[DCU] fix bad alloc

tests/pytorch/distributed/run_numerics.py

@@ -10,10 +10,10 @@ import os
 import sys
 from functools import wraps
 
-import transformer_engine.pytorch as te
 import torch
 from torch import nn
 import torch.distributed as dist
+import transformer_engine.pytorch as te
 import transformer_engine_torch as tex
 from transformer_engine.common.recipe import (
     MXFP8BlockScaling,

tests/pytorch/distributed/test_numerics.py

@@ -53,7 +53,7 @@ def test_distributed(quantization):
     if quantization == "fp8" and not fp8_available:
         pytest.skip(reason_for_no_fp8)
     if quantization == "fp8_cs" and not fp8_available:
-        pytest.skip(fp8_available)
+        pytest.skip(reason_for_no_fp8)
     if quantization == "mxfp8" and not mxfp8_available:
         pytest.skip(reason_for_no_mxfp8)
     _run_test(quantization)