fix(mxfp4): return is_monolithic=False when LoRA is enabled for Triton backend (#35382)

Signed-off-by: Seungho Yoon <yoonsnowdev@gmail.com>

fix(mxfp4): return is_monolithic=False when LoRA is enabled for Triton backend (#35382)
Signed-off-by: Seungho Yoon <yoonsnowdev@gmail.com>
5a435507 · Seungho Yoon · GitHub · 59d7af9c · 5a435507
Unverified Commit 5a435507 authored Mar 01, 2026 by Seungho Yoon Committed by GitHub Mar 01, 2026
Show whitespace changes
Inline Side-by-side

Showing with 2 additions and 0 deletions

vllm/model_executor/layers/quantization/mxfp4.py vllm/model_executor/layers/quantization/mxfp4.py +2 -0

No files found.
--- a/vllm/model_executor/layers/quantization/mxfp4.py
+++ b/vllm/model_executor/layers/quantization/mxfp4.py
@@ -1001,6 +1001,8 @@ class Mxfp4MoEMethod(FusedMoEMethodBase):

    @property
    def is_monolithic(self) -> bool:
+        if self.moe.is_lora_enabled:
+            return False
        return (
            self.mxfp4_backend == Mxfp4Backend.SM100_FI_MXFP4_MXFP8_TRTLLM
            or self.mxfp4_backend == Mxfp4Backend.SM100_FI_MXFP4_BF16