sgl-project/sglang · error · ValueError

DSPARK requires explicit layer_ids for aux hidden capture.

Error message

DSPARK requires explicit layer_ids for aux hidden capture.

What it means

DSPARK's aux hidden capture needs to know exactly which decoder layers' hidden states to capture; unlike some speculative paths it cannot infer them. If layer_ids is None (the caller omitted the argument) after the PP and last-rank checks, this ValueError is raised and capture_aux_hidden_states is never set.

Source

Thrown at python/sglang/srt/models/kimi_k3.py:2892

            )
        else:
            self.lm_head = PPMissingLayer()
        logit_scale = getattr(config, "logit_scale", 1.0)
        self.logits_processor = LogitsProcessor(config=config, logit_scale=logit_scale)
        self.capture_aux_hidden_states = False

    def get_input_embeddings(self):
        return self.model.embed_tokens

    def set_dspark_layers_to_capture(self, layer_ids: list[int]) -> None:
        if self.pp_group.world_size > 1:
            # Capture layers living on non-last PP ranks would be silently
            # skipped (the flag is only set on the last rank).
            raise NotImplementedError("DSPARK aux hidden capture requires PP=1.")
        if not self.pp_group.is_last_rank:
            return
        if layer_ids is None:
            raise ValueError(
                "DSPARK requires explicit layer_ids for aux hidden capture."
            )
        self.capture_aux_hidden_states = True
        self.model.dspark_layers_to_capture = list(layer_ids)

    @torch.no_grad()
    def forward(
        self,
        input_ids: torch.Tensor,
        positions: torch.Tensor,
        forward_batch: ForwardBatch,
        input_embeds: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        pp_proxy_tensors: Optional[PPProxyTensors] = None,
    ) -> torch.Tensor:
        embeds = input_embeds if input_embeds is not None else inputs_embeds
        hidden_states = self.model(
            input_ids, positions, forward_batch, embeds, pp_proxy_tensors

View on GitHub (pinned to 0132848349)

Solutions

1. Pass an explicit list of layer ids, e.g. set_dspark_layers_to_capture([0, 1, 2]) matching the DSPARK draft's tdt/esvd layer choices.
2. Upgrade the DSPARK runtime/spec worker to a version that supplies layer_ids for Kimi K3.
3. Validate layer_ids against model.config.num_hidden_layers before calling.

Example fix

# before
model.set_dspark_layers_to_capture(None)

# after
model.set_dspark_layers_to_capture([0, 1, 2])

assert layer_ids is not None and len(layer_ids) > 0, "DSPARK requires explicit layer_ids"
assert all(0 <= i < model.config.num_hidden_layers for i in layer_ids)

def valid_layer_ids(layer_ids, num_layers) -> bool:
    return layer_ids is not None and all(isinstance(i, int) and 0 <= i < num_layers for i in layer_ids)

When it happens

Trigger: Calling set_dspark_layers_to_capture(None) or omitting layer_ids from the DSPARK/spec worker when wiring up Kimi K3 — the method explicitly rejects None even though the signature types it as list[int].

Common situations: A DSPARK integration/spec-worker version that doesn't pass layer_ids; user scripts calling the capture API directly without layer selection; default-argument path after an sglang upgrade changed the call convention.

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

• DSpark requires --speculative-dspark-block-size to be positi
• DSpark speculative_num_draft_tokens must be >= 2 (= gamma +
• DSPARK aux hidden capture requires PP=1.
• Unknown match_type: '{match_type}'. Must be 'BFS' or 'PROB'.
• Kimi-K3 DCP + DSPARK currently requires SGLANG_RAGGED_VERIFY

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/b318b4f842689251. Report an issue: GitHub.