sgl-project/sglang · error · ValueError

scale_shift_table must have shape [9, D]

Error message

scale_shift_table must have shape [9, D]

What it means

ltx2_ada_values9 expects scale_shift_table with shape [9, D] — nine rows (one per modulation parameter) and hidden dim D. A tensor with a different rank or a first dimension != 9 fails this check.

Source

Thrown at python/sglang/kernels/ops/diffusion/modulate/ltx2_ada_values_triton.py:149

    tl.store(out4_ptr + base, (table4 + temb4).to(tl.bfloat16), mask=mask)
    tl.store(out5_ptr + base, (table5 + temb5).to(tl.bfloat16), mask=mask)
    tl.store(out6_ptr + base, (table6 + temb6).to(tl.bfloat16), mask=mask)
    tl.store(out7_ptr + base, (table7 + temb7).to(tl.bfloat16), mask=mask)
    tl.store(out8_ptr + base, (table8 + temb8).to(tl.bfloat16), mask=mask)


def ltx2_ada_values9(
    scale_shift_table: torch.Tensor,
    timestep: torch.Tensor,
) -> tuple[torch.Tensor, ...]:
    if timestep.ndim != 3:
        raise ValueError("timestep must have shape [B, S, 9 * D]")
    if not timestep.is_cuda or timestep.dtype != torch.bfloat16:
        raise ValueError("timestep must be a CUDA bfloat16 tensor")
    if not timestep.is_contiguous():
        raise ValueError("timestep must be contiguous")
    if scale_shift_table.ndim != 2 or scale_shift_table.shape[0] != 9:
        raise ValueError("scale_shift_table must have shape [9, D]")
    if (
        not scale_shift_table.is_cuda
        or scale_shift_table.dtype not in (torch.bfloat16, torch.float32)
        or scale_shift_table.stride(-1) != 1
    ):
        raise ValueError(
            "scale_shift_table must be CUDA, bf16/fp32, last-dim contiguous"
        )

    total_params = int(scale_shift_table.shape[0])
    hidden = int(scale_shift_table.shape[1])
    if hidden <= 0 or timestep.shape[-1] != total_params * hidden:
        raise ValueError("timestep last dim must equal 9 * hidden")
    if hidden % 256 != 0 or hidden > 8192:
        raise ValueError("hidden size is outside the supported LTX2 fast-path range")

    batch, seq, _ = timestep.shape
    rows = int(batch * seq)

View on GitHub (pinned to 0132848349)

Solutions

1. Verify scale_shift_table.shape == (9, hidden_dim) before calling
2. Load the correct LTX2 checkpoint parameter; transpose if the checkpoint stores [D, 9]
3. If porting a 6-param AdaLN model, use the appropriate kernel, not ltx2_ada_values9

Example fix

# before
vals = ltx2_ada_values9(table_6row, t)
# after
assert table.shape[0] == 9
vals = ltx2_ada_values9(table, t)

assert scale_shift_table.ndim == 2 and scale_shift_table.shape[0] == 9

def table_ok(t: torch.Tensor) -> bool:
    return t.dim() == 2 and t.shape[0] == 9

When it happens

Trigger: Passing a table with 6 rows (standard 6-parameter AdaLN instead of LTX2's 9-parameter variant), a transposed [D, 9] table, or a 1D/3D tensor.

Common situations: Reusing scale_shift_table from a non-LTX2 model (e.g. DiT with shift/scale/gate only); loading a checkpoint whose table layout is transposed; passing the wrong model parameter.

Related errors

• timestep must have shape [B, S, 9 * D]
• timestep last dim must equal 9 * hidden
• MiniMax H3 AdaLN cache is only compatible with unquantized w
• MiniMax H3 pruned curve checkpoints cannot use a separate Ad
• --model-variant {server_args.model_variant} requires '{cls._

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/2b372a17ec87beac. Report an issue: GitHub.