pin lora_b moe weights on cpu (#31317)

Signed-off-by: gnovack <gnovack@amazon.com>
This commit is contained in:
gnovack
2026-01-05 12:15:40 -08:00
committed by GitHub
parent fdcc5176be
commit c77a993cc2
2 changed files with 89 additions and 65 deletions
-11
View File
@@ -671,20 +671,9 @@ class FusedMoE3DWithLoRA(FusedMoEWithLoRA):
self.reset_lora(index)
self.adapter_enabled[index] = 1
num_experts = self.w13_lora_a_stacked[0].shape[1]
w13_lora_a, w2_lora_a = lora_a
w13_lora_b, w2_lora_b = lora_b
# (num_experts,rank,input_size)
w13_lora_a = w13_lora_a.reshape(num_experts, -1, w13_lora_a.shape[-1])
w2_lora_a = w2_lora_a.reshape(num_experts, -1, w2_lora_a.shape[-1])
# (output_size,rank,num_experts)
w13_lora_b = w13_lora_b.reshape(w13_lora_b.shape[0], -1, num_experts)
w2_lora_b = w2_lora_b.reshape(w2_lora_b.shape[0], -1, num_experts)
# (num_experts,output_size,rank)
w13_lora_b = w13_lora_b.permute(2, 0, 1)
w2_lora_b = w2_lora_b.permute(2, 0, 1)
sliced_w13_lora_a = self._slice_w13_a(w13_lora_a)
sliced_w13_lora_b = self._slice_w13_b(w13_lora_b)