diff --git a/vllm/model_executor/layers/deepseek_v4_attention.py b/vllm/model_executor/layers/deepseek_v4_attention.py index 996534b05f..b2d2ac8164 100644 --- a/vllm/model_executor/layers/deepseek_v4_attention.py +++ b/vllm/model_executor/layers/deepseek_v4_attention.py @@ -47,7 +47,7 @@ from vllm.logger import init_logger from vllm.model_executor.custom_op import PluggableLayer from vllm.model_executor.layers.attention_layer_base import AttentionLayerBase from vllm.model_executor.layers.deepseek_compressor import DeepseekCompressor -from vllm.model_executor.layers.layernorm import LayerNorm, RMSNorm +from vllm.model_executor.layers.layernorm import RMSNorm from vllm.model_executor.layers.quantization import QuantizationConfig from vllm.model_executor.layers.quantization.input_quant_fp8 import ( QuantFP8, @@ -1111,7 +1111,6 @@ class DeepseekV4Indexer(nn.Module): quant_config=None, prefix=f"{prefix}.weights_proj", ) - self.k_norm = LayerNorm(self.head_dim, eps=1e-6) self.softmax_scale = self.head_dim**-0.5 self.scale_fmt = "ue8m0"