diff --git a/.buildkite/check-torch-abi.py b/.buildkite/check-torch-abi.py
index cb580fb56f..6aa51d66bf 100644
--- a/.buildkite/check-torch-abi.py
+++ b/.buildkite/check-torch-abi.py
@@ -14,7 +14,6 @@ from torch_abi_audit.report import ExtensionReport, PackageReport
 # Shrink and remove over time.
 ALLOWED_UNSTABLE_LIBRARIES: tuple[str, ...] = (
     "_flashkda_C.abi3.so",
-    "vllm_flash_attn/_vllm_fa2_C.abi3.so",
     "vllm_flash_attn/_vllm_fa3_C.abi3.so",
     "third_party/deep_gemm/_C*.so",
 )
diff --git a/cmake/external_projects/vllm_flash_attn.cmake b/cmake/external_projects/vllm_flash_attn.cmake
index 15a4f10b7a..c7ebbb7fd0 100644
--- a/cmake/external_projects/vllm_flash_attn.cmake
+++ b/cmake/external_projects/vllm_flash_attn.cmake
@@ -39,7 +39,7 @@ else()
   FetchContent_Declare(
           vllm-flash-attn
           GIT_REPOSITORY https://github.com/vllm-project/flash-attention.git
-          GIT_TAG ec4a882d70b19235c824848db944d84a6cb8f9ee
+          GIT_TAG 28e862d21806bc3580207aa0ad4e2759151e9827
           GIT_PROGRESS TRUE
           # Don't share the vllm-flash-attn build between build types
           BINARY_DIR ${CMAKE_BINARY_DIR}/vllm-flash-attn