diff --git a/vllm/attention/ops/paged_attn.py b/vllm/attention/ops/paged_attn.py
index 076f151ffcb61..92023d5b75f5a 100644
--- a/vllm/attention/ops/paged_attn.py
+++ b/vllm/attention/ops/paged_attn.py
@@ -34,7 +34,7 @@ class PagedAttention:
 
     @staticmethod
     def get_supported_head_sizes() -> List[int]:
-        return [32, 64, 80, 96, 112, 120, 128, 192, 256]
+        return [64, 80, 96, 112, 120, 128, 192, 256]
 
     @staticmethod
     def get_kv_cache_shape(