sgl-project/sglang · error · ValueError

DFLASH requires explicit layer_ids for aux hidden capture.

Error message

DFLASH requires explicit layer_ids for aux hidden capture.

What it means

Same DFLASH guard as other Qwen models: enabling aux hidden state capture for DFLASH speculative decoding requires an explicit list of layer ids; None is rejected.

Source

Thrown at python/sglang/srt/models/qwen3_moe.py:1104

        self.capture_aux_hidden_states = True
        if layer_ids is None:
            num_layers = self.config.num_hidden_layers
            self.model.set_eagle3_layers_to_capture(
                [
                    2,
                    num_layers // 2,
                    num_layers - 3,
                ]
            )  # Specific layers for EAGLE3 support
        else:
            self.model.set_eagle3_layers_to_capture([val + 1 for val in layer_ids])

    def set_dflash_layers_to_capture(self, layer_ids: List[int]):
        if not self.pp_group.is_last_rank:
            return

        if layer_ids is None:
            raise ValueError(
                "DFLASH requires explicit layer_ids for aux hidden capture."
            )

        self.capture_aux_hidden_states = True
        self.model.set_dflash_layers_to_capture([val + 1 for val in layer_ids])

    def load_weights(
        self, weights: Iterable[Tuple[str, torch.Tensor]], is_mtp: bool = False
    ):
        stacked_params_mapping = [
            # (param_name, shard_name, shard_id)
            ("qkv_proj", "q_proj", "q"),
            ("qkv_proj", "k_proj", "k"),
            ("qkv_proj", "v_proj", "v"),
            ("gate_up_proj", "gate_proj", 0),
            ("gate_up_proj", "up_proj", 1),
        ]

View on GitHub (pinned to 0132848349)

Solutions

1. Pass explicit layer indices, e.g. via --dflash-capture-layers
2. Verify the spec algorithm config computes layer ids before model init
3. Update to a config where capture layers default to a computed set

Example fix

# before
model.set_dflash_layers_to_capture(None)
# after
model.set_dflash_layers_to_capture([0, 1, 2, 3])

if layer_ids is None: raise SystemExit("specify --dflash-capture-layers")
model.set_dflash_layers_to_capture(layer_ids)

def valid_layer_ids(x) -> bool:
    return isinstance(x, list) and all(isinstance(i, int) and i >= 0 for i in x) and len(x) > 0

When it happens

Trigger: Calling Qwen3MoeForCausalLM.set_dflash_layers_to_capture(None) on the last pipeline rank while enabling DFLASH.

Common situations: DFLASH/spec config missing the capture-layer list when serving a Qwen3-MoE model.

Understand the failure class

Background: "missing required argument" and "the following required arguments were not provided": what required-argument errors mean and how to fix them — this error's family across 20 libraries.

Related errors

• DFLASH speculative decoding only supports CUDA and NPU devic
• Currently DFLASH speculative decoding does not support dp at
• Currently DFLASH speculative decoding only supports pp_size
• DFLASH speculative decoding requires setting --speculative-d
• DFLASH requires --speculative-dflash-block-size to be positi

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/962a79938d81e089. Report an issue: GitHub.