From a9ca036b0e572ef43497f724c18d918b540c496f Mon Sep 17 00:00:00 2001
From: Hanzhi Zhou <hanzhi_zhou@apple.com>
Date: Tue, 7 Jan 2025 22:55:30 -0800
Subject: [PATCH] Move to top level

---
 axlearn/common/flash_attention/conftest.py => conftest.py | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
 rename axlearn/common/flash_attention/conftest.py => conftest.py (55%)

diff --git a/axlearn/common/flash_attention/conftest.py b/conftest.py
similarity index 55%
rename from axlearn/common/flash_attention/conftest.py
rename to conftest.py
index 5bf85d334..ce4577cec 100644
--- a/axlearn/common/flash_attention/conftest.py
+++ b/conftest.py
@@ -2,7 +2,11 @@
 
 """Configures pytest to distribute tests to multiple GPUs.
 
-Usage on 8 GPU machines: PARALLEL_GPU_TEST=1 pytest -n 8 gpu_attention_test.py
+This is not enabled by default and requires explicit opt-in by setting the environment variable
+PARALLEL_GPU_TEST. This is because not all GPU tests are single-GPU tests.
+
+Example usage on 8 GPU machines:
+PARALLEL_GPU_TEST=1 pytest -n 8 axlearn/common/flash_attention/gpu_attention_test.py
 """
 import os