added CI for unit testing (#69)

tests/test_layers/test_3d/checks_3d/check_layer_3d.py

@@ -13,7 +13,7 @@ from colossalai.utils import get_current_device, print_rank_0
 from colossalai.nn.layer.parallel_3d._utils import get_parallel_mode_from_env
 from colossalai.constants import INPUT_GROUP_3D, WEIGHT_GROUP_3D, OUTPUT_GROUP_3D
 
-from common import *
+from .common import *
 
 
 def check_linear():

tests/test_layers/test_3d/checks_3d/check_operation_3d.py

@@ -7,7 +7,7 @@ from colossalai.logging import get_dist_logger
 from colossalai.nn.layer.parallel_3d._operation import *
 from colossalai.utils import get_current_device
 
-from common import *
+from .common import *
 
 
 def check_AB():

tests/test_layers/test_3d/test.sh

@@ -1,22 +0,0 @@
-#!/bin/bash
-
-python -m torch.distributed.launch test_2d.py --nproc_per_node 8 test_3d.py --host $HOST --port 29516 --world_size 8
-
-# expected test output
-#  distributed environment initialized
-#  AB forward: pass
-#  AB backward: pass
-#  ABT forward: pass
-#  ABT backward: pass
-#  ATB forward: pass
-#  ATB backward: pass
-#  linear backward: pass
-#  linear backward: pass
-#  layer norm forward: pass
-#  layer norm backward: pass
-#  self attention forward: pass
-#  self attention backward: pass
-#  mlp forward: pass
-#  mlp backward: pass
-#  transformerlayer forward: pass
-#  transformerlayer backward: pass

tests/test_layers/test_3d/test_3d.py

@@ -1,11 +1,14 @@
 #!/usr/bin/env python
 # -*- encoding: utf-8 -*-
-
+import pytest
+import torch
+import torch.multiprocessing as mp
 from colossalai.initialize import launch, get_default_parser
 
-from test_layer import *
-from test_operation import *
+from checks_3d.check_layer_3d import *
+from checks_3d.check_operation_3d import *
 from colossalai.logging import get_dist_logger
+from functools import partial
 
 CONFIG = dict(parallel=dict(pipeline=1, tensor=dict(mode='3d', size=8)),
               seed=0)
@@ -38,26 +41,25 @@ def check_layer():
         ranks=[0])
 
 
-def _test_main():
-    # init dist
-    parser = get_default_parser()
-    args = parser.parse_args()
+def check_layer_and_operation(rank, world_size):
     launch(config=CONFIG,
-           rank=args.rank,
-           world_size=args.world_size,
-           host=args.host,
-           port=args.port,
-           backend=args.backend)
-    logger = get_dist_logger()
-    logger.info('Distributed environment is initialzied.', ranks=[0])
-    torch.backends.cudnn.benchmark = True
+           rank=rank,
+           world_size=world_size,
+           host='localhost',
+           port=29923,
+           backend='nccl')
 
-    # check operation
-    # check_operations()
-
-    # check layers
     check_layer()
+    gpc.destroy()
+    torch.cuda.empty_cache()
+
+
+@pytest.mark.dist
+def test_3d():
+    world_size = 8
+    run_func = partial(check_layer_and_operation, world_size=world_size)
+    mp.spawn(run_func, nprocs=world_size)
 
 
 if __name__ == '__main__':
-    _test_main()
+    test_3d()