts each element with int() and requires every value to be strictly positive. This fires when a shape/patch tuple contains a zero or negative entry, which would make patch grid math (divisions and row counts) invalid.

Source

Thrown at python/sglang/multimodal_gen/runtime/pipelines_core/stages/model_specific_stages/minimax_h3/packed_tokens.py:14

# SPDX-License-Identifier: Apache-2.0
from __future__ import annotations

from collections.abc import Sequence

import torch


def _int_tuple(value: Sequence[int], name: str, length: int) -> tuple[int, ...]:
    if len(value) != length:
        raise ValueError(f"{name} must have length {length}, got {list(value)!r}")
    out = tuple(int(item) for item in value)
    if any(item <= 0 for item in out):
        raise ValueError(f"{name} values must be positive, got {list(value)!r}")
    return out


def _rank(tensor: torch.Tensor, name: str, rank: int) -> None:
    if tensor.ndim != rank:
        raise ValueError(f"{name} must be rank {rank}, got shape={list(tensor.shape)}")


def minimax_h3_patchify_video_latent(
    latent: torch.Tensor,
    *,
    patch_size: Sequence[int],
) -> torch.Tensor:
    """Pack SGLang video latent [B,C,T,H,W] into DiT token rows."""

    _rank(latent, "video latent", 5)
    pt, ph, pw = _int_tuple(patch_size, "patch_size", 3)
    batch, channel, full_t, full_h, full_w = (int(dim) for dim in latent.shape)

View on GitHub (pinned to 0132848349)

Solutions

Fix the upstream producer so latent/patch dims are >= 1 (e.g. skip empty video segments).
Guard against frame_count == 0 before computing frame_count - 1 style values.
Validate config-loaded patch sizes at startup and fail with a clear config error.

Example fix

# before
patch_size = (t, h, w)  # t = num_latent_frames // pt, may be 0 for empty clip

# after
if t <= 0 or h <= 0 or w <= 0:
    raise ValueError("empty video segment")
patch_size = (t, h, w)

Defensive patterns

Strategy: validation

Validate before calling

assert all(x > 0 for x in patch_size), f"patch_size entries must be > 0, got {patch_size}"
assert all(x > 0 for x in latent_shape), f"latent_shape entries must be > 0, got {latent_shape}"

Type guard

def is_positive_tuple(v) -> bool:
    return len(v) > 0 and all(isinstance(x, int) and x > 0 for x in v)

Prevention

Skip empty media segments before computing latent shapes.
Never compute dims via expressions like n - k without guarding n > k.

When it happens

Trigger: Calling minimax_h3_patchify_video_latent with patch_size=(1, 0, 16) or minimax_h3_unpatchify_video_tokens with latent_shape containing a 0 dim (e.g. (0, 8, 8, 16)).

Common situations: Zero-dimensional latents from empty video segments or dropped frames; negative values from signed arithmetic upstream (e.g. frame_count - 1 when frame_count == 0); a default patch_size left as (0,0,0) in config before initialization.

Understand the failure class

Background: "Must be a positive integer", "Invalid value", "Unsupported": the invalid-argument-value error family, when a library rejects the value you pass — this error's family across 35 libraries.

Related errors

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/67fcf2e478191ceb. Report an issue: GitHub.