[MoE Refactor] FusedMoE/MoERunner inversion refactor (#41184)

Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
bnellnm
2026-06-08 10:42:58 -04:00
committed by GitHub
parent 753e9d55e6
commit dc68bd8c41
90 changed files with 2738 additions and 2031 deletions
+3
View File
@@ -45,6 +45,9 @@ class LoRAMapping:
def _get_lora_device(base_layer: nn.Module) -> torch.device:
# code borrowed from https://github.com/fmmoret/vllm/blob/fm-support-lora-on-quantized-models/vllm/lora/layers.py#L34
"""Returns the device for where to place the LoRA tensors."""
if hasattr(base_layer, "routed_experts"):
base_layer = base_layer.routed_experts
# unquantizedLinear
if hasattr(base_layer, "weight"):
return base_layer.weight.device