已合并
【fix】修复assertError类型报错的问题 #31687
xiaoqi-zhou创建于 3月12日
【fix】修复assertError类型报错的问题 #31687
已合并
共 1 个文件变更+3-3
| @@ -361,13 +361,13 @@ class TestDistributedFailure(ShardedTensorTestBase): | |||
| 361 | self._test_load(state_dict) | 361 | self._test_load(state_dict) |
| 362 | self._test_load(state_dict, fail_set_up_storage_reader=[0]) | 362 | self._test_load(state_dict, fail_set_up_storage_reader=[0]) |
| 363 | self._test_load(state_dict, fail_prepare_global_plan=[0]) | 363 | self._test_load(state_dict, fail_prepare_global_plan=[0]) |
| 364 | - self._test_load(state_dict, fail_read_metadata=[0]) | 364 | + self._test_load(state_dict, fail_read_metadata=[0], ignore_exception_type=True) |
| 365 | self._test_load(state_dict, fail_prepare_local_plan=[1]) | 365 | self._test_load(state_dict, fail_prepare_local_plan=[1]) |
| 366 | self._test_load(state_dict, fail_read_data=[3]) | 366 | self._test_load(state_dict, fail_read_data=[3]) |
| 367 | self._test_load(state_dict, fail_read_data_async=[1]) | 367 | self._test_load(state_dict, fail_read_data_async=[1]) |
| 368 | 368 | ||
| 369 | self._test_load(state_dict, coordinator=3, fail_set_up_storage_reader=[0]) | 369 | self._test_load(state_dict, coordinator=3, fail_set_up_storage_reader=[0]) |
| 370 | - self._test_load(state_dict, coordinator=1, fail_read_metadata=[3]) | 370 | + self._test_load(state_dict, coordinator=1, fail_read_metadata=[3], ignore_exception_type=True) |
| 371 | self._test_load(state_dict, coordinator=2, fail_read_data=[0]) | 371 | self._test_load(state_dict, coordinator=2, fail_read_data=[0]) |
| 372 | self._test_load(state_dict, coordinator=3, fail_read_data_async=[2]) | 372 | self._test_load(state_dict, coordinator=3, fail_read_data_async=[2]) |
| 373 | self._test_load(state_dict, coordinator=1, fail_prepare_global_plan=[1]) | 373 | self._test_load(state_dict, coordinator=1, fail_prepare_global_plan=[1]) |
| @@ -376,7 +376,7 @@ class TestDistributedFailure(ShardedTensorTestBase): | |||
| 376 | state_dict = {"replicated": torch.rand(10, 10), "bytes": [1, 2, 3, 4]} | 376 | state_dict = {"replicated": torch.rand(10, 10), "bytes": [1, 2, 3, 4]} |
| 377 | self._test_load(state_dict) | 377 | self._test_load(state_dict) |
| 378 | self._test_load(state_dict, fail_set_up_storage_reader=[0]) | 378 | self._test_load(state_dict, fail_set_up_storage_reader=[0]) |
| 379 | - self._test_load(state_dict, fail_read_metadata=[0]) | 379 | + self._test_load(state_dict, fail_read_metadata=[0], ignore_exception_type=True) |
| 380 | self._test_load(state_dict, fail_prepare_local_plan=[0]) | 380 | self._test_load(state_dict, fail_prepare_local_plan=[0]) |
| 381 | self._test_load(state_dict, fail_prepare_global_plan=[0]) | 381 | self._test_load(state_dict, fail_prepare_global_plan=[0]) |
| 382 | self._test_load(state_dict, fail_read_data=[0]) | 382 | self._test_load(state_dict, fail_read_data=[0]) |