A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 8, 12:30 UTC
[Bugfix] Fix weights offloading for sleep mode (#32947)
Signed-off-by: Jarno Seppänen <jseppanen@nvidia.com> Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com>
J Seppänen committed
Feb 10, 2026 at 22:38 UTC
506ad7d7c178ac20f2140cfaac1ae657683e8013
1 file changed
+4
-3
vllm/v1/worker/gpu_worker.py
+4
-3
@@ -286,9 +286,10 @@ class Worker(WorkerBase):
286
# to hijack tensor allocation.
287
def load_model(self) -> None:
288
eep_scale_up = os.environ.get("VLLM_ELASTIC_EP_SCALE_UP_LAUNCH") == "1"
289
- with self._maybe_get_memory_pool_context(
290
- tag="weights"
291
- ) and set_current_vllm_config(self.vllm_config):
289
+ with (
290
+ self._maybe_get_memory_pool_context(tag="weights"),
291
+ set_current_vllm_config(self.vllm_config),
292
+ ):
293
self.model_runner.load_model(eep_scale_up=eep_scale_up)
294
295
def update_config(self, overrides: dict[str, Any]) -> None: