diff --git a/vllm/v1/attention/backends/mla/common.py b/vllm/v1/attention/backends/mla/common.py
index a990cb2f1a97..893248afcc30 100755
--- a/vllm/v1/attention/backends/mla/common.py
+++ b/vllm/v1/attention/backends/mla/common.py
@@ -480,7 +480,7 @@ def __init__(self,
             # 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: