sgl-project/sglang · error · ValueError

The hpc_ops MoE runner backend does not support no_combine (

Error message

The hpc_ops MoE runner backend does not support no_combine (the fused kernel always reduces over top-k experts).

What it means

The hpc_ops fused MoE kernel always reduces (sums) outputs over the top-k experts, so it cannot honor no_combine=True (returning per-expert outputs without combining). _check_runner_config_supported raises when no_combine is set.

Source

Thrown at python/sglang/srt/layers/moe/moe_runner/hpc_ops.py:112

def _check_runner_config_supported(runner_config: MoeRunnerConfig) -> None:
    if runner_config.activation != "silu" or not runner_config.is_gated:
        raise ValueError(
            "The hpc_ops MoE runner backend only supports the gated silu "
            f"activation, got activation={runner_config.activation}, "
            f"is_gated={runner_config.is_gated}."
        )
    if runner_config.num_fused_shared_experts != 0:
        raise ValueError(
            "The hpc_ops MoE runner backend does not support fused shared experts."
        )
    if runner_config.apply_router_weight_on_input:
        raise ValueError(
            "The hpc_ops MoE runner backend does not support "
            "apply_router_weight_on_input."
        )
    if runner_config.no_combine:
        raise ValueError(
            "The hpc_ops MoE runner backend does not support no_combine "
            "(the fused kernel always reduces over top-k experts)."
        )
    if (
        runner_config.gemm1_alpha is not None
        or runner_config.gemm1_clamp_limit is not None
        or runner_config.swiglu_limit is not None
    ):
        raise ValueError(
            "The hpc_ops MoE runner backend runs a plain SiLU-and-mul; it does "
            "not support gemm1_alpha / gemm1_clamp_limit / swiglu_limit."
        )


@register_fused_func("none", "hpc_ops")
def fused_experts_none_to_hpc_ops(
    dispatch_output: StandardDispatchOutput,
    quant_info: HpcOpsMoeQuantInfo,

View on GitHub (pinned to 0132848349)

Solutions

1. Switch to a backend that supports no_combine (triton/flashinfer)
2. Disable the feature that requires uncombined outputs
3. Avoid hpc_ops for this model

if server_args.moe_runner_backend == 'hpc_ops' and runner_config.no_combine:
    raise SystemExit('hpc_ops always combines top-k outputs; pick triton')

When it happens

Trigger: Running with --moe-runner-backend hpc_ops on a model/feature path that sets runner_config.no_combine=True (e.g. certain speculative or per-expert output paths).

Common situations: Using hpc_ops with models or features (e.g. EAGLE-style draft layers over MoE) that need uncombined expert outputs.

Related errors

• The hpc_ops MoE runner backend does not support fused shared
• The hpc_ops MoE runner backend does not support apply_router
• The hpc_ops MoE runner backend runs a plain SiLU-and-mul; it
• The hpc_ops MoE runner backend only supports FP8-quantized M
• Unsupported activation: {self.activation}

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/bfd5639c9425f9c1. Report an issue: GitHub.