EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

fix: use DeepEPDispatcher on CUDA (#5180)

Browse Source
This commit is contained in:
Yineng Zhang
2025-04-08 21:56:53 -07:00
committed by GitHub
parent 6669d12707
commit 90caf06c00
1 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
python/sglang/srt/models/deepseek_v2.py
View File

@@ -50,7 +50,6 @@ from sglang.srt.layers.linear import (
) )
from sglang.srt.layers.logits_processor import LogitsProcessor from sglang.srt.layers.logits_processor import LogitsProcessor
from sglang.srt.layers.moe.ep_moe.layer import DeepEPMoE, EPMoE from sglang.srt.layers.moe.ep_moe.layer import DeepEPMoE, EPMoE
from sglang.srt.layers.moe.ep_moe.token_dispatcher import DeepEPDispatcher
from sglang.srt.layers.moe.fused_moe_triton import FusedMoE from sglang.srt.layers.moe.fused_moe_triton import FusedMoE
from sglang.srt.layers.moe.topk import select_experts from sglang.srt.layers.moe.topk import select_experts
from sglang.srt.layers.quantization.base_config import QuantizationConfig from sglang.srt.layers.quantization.base_config import QuantizationConfig
@@ -79,6 +78,8 @@ _is_cuda = is_cuda()
if _is_cuda: if _is_cuda:
from sgl_kernel import awq_dequantize, bmm_fp8 from sgl_kernel import awq_dequantize, bmm_fp8
from sglang.srt.layers.moe.ep_moe.token_dispatcher import DeepEPDispatcher
else: else:
from vllm import _custom_ops as ops from vllm import _custom_ops as ops