[MoE Refactor] Introduce MoERunner abstraction and move execution logic from FusedMoE to DefaultMoERunner (#32344)

Signed-off-by: Bill Nell <bnell@redhat.com>
2026-02-10 19:51:07 -05:00
parent dc6de33c3d
commit d1481ba783
25 changed files with 913 additions and 753 deletions
--- a/tests/kernels/moe/utils.py
+++ b/tests/kernels/moe/utils.py
@@ -52,6 +52,7 @@ def make_dummy_moe_config(
        hidden_dim=hidden_dim,
        intermediate_size_per_partition=intermediate_size_per_partition,
        num_local_experts=num_experts,
+        num_logical_experts=num_experts,
        moe_parallel_config=FusedMoEParallelConfig.make_no_parallel(),
        activation="silu",
        in_dtype=in_dtype,