From bb9d31f22d8930ddaa909fb9e4cd234358e7e8a8 Mon Sep 17 00:00:00 2001 From: syy-hw Date: Thu, 18 Jun 2026 17:55:50 +0800 Subject: [PATCH] [NPU] Add head_dim=256 to _can_use_tnd whitelist (#28635) --- .../sglang/srt/hardware_backend/npu/attention/ascend_backend.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/python/sglang/srt/hardware_backend/npu/attention/ascend_backend.py b/python/sglang/srt/hardware_backend/npu/attention/ascend_backend.py index 62bf02555..a0a708839 100644 --- a/python/sglang/srt/hardware_backend/npu/attention/ascend_backend.py +++ b/python/sglang/srt/hardware_backend/npu/attention/ascend_backend.py @@ -376,7 +376,7 @@ class AscendAttnBackend(AttentionBackend): """Check if TND layout is supported.""" d = layer.qk_head_dim v = layer.v_head_dim - return (d == v and d in (128, 192)) or (d == 192 and v == 128) + return (d == v and d in (128, 192, 256)) or (d == 192 and v == 128) def get_verify_buffers_to_fill_after_draft(self): """