diff --git a/server/text_generation_server/layers/moe/__init__.py b/server/text_generation_server/layers/moe/__init__.py index abf7ba6b..e148f165 100644 --- a/server/text_generation_server/layers/moe/__init__.py +++ b/server/text_generation_server/layers/moe/__init__.py @@ -28,7 +28,7 @@ from text_generation_server.utils.weights import ( if SYSTEM == "ipex": from .fused_moe_ipex import fused_topk, grouped_topk -if SYSTEM == "cuda": +elif SYSTEM == "cuda": moe_kernels = load_kernel(module="moe", repo_id="kernels-community/moe") fused_topk = moe_kernels.fused_topk grouped_topk = moe_kernels.grouped_topk