From 235a488c878307da650054b4300007ee3d0cc052 Mon Sep 17 00:00:00 2001 From: heziiop Date: Thu, 23 Jul 2026 14:07:14 +0800 Subject: [PATCH] [NPU] ascend fuseep use moe ep group (#32040) --- python/sglang/srt/hardware_backend/npu/moe/fuseep.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/python/sglang/srt/hardware_backend/npu/moe/fuseep.py b/python/sglang/srt/hardware_backend/npu/moe/fuseep.py index 941c84685..9c5c16c9d 100644 --- a/python/sglang/srt/hardware_backend/npu/moe/fuseep.py +++ b/python/sglang/srt/hardware_backend/npu/moe/fuseep.py @@ -12,7 +12,7 @@ from typing import TYPE_CHECKING import torch -from sglang.srt.distributed import get_tp_group +from sglang.srt.distributed import get_moe_ep_group from sglang.srt.environ import envs from sglang.srt.hardware_backend.npu.utils import npu_format_cast from sglang.srt.layers.moe.token_dispatcher.deepep import DeepEPBuffer @@ -30,7 +30,7 @@ _PARAMS_BYTES = 2 # bf16 — Ascend's Dispatch & Combine does not support fp16 def _get_fuseep_buffer(layer: FusedMoE): DeepEPBuffer.set_dispatch_mode_as_low_latency() return DeepEPBuffer.get_deepep_buffer( - get_tp_group().device_group, + get_moe_ep_group().device_group, layer.hidden_size, _PARAMS_BYTES, DeepEPMode.LOW_LATENCY,