Merge pull request #4006 from AI-Hypercomputer:decoupled-tpu7x-zero1-skip

PiperOrigin-RevId: 922847201

Author: Google-ML-Automation

tests/integration/train_tests.py

@@ -500,6 +500,8 @@ def test_base_model_shardy_false(self):
 
   @pytest.mark.integration_test
   @pytest.mark.tpu_only
+  # TODO(b/517509898): Skip ZeRo-1 compiler Segfault on TPU7x SparseCore platforms
+  @pytest.mark.skip_on_tpu7x
   def test_tpu_zero1_gradient_accumulation(self):
     zero1_ga = [  # tests Zero-1 optimizer sharding with gradient accumulation
         None,