From 4741239db7b75949d577f99eb1c3a9e463dc77f3 Mon Sep 17 00:00:00 2001 From: Wentao Ye <44945378+yewentao256@users.noreply.github.com> Date: Mon, 22 Sep 2025 22:04:15 -0400 Subject: [PATCH] [Bug] Fix Long Context OOM Issue (#25290) Signed-off-by: yewentao256 --- vllm/v1/attention/backends/mla/common.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/vllm/v1/attention/backends/mla/common.py b/vllm/v1/attention/backends/mla/common.py index 9bca87c81d179..a177117a50bd1 100755 --- a/vllm/v1/attention/backends/mla/common.py +++ b/vllm/v1/attention/backends/mla/common.py @@ -481,7 +481,7 @@ class MLACommonMetadataBuilder(AttentionMetadataBuilder[M]): # which would result in up-projected context being # 2*(192*128)*(64*1024) = 3gb # (assuming 192 QK head dim, 128 heads, and fp16) - 128 * 1024) + 64 * 1024) assert self.chunked_prefill_workspace_size >= \ scheduler_config.max_num_seqs * cache_config.block_size if self.dcp_world_size > 1: