Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Keep the Kaggle GPU harness tests independent of the caller's CUDA_VISIBLE_DEVICES (#12010)

Daniel Han3e7521c19f16
diff --git a/tests/kaggle/test_studio_gpu_harness.py b/tests/kaggle/test_studio_gpu_harness.py
index 58c8019b190..a0b77771102 100644
--- a/tests/kaggle/test_studio_gpu_harness.py
+++ b/tests/kaggle/test_studio_gpu_harness.py
@@ -56,6 +56,13 @@ build_kernel = _load("studio_ci_build_kernel", CI_DIR / "build_kernel.py")
collect_evidence = _load("studio_ci_collect_evidence", CI_DIR / "collect_evidence.py")
[email protected](autouse = True)
+def _no_caller_cuda_visible_devices(monkeypatch):
+ # The samplers scope nvidia-smi by it; a CPU-only run with it set to "" read as no card
+ # visible and failed the mocked-listing tests. Tests that need a value set it themselves.
+ monkeypatch.delenv("CUDA_VISIBLE_DEVICES", raising = False)
+
+
# --------------------------------------------------------------- nvidia-smi