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

[BugFix] add ut for fused_moe_degemm (#6840)

Browse Source 
* add ut

* add skip
This commit is contained in:
fxyfxy777
2026-03-16 12:22:18 +08:00
committed by GitHub
parent 091e3c815d
commit 4d39232553
2 changed files with 381 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/model_executor/layers/moe/fused_moe_deepgemm_backend.py
+2 -2
View File
@@ -497,7 +497,7 @@ class DeepGemmFusedMoeMethod(MoEMethodBase):
permute_scale,
m_indices,
) = paddle.nn.functional.moe_permute(
hidden_states=recv_x,
hidden_states=recv_x_value,
scale=recv_x_scale,
expert_routemap_topk=recv_topk_idx,
expert_prob_topk=recv_topk_weights,
@@ -581,7 +581,7 @@ class DeepGemmFusedMoeMethod(MoEMethodBase):
zipped_expertwise_rowmap=permute_indices_per_token,
expert_routemap_topk=recv_topk_idx,
token_prob_unzipped=dst_weights,
total_zipped_tokens=recv_x.shape[0],
total_zipped_tokens=recv_x_value.shape[0],
num_experts=layer.num_local_experts,
using_weighted_combine=True,
)