Skip to content

vllm.models.deepseek_v4.common.eplb_util

Shared EPLB MoE registration helpers for DeepSeek V4 family models.

Functions:

collect_moe_layers(moe_model, layers, config, *, decoder_layer_type, moe_type)

Populate MixtureOfExperts fields from decoder layers for EPLB.

Works for both target backbones and DSpark draft stacks. Draft models pass their shorter model.layers list; pipeline-parallel placeholder layers are skipped via decoder_layer_type.

Source code in vllm/models/deepseek_v4/common/eplb_util.py 
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
def collect_moe_layers(
    moe_model: Any,
    layers: Iterable[nn.Module],
    config: object,
    *,
    decoder_layer_type: type[nn.Module],
    moe_type: type[nn.Module],
) -> None:
    """Populate ``MixtureOfExperts`` fields from decoder layers for EPLB.

    Works for both target backbones and DSpark draft stacks. Draft models pass
    their shorter ``model.layers`` list; pipeline-parallel placeholder layers
    are skipped via ``decoder_layer_type``.
    """
    moe_model.num_expert_groups = getattr(config, "n_group", 1)
    moe_model.moe_layers = []
    moe_model.moe_mlp_layers = []
    example_moe = None
    for layer in layers:
        if not isinstance(layer, decoder_layer_type):
            continue
        if not isinstance(layer.ffn, moe_type):
            continue
        example_moe = layer.ffn
        moe_model.moe_mlp_layers.append(layer.ffn)
        moe_model.moe_layers.append(layer.ffn.experts)

    moe_model.num_moe_layers = len(moe_model.moe_layers)
    moe_model.extract_moe_parameters(example_moe)

dspark_draft_supports_eplb(draft_model_config)

Return whether a DSpark draft can share EPLB state with the target.

Only DeepSeek-V4 DSpark drafts reuse the target expert layout. V4.1 drafts use a smaller routed-expert count and cannot share EPLB state.

Source code in vllm/models/deepseek_v4/common/eplb_util.py 
13
14
15
16
17
18
19
def dspark_draft_supports_eplb(draft_model_config: ModelConfig) -> bool:
    """Return whether a DSpark draft can share EPLB state with the target.

    Only DeepSeek-V4 DSpark drafts reuse the target expert layout. V4.1 drafts
    use a smaller routed-expert count and cannot share EPLB state.
    """
    return getattr(draft_model_config.hf_config, "model_type", None) == "deepseek_v4"