From ea56e108bba75285c195c11023e1a1d983cedafb Mon Sep 17 00:00:00 2001 From: Daniel Galvez Date: Mon, 9 Mar 2026 23:56:50 -0700 Subject: [PATCH 1/3] Run smoke test with CUDA_LOG_FILE=stdout Hopefully this will help diagnose this failure: https://github.com/pytorch/pytorch/pull/168912#issuecomment-4019390053 --- test/smoke_test.py | 1 + 1 file changed, 1 insertion(+) diff --git a/test/smoke_test.py b/test/smoke_test.py index e2a3b5068ab..e9f179ded31 100644 --- a/test/smoke_test.py +++ b/test/smoke_test.py @@ -1,6 +1,7 @@ """Run smoke tests""" import os +os.environ["CUDA_LOG_FILE"] = "stdout" import sys import sysconfig from pathlib import Path From 003f234e8a9c61c840143fe1a62d9e1c25b31004 Mon Sep 17 00:00:00 2001 From: Daniel Galvez Date: Tue, 10 Mar 2026 21:04:50 -0700 Subject: [PATCH 2/3] TORCH_SHOW_CPP_STACKTRACES=1 Try to get an understanding of what precise C++ code is failing. --- test/smoke_test.py | 1 + 1 file changed, 1 insertion(+) diff --git a/test/smoke_test.py b/test/smoke_test.py index e9f179ded31..d237bb411ca 100644 --- a/test/smoke_test.py +++ b/test/smoke_test.py @@ -2,6 +2,7 @@ import os os.environ["CUDA_LOG_FILE"] = "stdout" +os.environ["TORCH_SHOW_CPP_STACKTRACES"] = "1" import sys import sysconfig from pathlib import Path From f906b5dc1fc57fb42360aa95fb20badfaca260c7 Mon Sep 17 00:00:00 2001 From: Daniel Galvez Date: Tue, 10 Mar 2026 22:17:10 -0700 Subject: [PATCH 3/3] call nvidia-smi I am wondering whether the windows driver type might have something to do with the error. --- test/smoke_test.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/test/smoke_test.py b/test/smoke_test.py index d237bb411ca..f66ff8df54a 100644 --- a/test/smoke_test.py +++ b/test/smoke_test.py @@ -3,6 +3,8 @@ import os os.environ["CUDA_LOG_FILE"] = "stdout" os.environ["TORCH_SHOW_CPP_STACKTRACES"] = "1" +import subprocess +print(subprocess.check_output(["nvidia-smi"], text=True)) import sys import sysconfig from pathlib import Path