From 506ad7d7c178ac20f2140cfaac1ae657683e8013 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?J=20Sepp=C3=A4nen?= <83203+jseppanen@users.noreply.github.com> Date: Tue, 10 Feb 2026 22:38:17 +0200 Subject: [PATCH] [Bugfix] Fix weights offloading for sleep mode (#32947) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Signed-off-by: Jarno Seppänen Co-authored-by: Tyler Michael Smith --- vllm/v1/worker/gpu_worker.py | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/vllm/v1/worker/gpu_worker.py b/vllm/v1/worker/gpu_worker.py index 9696271705..1c526bab9a 100644 --- a/vllm/v1/worker/gpu_worker.py +++ b/vllm/v1/worker/gpu_worker.py @@ -286,9 +286,10 @@ class Worker(WorkerBase): # to hijack tensor allocation. def load_model(self) -> None: eep_scale_up = os.environ.get("VLLM_ELASTIC_EP_SCALE_UP_LAUNCH") == "1" - with self._maybe_get_memory_pool_context( - tag="weights" - ) and set_current_vllm_config(self.vllm_config): + with ( + self._maybe_get_memory_pool_context(tag="weights"), + set_current_vllm_config(self.vllm_config), + ): self.model_runner.load_model(eep_scale_up=eep_scale_up) def update_config(self, overrides: dict[str, Any]) -> None: