From f5d17400303149bbb480f6abfb6f7bb646c1d895 Mon Sep 17 00:00:00 2001 From: Iryna Boiko Date: Sun, 18 Jan 2026 23:20:39 +0100 Subject: [PATCH] [Bugfix] Add OOT backend option (#32471) Signed-off-by: Iryna Boiko --- vllm/model_executor/layers/fused_moe/oracle/unquantized.py | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/vllm/model_executor/layers/fused_moe/oracle/unquantized.py b/vllm/model_executor/layers/fused_moe/oracle/unquantized.py index 2213461a7..138bfac28 100644 --- a/vllm/model_executor/layers/fused_moe/oracle/unquantized.py +++ b/vllm/model_executor/layers/fused_moe/oracle/unquantized.py @@ -33,6 +33,7 @@ class UnquantizedMoeBackend(Enum): CPU = "CPU" XPU = "XPU" TPU = "TPU" + OOT = "OOT" # NOTE(zyongye): Unsupported backend means backend @@ -42,6 +43,7 @@ UNSUPPORTED_BACKEND = [ UnquantizedMoeBackend.CPU, UnquantizedMoeBackend.XPU, UnquantizedMoeBackend.TPU, + UnquantizedMoeBackend.OOT, ] @@ -95,6 +97,8 @@ def select_unquantized_moe_backend( backend = UnquantizedMoeBackend.CPU if current_platform.is_tpu(): backend = UnquantizedMoeBackend.TPU + if current_platform.is_out_of_tree(): + backend = UnquantizedMoeBackend.OOT logger.info_once(_make_log_backend(backend), scope="local") return backend