diff --git a/python/sglang/srt/models/chatglm.py b/python/sglang/srt/models/chatglm.py
index d2ad02fbf..0a22f994b 100644
--- a/python/sglang/srt/models/chatglm.py
+++ b/python/sglang/srt/models/chatglm.py
@@ -24,8 +24,6 @@ from torch import nn
 from torch.nn import LayerNorm
 from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
-from vllm.model_executor.layers.activation import SiluAndMul
-from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (
     MergedColumnParallelLinear,
     QKVParallelLinear,
@@ -43,6 +41,8 @@ from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import SamplerOutput
 from vllm.transformers_utils.configs import ChatGLMConfig
 
+from sglang.srt.layers.activation import SiluAndMul
+from sglang.srt.layers.layernorm import RMSNorm
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
diff --git a/python/sglang/srt/models/commandr.py b/python/sglang/srt/models/commandr.py
index 1259285c4..f6d6f6e1f 100644
--- a/python/sglang/srt/models/commandr.py
+++ b/python/sglang/srt/models/commandr.py
@@ -50,7 +50,6 @@ from vllm.distributed import (
     get_tensor_model_parallel_rank,
     get_tensor_model_parallel_world_size,
 )
-from vllm.model_executor.layers.activation import SiluAndMul
 from vllm.model_executor.layers.linear import (
     MergedColumnParallelLinear,
     QKVParallelLinear,
@@ -62,6 +61,7 @@ from vllm.model_executor.layers.vocab_parallel_embedding import VocabParallelEmb
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 from vllm.model_executor.utils import set_weight_attrs
 
+from sglang.srt.layers.activation import SiluAndMul
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
diff --git a/python/sglang/srt/models/deepseek.py b/python/sglang/srt/models/deepseek.py
index 98dcfd28d..59fd1ec7e 100644
--- a/python/sglang/srt/models/deepseek.py
+++ b/python/sglang/srt/models/deepseek.py
@@ -27,9 +27,7 @@ from vllm.distributed import (
     get_tensor_model_parallel_world_size,
     tensor_model_parallel_all_reduce,
 )
-from vllm.model_executor.layers.activation import SiluAndMul
 from vllm.model_executor.layers.fused_moe import fused_moe
-from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (
     MergedColumnParallelLinear,
     QKVParallelLinear,
@@ -44,6 +42,8 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 )
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
+from sglang.srt.layers.activation import SiluAndMul
+from sglang.srt.layers.layernorm import RMSNorm
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
diff --git a/python/sglang/srt/models/deepseek_v2.py b/python/sglang/srt/models/deepseek_v2.py
index 739562730..2198428b8 100644
--- a/python/sglang/srt/models/deepseek_v2.py
+++ b/python/sglang/srt/models/deepseek_v2.py
@@ -26,9 +26,7 @@ from vllm.distributed import (
     get_tensor_model_parallel_world_size,
     tensor_model_parallel_all_reduce,
 )
-from vllm.model_executor.layers.activation import SiluAndMul
 from vllm.model_executor.layers.fused_moe import FusedMoE
-from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (
     ColumnParallelLinear,
     MergedColumnParallelLinear,
@@ -43,6 +41,8 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 )
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
+from sglang.srt.layers.activation import SiluAndMul
+from sglang.srt.layers.layernorm import RMSNorm
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.managers.schedule_batch import global_server_args_dict
diff --git a/python/sglang/srt/models/gemma.py b/python/sglang/srt/models/gemma.py
index ce3973115..990937f51 100644
--- a/python/sglang/srt/models/gemma.py
+++ b/python/sglang/srt/models/gemma.py
@@ -24,7 +24,6 @@ from transformers import PretrainedConfig
 from vllm.config import CacheConfig, LoRAConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.activation import GeluAndMul
-from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (
     MergedColumnParallelLinear,
     QKVParallelLinear,
@@ -35,6 +34,7 @@ from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import VocabParallelEmbedding
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
+from sglang.srt.layers.layernorm import RMSNorm
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
diff --git a/python/sglang/srt/models/grok.py b/python/sglang/srt/models/grok.py
index 38297b7d6..13d4330d4 100644
--- a/python/sglang/srt/models/grok.py
+++ b/python/sglang/srt/models/grok.py
@@ -31,7 +31,6 @@ from vllm.distributed import (
     get_tensor_model_parallel_world_size,
     tensor_model_parallel_all_reduce,
 )
-from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (
     QKVParallelLinear,
     ReplicatedLinear,
@@ -50,6 +49,7 @@ from vllm.model_executor.utils import set_weight_attrs
 from vllm.utils import print_warning_once
 
 from sglang.srt.layers.fused_moe import fused_moe
+from sglang.srt.layers.layernorm import RMSNorm
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
diff --git a/python/sglang/srt/models/minicpm.py b/python/sglang/srt/models/minicpm.py
index bf572855e..49ff1926f 100644
--- a/python/sglang/srt/models/minicpm.py
+++ b/python/sglang/srt/models/minicpm.py
@@ -22,8 +22,6 @@ import torch
 from torch import nn
 from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
-from vllm.model_executor.layers.activation import SiluAndMul
-from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (
     MergedColumnParallelLinear,
     QKVParallelLinear,
@@ -37,6 +35,8 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 )
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
+from sglang.srt.layers.activation import SiluAndMul
+from sglang.srt.layers.layernorm import RMSNorm
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
diff --git a/python/sglang/srt/models/mixtral.py b/python/sglang/srt/models/mixtral.py
index 63053ac50..876c7a09d 100644
--- a/python/sglang/srt/models/mixtral.py
+++ b/python/sglang/srt/models/mixtral.py
@@ -31,7 +31,6 @@ from vllm.distributed import (
     tensor_model_parallel_all_reduce,
 )
 from vllm.model_executor.layers.fused_moe import fused_moe
-from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (
     QKVParallelLinear,
     ReplicatedLinear,
@@ -48,6 +47,7 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 from vllm.model_executor.utils import set_weight_attrs
 from vllm.utils import print_warning_once
 
+from sglang.srt.layers.layernorm import RMSNorm
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
diff --git a/python/sglang/srt/models/mixtral_quant.py b/python/sglang/srt/models/mixtral_quant.py
index 07caf3833..115fce1d6 100644
--- a/python/sglang/srt/models/mixtral_quant.py
+++ b/python/sglang/srt/models/mixtral_quant.py
@@ -29,7 +29,6 @@ from vllm.distributed import (
     get_tensor_model_parallel_world_size,
     tensor_model_parallel_all_reduce,
 )
-from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (
     QKVParallelLinear,
     ReplicatedLinear,
@@ -43,6 +42,7 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 )
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
+from sglang.srt.layers.layernorm import RMSNorm
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
diff --git a/python/sglang/srt/models/qwen.py b/python/sglang/srt/models/qwen.py
index ffc512b1c..93dae9585 100644
--- a/python/sglang/srt/models/qwen.py
+++ b/python/sglang/srt/models/qwen.py
@@ -22,8 +22,6 @@ from torch import nn
 from transformers import PretrainedConfig
 from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
-from vllm.model_executor.layers.activation import SiluAndMul
-from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (
     MergedColumnParallelLinear,
     QKVParallelLinear,
@@ -37,6 +35,8 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 )
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
+from sglang.srt.layers.activation import SiluAndMul
+from sglang.srt.layers.layernorm import RMSNorm
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
diff --git a/python/sglang/srt/models/qwen2.py b/python/sglang/srt/models/qwen2.py
index dec962bf0..d1295bd8c 100644
--- a/python/sglang/srt/models/qwen2.py
+++ b/python/sglang/srt/models/qwen2.py
@@ -22,8 +22,6 @@ import torch
 from torch import nn
 from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
-from vllm.model_executor.layers.activation import SiluAndMul
-from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (
     MergedColumnParallelLinear,
     QKVParallelLinear,
@@ -37,6 +35,8 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 )
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
+from sglang.srt.layers.activation import SiluAndMul
+from sglang.srt.layers.layernorm import RMSNorm
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
diff --git a/python/sglang/srt/models/qwen2_moe.py b/python/sglang/srt/models/qwen2_moe.py
index f96f7e0e4..9bdbd7506 100644
--- a/python/sglang/srt/models/qwen2_moe.py
+++ b/python/sglang/srt/models/qwen2_moe.py
@@ -28,9 +28,7 @@ from vllm.distributed import (
     get_tensor_model_parallel_world_size,
     tensor_model_parallel_all_reduce,
 )
-from vllm.model_executor.layers.activation import SiluAndMul
 from vllm.model_executor.layers.fused_moe import FusedMoE
-from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (
     MergedColumnParallelLinear,
     QKVParallelLinear,
@@ -47,6 +45,8 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 )
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
+from sglang.srt.layers.activation import SiluAndMul
+from sglang.srt.layers.layernorm import RMSNorm
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata
diff --git a/python/sglang/srt/models/stablelm.py b/python/sglang/srt/models/stablelm.py
index aeaa46ab1..9e10f12f2 100644
--- a/python/sglang/srt/models/stablelm.py
+++ b/python/sglang/srt/models/stablelm.py
@@ -24,7 +24,6 @@ from torch import nn
 from transformers import PretrainedConfig
 from vllm.config import CacheConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
-from vllm.model_executor.layers.activation import SiluAndMul
 from vllm.model_executor.layers.linear import (
     MergedColumnParallelLinear,
     QKVParallelLinear,
@@ -38,6 +37,7 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
 )
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
+from sglang.srt.layers.activation import SiluAndMul
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
 from sglang.srt.model_executor.forward_batch_info import InputMetadata