From 0dd5f3e17d2eb2c79ca30d6b5e1befb78c5296e0 Mon Sep 17 00:00:00 2001 From: tzhouam Date: Sat, 3 Oct 2026 22:34:10 +0800 Subject: [PATCH] Use blockwise Flux offload within single-L4 memory capacity Signed-off-by: tzhouam --- tests/e2e/online_serving/test_flux_kontext_expansion.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tests/e2e/online_serving/test_flux_kontext_expansion.py b/tests/e2e/online_serving/test_flux_kontext_expansion.py index b7a360a40..f27bb8da9 100644 --- a/tests/e2e/online_serving/test_flux_kontext_expansion.py +++ b/tests/e2e/online_serving/test_flux_kontext_expansion.py @@ -24,7 +24,7 @@ def _get_diffusion_feature_cases(model: str): return [ pytest.param( # The unsharded transformer alone exceeds the L4's 22 GiB. - OmniServerParams(model=model, server_args=["--enable-cpu-offload"]), + OmniServerParams(model=model, server_args=["--enable-layerwise-offload"]), id="base", marks=BASE_FEATURE_MARKS, ),