github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-10-25 01:40:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

[test] fixed rerun_on_exception and adapted test cases (#487)

Browse Source
This commit is contained in:
Frank Lee
2022-03-25 17:25:12 +08:00
committed by GitHub
parent 4d322b79da
commit 3601b2bad0
31 changed files with 143 additions and 135 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
tests/test_zero_data_parallel/test_zero_engine.py
View File

@@ -13,6 +13,7 @@ from colossalai.utils import free_port
from colossalai.zero.init_ctx import ZeroInitContext
from colossalai.zero.sharded_model.utils import col_model_deepcopy
from colossalai.zero.sharded_optim._utils import has_inf_or_nan
from colossalai.testing import rerun_on_exception
from tests.components_to_test.registry import non_distributed_component_funcs
from torch.nn.parallel import DistributedDataParallel as DDP
@@ -96,6 +97,7 @@ def run_dist(rank, world_size, port, parallel_config):
@pytest.mark.skip
@pytest.mark.dist
@pytest.mark.parametrize("world_size", [2, 4])
@rerun_on_exception(exception_type=mp.ProcessRaisedException, pattern=".*Address already in use.*")
def test_mp_engine(world_size):
run_func = partial(run_dist, world_size=world_size, port=free_port(), parallel_config=MP_PARALLEL_CONFIG)
mp.spawn(run_func, nprocs=world_size)
@@ -103,6 +105,7 @@ def test_mp_engine(world_size):
@pytest.mark.dist
@pytest.mark.parametrize("world_size", [1, 2])
@rerun_on_exception(exception_type=mp.ProcessRaisedException, pattern=".*Address already in use.*")
def test_zero_engine(world_size):
run_func = partial(run_dist, world_size=world_size, port=free_port(), parallel_config=ZERO_PARALLEL_CONFIG)
mp.spawn(run_func, nprocs=world_size)