diff --git a/tests/unit_tests/transformer/experimental_attention_variant/test_attention_variant_dsa.py b/tests/unit_tests/transformer/experimental_attention_variant/test_attention_variant_dsa.py
index 96253a4ca10..45470d4dd6c 100644
--- a/tests/unit_tests/transformer/experimental_attention_variant/test_attention_variant_dsa.py
+++ b/tests/unit_tests/transformer/experimental_attention_variant/test_attention_variant_dsa.py
@@ -423,6 +423,7 @@ def test_fused_indexer_loss_gradient_matches_autograd(self, seqlen_and_topk, spa
 class TestFusedDSAIndexerLossGradientTP:
     """Test FusedDSAIndexerLoss gradient consistency across different TP sizes."""
 
+    @pytest.mark.flaky_in_dev
     @pytest.mark.skipif(not torch.cuda.is_available(), reason="CUDA not available")
     def test_fused_indexer_loss_gradient_tp_consistency(
         self, tensor_model_parallel_size, sparse_loss