[Bugfix] Disable RoutingMethodType.[Renormalize,RenormalizeNaive] TRTLLM per-tensor FP8 MoE (#33620)

Signed-off-by: mgoin <mgoin64@gmail.com>
2026-02-03 05:37:15 -05:00
parent 83449a5ff0
commit e346e2d056
1 changed files with 4 additions and 2 deletions
--- a/vllm/model_executor/layers/fused_moe/flashinfer_trtllm_moe.py
+++ b/vllm/model_executor/layers/fused_moe/flashinfer_trtllm_moe.py
@@ -72,8 +72,10 @@ def _supports_routing_method(
        # NOTE(dbari): as above, potentially allow others here.
        return routing_method in [
            RoutingMethodType.Llama4,
-            RoutingMethodType.Renormalize,
-            RoutingMethodType.RenormalizeNaive,
+            # NOTE(mgoin): Disabled to investigate accuracy issues.
+            # See https://github.com/vllm-project/vllm/issues/33532
+            # RoutingMethodType.Renormalize,
+            # RoutingMethodType.RenormalizeNaive,
        ]
    else:
        raise ValueError("Unsupported quantization scheme.")