From dd0fa02547aae9f57e1cb1c80b7db50a4161b8d2 Mon Sep 17 00:00:00 2001 From: Du Bin Date: Wed, 29 Apr 2026 09:10:51 +0800 Subject: [PATCH] [Bugfix] Add missing .gpu accessor for inputs_embeds CpuGpuBuffer in prefill overlay (#2068) Signed-off-by: dubin555 --- vllm_omni/worker/gpu_model_runner.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/vllm_omni/worker/gpu_model_runner.py b/vllm_omni/worker/gpu_model_runner.py index efad5f3d4..893740b75 100644 --- a/vllm_omni/worker/gpu_model_runner.py +++ b/vllm_omni/worker/gpu_model_runner.py @@ -1120,7 +1120,7 @@ class OmniGPUModelRunner(GPUModelRunner): dtype=self.dtype, device=self.device, non_blocking=True ) start_offset = int(self.query_start_loc.cpu[req_index]) - self.inputs_embeds[start_offset : start_offset + overlay_len].copy_(src) + self.inputs_embeds.gpu[start_offset : start_offset + overlay_len].copy_(src) def _update_additional_information(self, scheduler_output: "SchedulerOutput") -> None: for new_req in scheduler_output.scheduled_new_reqs: