[Bugfix] Use FusedMoEWithLoRA for shared MoE LoRAs on 3D models (#60098)

Signed-off-by: ARAVINDHAN T <arvindhant01@gmail.com>
Co-authored-by: linitra24 <renshuang.zhou@daocloud.io>
This commit is contained in:
ARAVINDHAN T
2026-10-08 15:04:32 +08:00
committed by GitHub
co-authored by linitra24
parent 032d39e9b4
commit e58f8dde7f
2 changed files with 7 additions and 8 deletions
+4 -6
View File
@@ -144,20 +144,18 @@ class LoRAModelManager:
else set()
)
# When the engine is started with enable_mixed_moe_lora_format=True
# we force the universal 2D wrapper (FusedMoEWithLoRA) regardless of
# the model's 3D flag, so 2D and 3D adapters can coexist.
# Mixed-format and shared-outer adapters both use FusedMoEWithLoRA.
# FusedMoE3DWithLoRA only implements the fused gate/up PEFT pair.
self._enable_mixed_moe_lora_format = (
is_moe and lora_config.enable_mixed_moe_lora_format
)
self._enable_moe_shared_loras = is_moe and lora_config.enable_moe_shared_loras
self._is_3d_moe_model = (
self._is_moe
and self.model.is_3d_moe_weight
and not self._enable_mixed_moe_lora_format
and not self._enable_moe_shared_loras
)
# Shared MoE adapters: w13 lora_A / w2 lora_B shared across experts,
# stored as pre-stacked experts.w{1,2,3} tensors (startup opt-in).
self._enable_moe_shared_loras = is_moe and lora_config.enable_moe_shared_loras
self.packed_modules_mapping = process_packed_modules_mapping(
self.model,
force_2d_moe=self._enable_mixed_moe_lora_format,