apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 16:48:03 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[GCU] Support gcu platform (#2702)

Browse Source 
baseline: e7fa57ebae

Co-authored-by: yongqiangma <xing.wo@163.com>
This commit is contained in:
EnflameGCU
2025-07-08 13:00:52 +08:00
committed by GitHub
parent 26d5d737dd
commit d0f4d6ba3a
33 changed files with 2988 additions and 85 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
fastdeploy/model_executor/layers/moe/moe.py
View File

@@ -20,6 +20,7 @@ from paddleformers.utils.log import logger
from fastdeploy import envs
from fastdeploy.model_executor.layers.utils import get_tensor
from fastdeploy.platforms import current_platform
class FusedMoE(nn.Layer):
@@ -95,8 +96,13 @@ class FusedMoE(nn.Layer):
self.moe_quant_type = moe_quant_config.name()
else:
# now, no quant method(w_fp16 a_fp16) can't get from quant_config, we will optimize it in future
from .fused_moe_cutlass_backend import CutlassMoEMethod
self.quant_method = CutlassMoEMethod(None)
if current_platform.is_cuda():
from .fused_moe_cutlass_backend import CutlassMoEMethod
self.quant_method = CutlassMoEMethod(None)
elif current_platform.is_gcu():
from fastdeploy.model_executor.layers.backends import \
GCUFusedMoeMethod
self.quant_method = GCUFusedMoeMethod(None)
if self.ep_size > 1:
self.quant_method.init_ep(self)