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.

Parallel-isolation guard: exempt the nvidia-smi fake's poll deadline (#12353)

Daniel Han3d7dde02e320
diff --git a/tests/studio/test_backend_ci_parallel_isolation.py b/tests/studio/test_backend_ci_parallel_isolation.py
index abfa115cf3d..524801d8ecf 100644
--- a/tests/studio/test_backend_ci_parallel_isolation.py
+++ b/tests/studio/test_backend_ci_parallel_isolation.py
@@ -273,6 +273,9 @@ BENIGN_TIMING = {
"test_account_local_model_resolver.py",
"test_a_warm_scan_queues_behind_another_accounts_scan",
),
+ # A poll deadline: FakeSmi.wait_for_call loops until the fake nvidia-smi has logged the expected children and only
+ # asserts `time.monotonic() < deadline` against a 30s budget, so a descheduled worker just polls longer.
+ ("test_gpu_query_cache.py", "wait_for_call"),
}