apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 00:17:25 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity

[Intel HPU] enable MoE EP for hpu (#5855)

Browse Source 
* enable HPU MoE EP

* MoE intermediate_scale stack

* enable loader_v1 esp for tensor_wise_fp8 TP or EP

* modify activation_scale name
This commit is contained in:
Cheng Yanfei
2026-01-15 13:08:00 +08:00
committed by GitHub
parent 7c56041272
commit fbcccaa750
9 changed files with 177 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/model_executor/layers/moe/moe.py
+7 -4
View File
@@ -659,6 +659,12 @@ class FusedMoE(nn.Layer):
tp_group=self.fd_config.parallel_config.tp_group,
)
if current_platform.is_intel_hpu():
out = self.forward_normal(x, gate, forward_meta, topk_ids_hookfunc=topk_ids_hookfunc)
if self.reduce_results and (self.ep_size > 1 or self.tp_size > 1):
tensor_model_parallel_all_reduce_custom(out)
return out
token_num = x.shape[0]
if (
self.ep_size > 1
@@ -678,10 +684,7 @@ class FusedMoE(nn.Layer):
out = self.forward_normal(x, gate, forward_meta, topk_ids_hookfunc=topk_ids_hookfunc)
if self.reduce_results and self.tp_size > 1:
if current_platform.is_intel_hpu():
tensor_model_parallel_all_reduce_custom(out)
else:
out = tensor_model_parallel_all_reduce(out, self.tp_group)
out = tensor_model_parallel_all_reduce(out, self.tp_group)
return out
def forward_chunked_moe(