[Bugfix][NPU] Fix/Refactor routed scaling factor application in MoE routing (#31449)

2026-07-18 10:59:13 +08:00
parent 67e7f8d13a
commit 359009fa00
3 changed files with 4 additions and 1 deletions
@@ -29,6 +29,7 @@ class TestLLaDA2Mini(GSM8KAscendMixin, CustomTestCase):
        "ascend",
        "--dllm-algorithm",
        "LowConfidence",  # TODO: Add dLLM configurations
+        "--no-dllm-fdfo",  # FDFO (PR #27551) halves single-batch speed on NPU; use sync mode
    ]
    env = {
        **os.environ,