diff --git a/python/sglang/srt/models/deepseek_v2.py b/python/sglang/srt/models/deepseek_v2.py index 325bc9ebb..646a943e8 100644 --- a/python/sglang/srt/models/deepseek_v2.py +++ b/python/sglang/srt/models/deepseek_v2.py @@ -2983,9 +2983,9 @@ class DeepseekV2ForCausalLM(nn.Module, DeepseekV2WeightLoaderMixin): self._routed_experts_weights_of_layer = LazyValue( lambda: { - layer_id: layer.mlp.get_moe_weights() - for layer_id, layer in enumerate(self.model.layers) - if isinstance(layer.mlp, DeepseekV2MoE) + layer_id: self.model.layers[layer_id].mlp.get_moe_weights() + for layer_id in range(self.model.start_layer, self.model.end_layer) + if isinstance(self.model.layers[layer_id].mlp, DeepseekV2MoE) } ) self.capture_aux_hidden_states = False