From 908ab01672dd9a69aaff48b96137184cdf20bf4e Mon Sep 17 00:00:00 2001 From: Yifan Qiao Date: Sat, 25 Apr 2026 04:44:09 +0000 Subject: [PATCH] chore: pass mypy Signed-off-by: Yifan Qiao --- vllm/model_executor/layers/deepseek_v4_attention.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/vllm/model_executor/layers/deepseek_v4_attention.py b/vllm/model_executor/layers/deepseek_v4_attention.py index 7c6b1e9c13b..5613f899b2a 100644 --- a/vllm/model_executor/layers/deepseek_v4_attention.py +++ b/vllm/model_executor/layers/deepseek_v4_attention.py @@ -374,8 +374,9 @@ class DeepseekV4MultiHeadLatentAttentionWrapper(PluggableLayer): ) elif self.compressor is not None: # Compressor on default, kv_insert on aux. + compressor = self.compressor maybe_execute_in_parallel( - lambda: self.compressor(hidden_states, positions, self.rotary_emb), + lambda: compressor(hidden_states, positions, self.rotary_emb), lambda: self._fused_qnorm_rope_kv_insert( q, kv, positions, attn_metadata ),