From 1668e5976db7457631b84a6a570c64319581b1e7 Mon Sep 17 00:00:00 2001 From: "zhao, zhenhui" Date: Mon, 28 Sep 2026 13:31:42 +0800 Subject: [PATCH] [CPU] Use accelerator memory API in DiffusionGemma (#58964) Signed-off-by: zhejiangxiaomai Co-authored-by: Li, Jiang --- vllm/model_executor/models/diffusion_gemma.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/vllm/model_executor/models/diffusion_gemma.py b/vllm/model_executor/models/diffusion_gemma.py index 1ae1cc7116..8f26edf9e4 100644 --- a/vllm/model_executor/models/diffusion_gemma.py +++ b/vllm/model_executor/models/diffusion_gemma.py @@ -1493,7 +1493,7 @@ class DiffusionSampler: # [tile * W, vocab] copies, so a tile is also bounded by free memory. widths_np = states.canvas_width_np[decode_slots_np] order = np.argsort(widths_np, kind="stable") - free = current_platform.mem_get_info()[0] if num_decode > 0 else 0 + free = torch.accelerator.get_memory_info()[0] if num_decode > 0 else 0 run_start = 0 while run_start < num_decode: W = int(widths_np[order[run_start]])