From 047935080dbc46c671a943a2f07c66991781bee0 Mon Sep 17 00:00:00 2001
From: ErsongWang <158176536+ErsongWang@users.noreply.github.com>
Date: Sun, 23 Nov 2025 06:25:52 +0800
Subject: [PATCH] Revert "Fix RMSNorm API CALL mismatch issue. (#10032)"
 (#13727)

---
 python/sglang/srt/layers/layernorm.py | 21 ++-------------------
 1 file changed, 2 insertions(+), 19 deletions(-)

diff --git a/python/sglang/srt/layers/layernorm.py b/python/sglang/srt/layers/layernorm.py
index 7569f2b97..fcbeb8ae1 100644
--- a/python/sglang/srt/layers/layernorm.py
+++ b/python/sglang/srt/layers/layernorm.py
@@ -19,7 +19,6 @@ from typing import Optional, Tuple, Union
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
-from packaging.version import Version
 
 from sglang.srt.batch_invariant_ops import (
     is_batch_invariant_mode_enabled,
@@ -70,11 +69,8 @@ if _use_aiter:
     from aiter import rmsnorm2d_fwd as rms_norm
     from aiter import rmsnorm2d_fwd_with_add as fused_add_rms_norm
 elif _is_hip:
-    import vllm
     from vllm._custom_ops import fused_add_rms_norm, rms_norm
 
-    _vllm_version = Version(vllm.__version__)
-
 logger = logging.getLogger(__name__)
 
 if _is_npu:
@@ -169,21 +165,8 @@ class RMSNorm(CustomOp):
             # NOTE: Remove this if aiter kernel supports discontinuous input
             x = x.contiguous()
         if residual is not None:
-            if _vllm_version < Version("0.9"):
-                fused_add_rms_norm(x, residual, self.weight.data, self.variance_epsilon)
-                return x, residual
-            else:
-                residual_out = torch.empty_like(x)
-                output = torch.empty_like(x)
-                fused_add_rms_norm(
-                    output,
-                    x,
-                    residual_out,
-                    residual,
-                    self.weight.data,
-                    self.variance_epsilon,
-                )
-                return output, residual_out
+            fused_add_rms_norm(x, residual, self.weight.data, self.variance_epsilon)
+            return x, residual
         out = torch.empty_like(x)
         rms_norm(out, x, self.weight.data, self.variance_epsilon)
         return out