zhangyiss/mlx-examples
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx-examples.git synced 2025-08-28 16:16:27 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

faster generation

Browse Source
This commit is contained in:
Goekdeniz-Guelmez 2025-03-04 21:26:21 +01:00
parent fd63c68280
commit 8b6beea3be
1 changed files with 2 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
llms/mlx_lm/models/olmoe.py
View File

@ -124,9 +124,8 @@ class OlmoeSparseMoeBlock(nn.Module):
final_hidden_states = mx.zeros_like(x)
for expert_idx in range(self.num_experts):
expert_weights = routing_weights[:, expert_idx:expert_idx+1]
if mx.max(expert_weights) > 1e-5:
expert_output = self.experts[expert_idx](x)
final_hidden_states += expert_output * expert_weights
expert_output = self.experts[expert_idx](x)
final_hidden_states += expert_output * expert_weights
return final_hidden_states.reshape(B, L, D)