From 0a5ad6f0d42c1ace54f27d7dd6a852be86e42148 Mon Sep 17 00:00:00 2001 From: Woosuk Kwon Date: Wed, 26 Aug 2026 11:59:09 -0700 Subject: [PATCH] [Model] Remove unused DeepSeek V4 top-k buffer helper (#53697) --- vllm/models/deepseek_v4/attention.py | 7 ------- 1 file changed, 7 deletions(-) diff --git a/vllm/models/deepseek_v4/attention.py b/vllm/models/deepseek_v4/attention.py index a7e8783873..81897398e3 100644 --- a/vllm/models/deepseek_v4/attention.py +++ b/vllm/models/deepseek_v4/attention.py @@ -685,13 +685,6 @@ class DeepseekV4Attention(nn.Module, AttentionLayerBase, ABC): ) return q_fp8 - def _global_topk_output_buffers( - self, topk_indices: torch.Tensor - ) -> tuple[torch.Tensor, torch.Tensor] | None: - if self.compress_ratio != 4 or self.eager_scratch_pool is None: - return None - return self.eager_scratch_pool.global_topk_outputs(topk_indices) - def bind_kv_cache(self, kv_cache: torch.Tensor) -> None: # [B, H=1, N, C] -> [B, N, C] self.kv_cache = kv_cache.squeeze(1)