From 6147c70224394a620ff3a56190a16539bdebdc3d Mon Sep 17 00:00:00 2001 From: Wentao Ye <44945378+yewentao256@users.noreply.github.com> Date: Fri, 15 May 2026 12:41:23 -0400 Subject: [PATCH] [Model Runner v2] Support reload weights (sleep mode) (#42673) Signed-off-by: yewentao256 Signed-off-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> --- vllm/v1/worker/gpu/model_runner.py | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/vllm/v1/worker/gpu/model_runner.py b/vllm/v1/worker/gpu/model_runner.py index 4fc5a754db..091138d9ab 100644 --- a/vllm/v1/worker/gpu/model_runner.py +++ b/vllm/v1/worker/gpu/model_runner.py @@ -329,6 +329,14 @@ class GPUModelRunner(LoRAModelRunnerMixin): def get_model(self) -> nn.Module: return self.model + def reload_weights(self, *args, **kwargs) -> None: + # TODO(Wentao): Use full version instead of import when fully migrated to v2 + from vllm.v1.worker.gpu_model_runner import GPUModelRunner as GPUModelRunnerV1 + + GPUModelRunnerV1.reload_weights(self, *args, **kwargs) # type: ignore[arg-type] + self.reset_encoder_cache() + self.reset_mm_cache() + @functools.cached_property def main_stream(self) -> torch.cuda.Stream: # Cache the default CUDA stream to avoid lookup overhead.