fix

.github/workflows/build_on_pr.yml

@@ -161,7 +161,7 @@ jobs:
           --ignore tests/test_infer_ops \
           --ignore tests/test_legacy \
           --ignore tests/test_smoothquant \
-          tests/test_fp8/
+          tests/
         env:
           LD_LIBRARY_PATH: /github/home/.tensornvme/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64
           LLAMA_PATH: /data/scratch/llama-tiny

tests/test_device/test_init_logical_pg.py

@@ -1,4 +1,3 @@
-import pytest
 import torch
 import torch.distributed as dist
 from torch.distributed import ReduceOp
@@ -27,7 +26,6 @@ def check_layer(rank, world_size, port):
         assert tensor.equal(tensor_to_check)
 
 
-@pytest.mark.skip("tested in corresponding sharderformer")
 @pytest.mark.dist
 @rerun_if_address_is_in_use()
 def test_logical_pg():

tests/test_fp8/test_fp8_allgather.py

@@ -13,7 +13,7 @@ from colossalai.testing import parameterize, rerun_if_address_is_in_use, spawn,
     "shape",
     [(3, 7, 16)],
 )
-
+@clear_cache_before_run()
 @parameterize("dtype", [torch.bfloat16, torch.float16])
 @parameterize("fp8_format", ["e4m3", "e5m2"])
 @parameterize("async_op", [True, False])