apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-24 01:29:57 +08:00
Code Issues Actions 9 Packages Projects Releases Wiki Activity

[RL] support moe-topk use topk_reduce_func (#7218)

Browse Source 
* support moe-topk use topk_reduce_func

* fix ep error

* fix ut

* fix ut
This commit is contained in:
JYChen
2026-04-09 11:01:03 +08:00
committed by GitHub
parent 7005404ce3
commit 43ace7af25
7 changed files with 66 additions and 112 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/model_executor/layers/moe/ep.py
+2
View File
@@ -509,6 +509,7 @@ class EPRunner:
expert_in_rank_num_list=expert_in_rank_num_list,
tokens_per_expert_stats_list=tokens_per_expert_stats_list,
redundant_ep_rank_num_plus_one=layer.fd_config.eplb_config.redundant_experts_num + 1,
topk_reduce_func=getattr(layer, "topk_reduce_func", None),
)
else:
topk_idx, topk_weights = fastdeploy.model_executor.ops.gpu.moe_redundant_topk_select(
@@ -534,6 +535,7 @@ class EPRunner:
layer.routed_scaling_factor,
layer.gate_correction_bias,
getattr(layer, "renormalize", True),
topk_reduce_func=getattr(layer, "topk_reduce_func", None),
)
else:
topk_idx, topk_weights = fastdeploy.model_executor.ops.gpu.moe_topk_select(