diff --git a/vllm/model_executor/layers/deepseek_v4_attention.py b/vllm/model_executor/layers/deepseek_v4_attention.py
index 847c3eee55a8..759578f6eb84 100644
--- a/vllm/model_executor/layers/deepseek_v4_attention.py
+++ b/vllm/model_executor/layers/deepseek_v4_attention.py
@@ -352,7 +352,6 @@ def compressor_kv_score() -> torch.Tensor:
                 return torch.mm(
                     hidden_states,
                     compressor.fused_wkv_wgate.weight.T,
-                    out_dtype=torch.float32,
                 )
 
             aux_fns[0] = compressor_kv_score
@@ -369,7 +368,6 @@ def indexer_compressor_kv_score() -> torch.Tensor:
                 return torch.mm(
                     hidden_states,
                     indexer.compressor.fused_wkv_wgate.weight.T,
-                    out_dtype=torch.float32,
                 )
 
             aux_fns[1] = indexer_weights_proj