diff --git a/src/transformers/integrations/hub_kernels.py b/src/transformers/integrations/hub_kernels.py
index b58adc35c063..0b89aafb2457 100644
--- a/src/transformers/integrations/hub_kernels.py
+++ b/src/transformers/integrations/hub_kernels.py
@@ -98,9 +98,9 @@
     )
     from kernels import use_kernelized_func as _kernels_use_kernelized_func
 
-    def use_kernel_forward_from_hub(layer_name: str):
+    def use_kernel_forward_from_hub(layer_name: str, condition: Callable[["nn.Module"], bool] | None = None):
         if _kernels_enabled:
-            return _kernels_use_kernel_forward_from_hub(layer_name)
+            return _kernels_use_kernel_forward_from_hub(layer_name, condition)
         else:
             logger.warning_once(
                 f"kernels hub usage is disabled through the environment USE_HUB_KERNELS={_TRANSFORMERS_USE_HUB_KERNELS}"
diff --git a/src/transformers/models/afmoe/modeling_afmoe.py b/src/transformers/models/afmoe/modeling_afmoe.py
index bf79f260355b..fc8fc35db66c 100644
--- a/src/transformers/models/afmoe/modeling_afmoe.py
+++ b/src/transformers/models/afmoe/modeling_afmoe.py
@@ -119,6 +119,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class AfmoeMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/aimv2/modeling_aimv2.py b/src/transformers/models/aimv2/modeling_aimv2.py
index ec1e30d493f7..8557e3611650 100644
--- a/src/transformers/models/aimv2/modeling_aimv2.py
+++ b/src/transformers/models/aimv2/modeling_aimv2.py
@@ -97,6 +97,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Aimv2MLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/aria/modeling_aria.py b/src/transformers/models/aria/modeling_aria.py
index 59f307883293..6a29680de498 100644
--- a/src/transformers/models/aria/modeling_aria.py
+++ b/src/transformers/models/aria/modeling_aria.py
@@ -215,6 +215,7 @@ def forward(self, key_value_states: torch.Tensor, attn_mask: torch.Tensor | None
         return out
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class AriaSharedExpertsMLP(nn.Module):
     """
     Shared Expert MLP for shared experts.
diff --git a/src/transformers/models/axk1/modeling_axk1.py b/src/transformers/models/axk1/modeling_axk1.py
index dd65c526e998..14075e5cd4b1 100644
--- a/src/transformers/models/axk1/modeling_axk1.py
+++ b/src/transformers/models/axk1/modeling_axk1.py
@@ -126,6 +126,7 @@ def forward(self, x, position_ids):
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class AXK1MLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/axk2/modeling_axk2.py b/src/transformers/models/axk2/modeling_axk2.py
index a9e5238eafa9..9b8ebed35bab 100644
--- a/src/transformers/models/axk2/modeling_axk2.py
+++ b/src/transformers/models/axk2/modeling_axk2.py
@@ -381,6 +381,7 @@ def forward(
         return final_hidden_states
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class AXK2MLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/bamba/modeling_bamba.py b/src/transformers/models/bamba/modeling_bamba.py
index 2a06e67cd53e..0f9a180100ac 100644
--- a/src/transformers/models/bamba/modeling_bamba.py
+++ b/src/transformers/models/bamba/modeling_bamba.py
@@ -795,6 +795,7 @@ def forward(
         return contextualized_states
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class BambaMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/bitnet/modular_bitnet.py b/src/transformers/models/bitnet/modular_bitnet.py
index 7f3a248766b2..dfaa5a16227f 100644
--- a/src/transformers/models/bitnet/modular_bitnet.py
+++ b/src/transformers/models/bitnet/modular_bitnet.py
@@ -22,6 +22,7 @@
 from ...modeling_utils import ALL_ATTENTION_FUNCTIONS
 from ...processing_utils import Unpack
 from ...utils import logging
+from ...utils.generic import no_inherit_decorator
 from ..gemma.modeling_gemma import GemmaMLP
 from ..llama.modeling_llama import (
     LlamaAttention,
@@ -42,6 +43,7 @@ class BitNetRMSNorm(LlamaRMSNorm):
     pass
 
 
+@no_inherit_decorator
 class BitNetMLP(GemmaMLP):
     def __init__(self, config: BitNetConfig):
         super().__init__(config)
diff --git a/src/transformers/models/cohere/modeling_cohere.py b/src/transformers/models/cohere/modeling_cohere.py
index c2850690f0b9..1bfe3d55ce2b 100644
--- a/src/transformers/models/cohere/modeling_cohere.py
+++ b/src/transformers/models/cohere/modeling_cohere.py
@@ -34,6 +34,7 @@
 from ...activations import ACT2FN
 from ...cache_utils import Cache, DynamicCache
 from ...generation import GenerationMixin
+from ...integrations import use_kernel_forward_from_hub
 from ...masking_utils import create_causal_mask
 from ...modeling_flash_attention_utils import FlashAttentionKwargs
 from ...modeling_layers import GradientCheckpointingLayer
@@ -119,6 +120,7 @@ def forward(self, x, position_ids):
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class CohereMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/cohere2/modeling_cohere2.py b/src/transformers/models/cohere2/modeling_cohere2.py
index eed7e377a706..00e572e13bad 100644
--- a/src/transformers/models/cohere2/modeling_cohere2.py
+++ b/src/transformers/models/cohere2/modeling_cohere2.py
@@ -26,6 +26,7 @@
 from ...activations import ACT2FN
 from ...cache_utils import Cache, DynamicCache
 from ...generation import GenerationMixin
+from ...integrations import use_kernel_forward_from_hub
 from ...masking_utils import create_causal_mask, create_sliding_window_causal_mask
 from ...modeling_layers import GradientCheckpointingLayer
 from ...modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
@@ -253,6 +254,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Cohere2MLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/cohere2_moe/modeling_cohere2_moe.py b/src/transformers/models/cohere2_moe/modeling_cohere2_moe.py
index e7175efe436f..f66ead99d97a 100644
--- a/src/transformers/models/cohere2_moe/modeling_cohere2_moe.py
+++ b/src/transformers/models/cohere2_moe/modeling_cohere2_moe.py
@@ -82,6 +82,7 @@ def forward(self, hidden_states):
         return hidden_states.to(input_dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Cohere2MoeMLP(nn.Module):
     def __init__(self, config: Cohere2MoeConfig, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/cohere_compass/modeling_cohere_compass.py b/src/transformers/models/cohere_compass/modeling_cohere_compass.py
index 0f335150ae2a..035eaabdbd71 100644
--- a/src/transformers/models/cohere_compass/modeling_cohere_compass.py
+++ b/src/transformers/models/cohere_compass/modeling_cohere_compass.py
@@ -160,6 +160,7 @@ def apply_interleaved_mrope(self, freqs, mrope_section):
         return freqs_t
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class CohereCompassMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/csm/modeling_csm.py b/src/transformers/models/csm/modeling_csm.py
index 3b61827a7c0b..8f84fd81d525 100644
--- a/src/transformers/models/csm/modeling_csm.py
+++ b/src/transformers/models/csm/modeling_csm.py
@@ -174,6 +174,7 @@ def forward(self, x, position_ids):
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class CsmMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/cwm/modeling_cwm.py b/src/transformers/models/cwm/modeling_cwm.py
index 546d9a14c6ce..f67d1fbe17ec 100644
--- a/src/transformers/models/cwm/modeling_cwm.py
+++ b/src/transformers/models/cwm/modeling_cwm.py
@@ -249,6 +249,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class CwmMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/deepseek_ocr2/modeling_deepseek_ocr2.py b/src/transformers/models/deepseek_ocr2/modeling_deepseek_ocr2.py
index 90676efdd97c..0d17be2f5c44 100644
--- a/src/transformers/models/deepseek_ocr2/modeling_deepseek_ocr2.py
+++ b/src/transformers/models/deepseek_ocr2/modeling_deepseek_ocr2.py
@@ -1118,6 +1118,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class DeepseekOcr2TextMLP(nn.Module):
     def __init__(self, config: DeepseekOcr2TextConfig, hidden_size=None, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/deepseek_ocr2/modular_deepseek_ocr2.py b/src/transformers/models/deepseek_ocr2/modular_deepseek_ocr2.py
index 989034e8431b..6b2e952a43f2 100644
--- a/src/transformers/models/deepseek_ocr2/modular_deepseek_ocr2.py
+++ b/src/transformers/models/deepseek_ocr2/modular_deepseek_ocr2.py
@@ -40,7 +40,7 @@
 from ...modeling_utils import PreTrainedModel
 from ...processing_utils import Unpack
 from ...utils import TensorType, TransformersKwargs, auto_docstring, can_return_tuple, logging
-from ...utils.generic import merge_with_config_defaults
+from ...utils.generic import merge_with_config_defaults, no_inherit_decorator
 from ...utils.import_utils import requires
 from ...utils.output_capturing import capture_outputs
 from ..deepseek_v2.configuration_deepseek_v2 import DeepseekV2Config
@@ -783,6 +783,7 @@ def forward(self, pixel_values: torch.FloatTensor, **kwargs) -> BaseModelOutput:
         return BaseModelOutput(last_hidden_state=hidden_states)
 
 
+@no_inherit_decorator
 class DeepseekOcr2VisionMLP(Qwen2MLP):
     pass
 
diff --git a/src/transformers/models/deepseek_v2/modeling_deepseek_v2.py b/src/transformers/models/deepseek_v2/modeling_deepseek_v2.py
index 32a5c15fd72a..6551549bb915 100644
--- a/src/transformers/models/deepseek_v2/modeling_deepseek_v2.py
+++ b/src/transformers/models/deepseek_v2/modeling_deepseek_v2.py
@@ -137,6 +137,7 @@ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
         return hidden_states
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class DeepseekV2MLP(nn.Module):
     def __init__(self, config: DeepseekV2Config, hidden_size=None, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/deepseek_v3/modeling_deepseek_v3.py b/src/transformers/models/deepseek_v3/modeling_deepseek_v3.py
index 76bb3e7a16a3..338db92101f0 100644
--- a/src/transformers/models/deepseek_v3/modeling_deepseek_v3.py
+++ b/src/transformers/models/deepseek_v3/modeling_deepseek_v3.py
@@ -112,6 +112,7 @@ def forward(self, x, position_ids):
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class DeepseekV3MLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/deepseek_v32/modeling_deepseek_v32.py b/src/transformers/models/deepseek_v32/modeling_deepseek_v32.py
index 8bf0f009b1a5..71f029483544 100644
--- a/src/transformers/models/deepseek_v32/modeling_deepseek_v32.py
+++ b/src/transformers/models/deepseek_v32/modeling_deepseek_v32.py
@@ -503,6 +503,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class DeepseekV32MLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/deepseek_v4/modeling_deepseek_v4.py b/src/transformers/models/deepseek_v4/modeling_deepseek_v4.py
index a399e426feea..d500fe2158a9 100644
--- a/src/transformers/models/deepseek_v4/modeling_deepseek_v4.py
+++ b/src/transformers/models/deepseek_v4/modeling_deepseek_v4.py
@@ -962,6 +962,7 @@ def forward(self, x: torch.Tensor) -> torch.Tensor:
         return (pre.unsqueeze(-1) * x).sum(dim=2).to(x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class DeepseekV4MLP(nn.Module):
     def __init__(self, config: DeepseekV4Config):
         super().__init__()
diff --git a/src/transformers/models/deimv2/modeling_deimv2.py b/src/transformers/models/deimv2/modeling_deimv2.py
index 44905ad1ed37..57676d5de5c5 100644
--- a/src/transformers/models/deimv2/modeling_deimv2.py
+++ b/src/transformers/models/deimv2/modeling_deimv2.py
@@ -177,6 +177,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Deimv2SwiGLUFFN(nn.Module):
     def __init__(self, config: Deimv2Config):
         super().__init__()
diff --git a/src/transformers/models/diffllama/modeling_diffllama.py b/src/transformers/models/diffllama/modeling_diffllama.py
index 2f2f85b62052..4f9b5f4c658e 100644
--- a/src/transformers/models/diffllama/modeling_diffllama.py
+++ b/src/transformers/models/diffllama/modeling_diffllama.py
@@ -49,6 +49,7 @@
 from .configuration_diffllama import DiffLlamaConfig
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class DiffLlamaMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/diffusion_gemma/configuration_diffusion_gemma.py b/src/transformers/models/diffusion_gemma/configuration_diffusion_gemma.py
index cc096cea1ca3..b71cc6876abb 100644
--- a/src/transformers/models/diffusion_gemma/configuration_diffusion_gemma.py
+++ b/src/transformers/models/diffusion_gemma/configuration_diffusion_gemma.py
@@ -75,6 +75,7 @@ class DiffusionGemmaTextConfig(PreTrainedConfig):
         "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
         "norm": (["hidden_states"], ["hidden_states"]),
     }
+    attribute_map = {"hidden_act": "hidden_activation"}
 
     vocab_size: int = 262_144
     hidden_size: int = 2304
diff --git a/src/transformers/models/diffusion_gemma/modeling_diffusion_gemma.py b/src/transformers/models/diffusion_gemma/modeling_diffusion_gemma.py
index dedcf39a4c05..e8d8d477d8c1 100644
--- a/src/transformers/models/diffusion_gemma/modeling_diffusion_gemma.py
+++ b/src/transformers/models/diffusion_gemma/modeling_diffusion_gemma.py
@@ -28,7 +28,7 @@
 from ...activations import ACT2FN
 from ...cache_utils import Cache, DynamicCache
 from ...configuration_utils import PreTrainedConfig
-from ...integrations import use_experts_implementation
+from ...integrations import use_experts_implementation, use_kernel_forward_from_hub
 from ...masking_utils import (
     ALL_MASK_ATTENTION_FUNCTIONS,
     bidirectional_mask_function,
@@ -497,6 +497,7 @@ def append_to_cache(
         return keys, values
 
 
+@use_kernel_forward_from_hub("GeGLUMLP", condition=lambda module: module.config.hidden_act == "gelu_pytorch_tanh")
 class DiffusionGemmaText4MLP(nn.Module):
     def __init__(self, config: DiffusionGemmaTextConfig, layer_idx: int):
         super().__init__()
@@ -506,7 +507,7 @@ def __init__(self, config: DiffusionGemmaTextConfig, layer_idx: int):
         self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
-        self.act_fn = ACT2FN[config.hidden_activation]
+        self.act_fn = ACT2FN[config.hidden_act]
 
     def forward(self, x):
         down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
diff --git a/src/transformers/models/diffusion_gemma/modular_diffusion_gemma.py b/src/transformers/models/diffusion_gemma/modular_diffusion_gemma.py
index b0b864d5c93f..00d8794a84a9 100644
--- a/src/transformers/models/diffusion_gemma/modular_diffusion_gemma.py
+++ b/src/transformers/models/diffusion_gemma/modular_diffusion_gemma.py
@@ -464,7 +464,7 @@ def __init__(self, config: DiffusionGemmaTextConfig, layer_idx: int):
         self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
-        self.act_fn = ACT2FN[config.hidden_activation]
+        self.act_fn = ACT2FN[config.hidden_act]
 
 
 class DiffusionGemmaTextRouter(Gemma4TextRouter):
diff --git a/src/transformers/models/dinov3_vit/modular_dinov3_vit.py b/src/transformers/models/dinov3_vit/modular_dinov3_vit.py
index 1af9b58d2d2e..73df746eb4af 100644
--- a/src/transformers/models/dinov3_vit/modular_dinov3_vit.py
+++ b/src/transformers/models/dinov3_vit/modular_dinov3_vit.py
@@ -38,7 +38,7 @@
 from ...processing_utils import Unpack
 from ...pytorch_utils import compile_compatible_method_lru_cache
 from ...utils import TransformersKwargs, auto_docstring, logging
-from ...utils.generic import can_return_tuple, maybe_autocast, merge_with_config_defaults
+from ...utils.generic import can_return_tuple, maybe_autocast, merge_with_config_defaults, no_inherit_decorator
 from ...utils.output_capturing import capture_outputs
 from ..swin.modeling_swin import SwinDropPath
 from .configuration_dinov3_vit import DINOv3ViTConfig
@@ -300,6 +300,7 @@ class DINOv3ViTMLP(ArceeMLP):
     pass
 
 
+@no_inherit_decorator
 class DINOv3ViTGatedMLP(LlamaMLP):
     pass
 
diff --git a/src/transformers/models/doge/modeling_doge.py b/src/transformers/models/doge/modeling_doge.py
index 6fd46ac81f9f..58816798f072 100644
--- a/src/transformers/models/doge/modeling_doge.py
+++ b/src/transformers/models/doge/modeling_doge.py
@@ -371,6 +371,7 @@ def prepare_dynamic_mask(
         return attn_mask
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class DogeMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/dots1/modeling_dots1.py b/src/transformers/models/dots1/modeling_dots1.py
index 04fc35253bbc..6ad4d83368bc 100644
--- a/src/transformers/models/dots1/modeling_dots1.py
+++ b/src/transformers/models/dots1/modeling_dots1.py
@@ -260,6 +260,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Dots1MLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/emu3/modeling_emu3.py b/src/transformers/models/emu3/modeling_emu3.py
index f3fd4ce0f9f0..bd6b37117a2e 100644
--- a/src/transformers/models/emu3/modeling_emu3.py
+++ b/src/transformers/models/emu3/modeling_emu3.py
@@ -216,6 +216,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Emu3MLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/ernie4_5/modeling_ernie4_5.py b/src/transformers/models/ernie4_5/modeling_ernie4_5.py
index 86798b16a5c8..e4f381a68be9 100644
--- a/src/transformers/models/ernie4_5/modeling_ernie4_5.py
+++ b/src/transformers/models/ernie4_5/modeling_ernie4_5.py
@@ -95,6 +95,7 @@ def forward(self, x, position_ids):
         return cos, sin
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Ernie4_5MLP(nn.Module):
     def __init__(self, config: Ernie4_5Config):
         super().__init__()
diff --git a/src/transformers/models/ernie4_5_moe/modeling_ernie4_5_moe.py b/src/transformers/models/ernie4_5_moe/modeling_ernie4_5_moe.py
index a347f9a7358f..91776332136b 100644
--- a/src/transformers/models/ernie4_5_moe/modeling_ernie4_5_moe.py
+++ b/src/transformers/models/ernie4_5_moe/modeling_ernie4_5_moe.py
@@ -63,6 +63,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Ernie4_5_MoeMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/ernie4_5_vl_moe/modeling_ernie4_5_vl_moe.py b/src/transformers/models/ernie4_5_vl_moe/modeling_ernie4_5_vl_moe.py
index c9decaad4383..0835bcbe7f7a 100644
--- a/src/transformers/models/ernie4_5_vl_moe/modeling_ernie4_5_vl_moe.py
+++ b/src/transformers/models/ernie4_5_vl_moe/modeling_ernie4_5_vl_moe.py
@@ -290,6 +290,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Ernie4_5_VLMoeMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/esmc/modeling_esmc.py b/src/transformers/models/esmc/modeling_esmc.py
index 04d8eb0c5443..5ed3be518c0f 100644
--- a/src/transformers/models/esmc/modeling_esmc.py
+++ b/src/transformers/models/esmc/modeling_esmc.py
@@ -110,6 +110,7 @@ def forward(self, x, position_ids):
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class EsmcMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/eurobert/modeling_eurobert.py b/src/transformers/models/eurobert/modeling_eurobert.py
index 931c9e4366ef..700aa4d14a40 100644
--- a/src/transformers/models/eurobert/modeling_eurobert.py
+++ b/src/transformers/models/eurobert/modeling_eurobert.py
@@ -200,6 +200,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class EuroBertMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/evolla/modeling_evolla.py b/src/transformers/models/evolla/modeling_evolla.py
index 45ad92b89f57..e732d299b18c 100644
--- a/src/transformers/models/evolla/modeling_evolla.py
+++ b/src/transformers/models/evolla/modeling_evolla.py
@@ -1083,6 +1083,7 @@ def forward(self, x, position_ids):
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class EvollaMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/exaone4/modeling_exaone4.py b/src/transformers/models/exaone4/modeling_exaone4.py
index 6a35803a94b2..a6e778933de5 100644
--- a/src/transformers/models/exaone4/modeling_exaone4.py
+++ b/src/transformers/models/exaone4/modeling_exaone4.py
@@ -265,6 +265,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Exaone4MLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/exaone_moe/modeling_exaone_moe.py b/src/transformers/models/exaone_moe/modeling_exaone_moe.py
index 5033eae0cdd3..70847e7f5e76 100644
--- a/src/transformers/models/exaone_moe/modeling_exaone_moe.py
+++ b/src/transformers/models/exaone_moe/modeling_exaone_moe.py
@@ -208,6 +208,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class ExaoneMoeMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/falcon_h1/modeling_falcon_h1.py b/src/transformers/models/falcon_h1/modeling_falcon_h1.py
index 9edbd2b59b32..bb7d19892f12 100644
--- a/src/transformers/models/falcon_h1/modeling_falcon_h1.py
+++ b/src/transformers/models/falcon_h1/modeling_falcon_h1.py
@@ -822,6 +822,7 @@ def forward(
         return contextualized_states
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class FalconH1MLP(nn.Module):
     def __init__(self, config: FalconH1Config):
         super().__init__()
diff --git a/src/transformers/models/flex_olmo/modeling_flex_olmo.py b/src/transformers/models/flex_olmo/modeling_flex_olmo.py
index 2654532668b9..8ac5b24fdb68 100644
--- a/src/transformers/models/flex_olmo/modeling_flex_olmo.py
+++ b/src/transformers/models/flex_olmo/modeling_flex_olmo.py
@@ -117,6 +117,7 @@ def forward(self, x, position_ids):
         return cos, sin
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class FlexOlmoMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/gemma/modeling_gemma.py b/src/transformers/models/gemma/modeling_gemma.py
index 91bbfcd58476..520c232e9be0 100644
--- a/src/transformers/models/gemma/modeling_gemma.py
+++ b/src/transformers/models/gemma/modeling_gemma.py
@@ -81,6 +81,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.eps}"
 
 
+@use_kernel_forward_from_hub("GeGLUMLP", condition=lambda module: module.config.hidden_act == "gelu_pytorch_tanh")
 class GemmaMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/gemma/modular_gemma.py b/src/transformers/models/gemma/modular_gemma.py
index b47fca63a2d1..879663989f74 100644
--- a/src/transformers/models/gemma/modular_gemma.py
+++ b/src/transformers/models/gemma/modular_gemma.py
@@ -21,6 +21,7 @@
 from ... import initialization as init
 from ...cache_utils import Cache, DynamicCache
 from ...configuration_utils import PreTrainedConfig
+from ...integrations import use_kernel_forward_from_hub
 from ...masking_utils import create_causal_mask
 from ...modeling_outputs import BaseModelOutputWithPast
 from ...modeling_rope_utils import RopeParameters
@@ -136,6 +137,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.eps}"
 
 
+@use_kernel_forward_from_hub("GeGLUMLP", condition=lambda module: module.config.hidden_act == "gelu_pytorch_tanh")
 class GemmaMLP(LlamaMLP):
     def __init__(self, config):
         super().__init__(config)
diff --git a/src/transformers/models/gemma2/configuration_gemma2.py b/src/transformers/models/gemma2/configuration_gemma2.py
index 11d7b012099f..a04dcbb14261 100644
--- a/src/transformers/models/gemma2/configuration_gemma2.py
+++ b/src/transformers/models/gemma2/configuration_gemma2.py
@@ -64,6 +64,7 @@ class Gemma2Config(PreTrainedConfig):
         "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
         "norm": (["hidden_states"], ["hidden_states"]),
     }
+    attribute_map = {"hidden_act": "hidden_activation"}
 
     vocab_size: int = 256000
     hidden_size: int = 2304
diff --git a/src/transformers/models/gemma2/modeling_gemma2.py b/src/transformers/models/gemma2/modeling_gemma2.py
index fe8f41724b9d..5a295e03442e 100644
--- a/src/transformers/models/gemma2/modeling_gemma2.py
+++ b/src/transformers/models/gemma2/modeling_gemma2.py
@@ -66,6 +66,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.eps}"
 
 
+@use_kernel_forward_from_hub("GeGLUMLP", condition=lambda module: module.config.hidden_act == "gelu_pytorch_tanh")
 class Gemma2MLP(nn.Module):
     def __init__(self, config):
         super().__init__()
@@ -75,7 +76,7 @@ def __init__(self, config):
         self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
-        self.act_fn = ACT2FN[config.hidden_activation]
+        self.act_fn = ACT2FN[config.hidden_act]
 
     def forward(self, x):
         down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
diff --git a/src/transformers/models/gemma2/modular_gemma2.py b/src/transformers/models/gemma2/modular_gemma2.py
index d798196f1807..3d01ad1e49e2 100644
--- a/src/transformers/models/gemma2/modular_gemma2.py
+++ b/src/transformers/models/gemma2/modular_gemma2.py
@@ -18,7 +18,6 @@
 import torch.nn as nn
 from huggingface_hub.dataclasses import strict
 
-from ...activations import ACT2FN
 from ...cache_utils import Cache, DynamicCache
 from ...configuration_utils import PreTrainedConfig
 from ...masking_utils import create_causal_mask, create_sliding_window_causal_mask
@@ -92,6 +91,7 @@ class Gemma2Config(PreTrainedConfig):
         "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
         "norm": (["hidden_states"], ["hidden_states"]),
     }
+    attribute_map = {"hidden_act": "hidden_activation"}
 
     vocab_size: int = 256000
     hidden_size: int = 2304
@@ -141,9 +141,7 @@ class Gemma2RMSNorm(GemmaRMSNorm):
 
 
 class Gemma2MLP(GemmaMLP):
-    def __init__(self, config):
-        super().__init__(config)
-        self.act_fn = ACT2FN[config.hidden_activation]
+    pass
 
 
 class Gemma2RotaryEmbedding(GemmaRotaryEmbedding):
diff --git a/src/transformers/models/gemma3/configuration_gemma3.py b/src/transformers/models/gemma3/configuration_gemma3.py
index b640a906fbef..9a40687acc83 100644
--- a/src/transformers/models/gemma3/configuration_gemma3.py
+++ b/src/transformers/models/gemma3/configuration_gemma3.py
@@ -73,6 +73,7 @@ class Gemma3TextConfig(PreTrainedConfig):
         "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
         "norm": (["hidden_states"], ["hidden_states"]),
     }
+    attribute_map = {"hidden_act": "hidden_activation"}
 
     vocab_size: int = 262_208
     hidden_size: int = 2304
diff --git a/src/transformers/models/gemma3/modeling_gemma3.py b/src/transformers/models/gemma3/modeling_gemma3.py
index e870ec1ad81a..cc43497588c8 100644
--- a/src/transformers/models/gemma3/modeling_gemma3.py
+++ b/src/transformers/models/gemma3/modeling_gemma3.py
@@ -117,6 +117,7 @@ def forward(self, input_ids: torch.Tensor):
         return super().forward(input_ids) * self.embed_scale.to(self.weight.dtype)
 
 
+@use_kernel_forward_from_hub("GeGLUMLP", condition=lambda module: module.config.hidden_act == "gelu_pytorch_tanh")
 class Gemma3MLP(nn.Module):
     def __init__(self, config: Gemma3TextConfig):
         super().__init__()
@@ -126,7 +127,7 @@ def __init__(self, config: Gemma3TextConfig):
         self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
-        self.act_fn = ACT2FN[config.hidden_activation]
+        self.act_fn = ACT2FN[config.hidden_act]
 
     def forward(self, x):
         down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
diff --git a/src/transformers/models/gemma3n/configuration_gemma3n.py b/src/transformers/models/gemma3n/configuration_gemma3n.py
index 8f197fa3bb3f..462c3a962502 100644
--- a/src/transformers/models/gemma3n/configuration_gemma3n.py
+++ b/src/transformers/models/gemma3n/configuration_gemma3n.py
@@ -94,6 +94,7 @@ class Gemma3nTextConfig(PreTrainedConfig):
         "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
         "norm": (["hidden_states"], ["hidden_states"]),
     }
+    attribute_map = {"hidden_act": "hidden_activation"}
 
     vocab_size: int = 262_400
     hidden_size: int = 2048
diff --git a/src/transformers/models/gemma3n/modeling_gemma3n.py b/src/transformers/models/gemma3n/modeling_gemma3n.py
index 57dfb6b2d4e4..837ca6b29497 100644
--- a/src/transformers/models/gemma3n/modeling_gemma3n.py
+++ b/src/transformers/models/gemma3n/modeling_gemma3n.py
@@ -957,7 +957,7 @@ def __init__(self, config: Gemma3nTextConfig, layer_idx: int = 0):
         self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
-        self.act_fn = ACT2FN[config.hidden_activation]
+        self.act_fn = ACT2FN[config.hidden_act]
         self.activation_sparsity = config.activation_sparsity_pattern[layer_idx]
 
     def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
diff --git a/src/transformers/models/gemma3n/modular_gemma3n.py b/src/transformers/models/gemma3n/modular_gemma3n.py
index 3e4f2331d148..9132eece54d8 100644
--- a/src/transformers/models/gemma3n/modular_gemma3n.py
+++ b/src/transformers/models/gemma3n/modular_gemma3n.py
@@ -40,7 +40,7 @@
     logging,
     torch_compilable_check,
 )
-from ...utils.generic import merge_with_config_defaults
+from ...utils.generic import merge_with_config_defaults, no_inherit_decorator
 from ...utils.output_capturing import capture_outputs
 from ..auto import AutoModel
 from ..gemma2.modeling_gemma2 import (
@@ -1303,6 +1303,7 @@ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
         return hidden_states + normed_laurel_hidden_states
 
 
+@no_inherit_decorator
 class Gemma3nTextMLP(Gemma2MLP):
     def __init__(self, config: Gemma3nTextConfig, layer_idx: int = 0):
         super().__init__(config)
diff --git a/src/transformers/models/gemma4/configuration_gemma4.py b/src/transformers/models/gemma4/configuration_gemma4.py
index ff970b4c7ef5..c381eab0dc28 100644
--- a/src/transformers/models/gemma4/configuration_gemma4.py
+++ b/src/transformers/models/gemma4/configuration_gemma4.py
@@ -148,6 +148,7 @@ class Gemma4TextConfig(PreTrainedConfig):
         "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
         "norm": (["hidden_states"], ["hidden_states"]),
     }
+    attribute_map = {"hidden_act": "hidden_activation"}
 
     vocab_size: int = 262_144
     hidden_size: int = 2304
@@ -266,6 +267,7 @@ class Gemma4VisionConfig(PreTrainedConfig):
         "encoder.layers.*.mlp.up_proj": "colwise",
         "encoder.layers.*.mlp.down_proj": "rowwise",
     }
+    attribute_map = {"hidden_act": "hidden_activation"}
 
     default_theta = 100.0
 
diff --git a/src/transformers/models/gemma4/modeling_gemma4.py b/src/transformers/models/gemma4/modeling_gemma4.py
index 79f626e86996..7ae059bac03a 100644
--- a/src/transformers/models/gemma4/modeling_gemma4.py
+++ b/src/transformers/models/gemma4/modeling_gemma4.py
@@ -33,7 +33,7 @@
 from ...cache_utils import Cache, DynamicCache
 from ...configuration_utils import PreTrainedConfig
 from ...generation import GenerationMixin
-from ...integrations import use_experts_implementation
+from ...integrations import use_experts_implementation, use_kernel_forward_from_hub
 from ...masking_utils import (
     _preprocess_mask_arguments,
     blockwise_overlay,
@@ -688,6 +688,7 @@ def forward(
         return hidden_states, padding_positions
 
 
+@use_kernel_forward_from_hub("GeGLUMLP", condition=lambda module: module.config.hidden_act == "gelu_pytorch_tanh")
 class Gemma4VisionMLP(nn.Module):
     def __init__(self, config: Gemma4VisionConfig):
         super().__init__()
@@ -697,7 +698,7 @@ def __init__(self, config: Gemma4VisionConfig):
         self.gate_proj = Gemma4ClippableLinear(config, self.hidden_size, self.intermediate_size)
         self.up_proj = Gemma4ClippableLinear(config, self.hidden_size, self.intermediate_size)
         self.down_proj = Gemma4ClippableLinear(config, self.intermediate_size, self.hidden_size)
-        self.act_fn = ACT2FN[config.hidden_activation]
+        self.act_fn = ACT2FN[config.hidden_act]
 
     def forward(self, x):
         down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
@@ -1062,6 +1063,7 @@ def forward(
         return BaseModelOutputWithPast(last_hidden_state=hidden_states)
 
 
+@use_kernel_forward_from_hub("GeGLUMLP", condition=lambda module: module.config.hidden_act == "gelu_pytorch_tanh")
 class Gemma4TextMLP(nn.Module):
     def __init__(self, config: Gemma4TextConfig, layer_idx: int):
         super().__init__()
@@ -1074,7 +1076,7 @@ def __init__(self, config: Gemma4TextConfig, layer_idx: int):
         self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
-        self.act_fn = ACT2FN[config.hidden_activation]
+        self.act_fn = ACT2FN[config.hidden_act]
 
     def forward(self, x):
         down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
diff --git a/src/transformers/models/gemma4_unified/configuration_gemma4_unified.py b/src/transformers/models/gemma4_unified/configuration_gemma4_unified.py
index 0b93b9a1701d..70968133126e 100644
--- a/src/transformers/models/gemma4_unified/configuration_gemma4_unified.py
+++ b/src/transformers/models/gemma4_unified/configuration_gemma4_unified.py
@@ -120,6 +120,7 @@ class Gemma4UnifiedTextConfig(PreTrainedConfig):
         "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
         "norm": (["hidden_states"], ["hidden_states"]),
     }
+    attribute_map = {"hidden_act": "hidden_activation"}
 
     vocab_size: int = 262_144
     hidden_size: int = 2304
diff --git a/src/transformers/models/gemma4_unified/modeling_gemma4_unified.py b/src/transformers/models/gemma4_unified/modeling_gemma4_unified.py
index 9a897479a360..896ab6cf6370 100644
--- a/src/transformers/models/gemma4_unified/modeling_gemma4_unified.py
+++ b/src/transformers/models/gemma4_unified/modeling_gemma4_unified.py
@@ -29,6 +29,7 @@
 from ...cache_utils import Cache, DynamicCache
 from ...configuration_utils import PreTrainedConfig
 from ...generation import GenerationMixin
+from ...integrations import use_kernel_forward_from_hub
 from ...masking_utils import (
     _preprocess_mask_arguments,
     blockwise_overlay,
@@ -455,6 +456,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("GeGLUMLP", condition=lambda module: module.config.hidden_act == "gelu_pytorch_tanh")
 class Gemma4UnifiedTextMLP(nn.Module):
     def __init__(self, config: Gemma4UnifiedTextConfig, layer_idx: int):
         super().__init__()
@@ -467,7 +469,7 @@ def __init__(self, config: Gemma4UnifiedTextConfig, layer_idx: int):
         self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
-        self.act_fn = ACT2FN[config.hidden_activation]
+        self.act_fn = ACT2FN[config.hidden_act]
 
     def forward(self, x):
         down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
diff --git a/src/transformers/models/glm4_moe/modeling_glm4_moe.py b/src/transformers/models/glm4_moe/modeling_glm4_moe.py
index 539da895bd6d..ba591e80508d 100644
--- a/src/transformers/models/glm4_moe/modeling_glm4_moe.py
+++ b/src/transformers/models/glm4_moe/modeling_glm4_moe.py
@@ -260,6 +260,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Glm4MoeMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/glm4_moe_lite/modeling_glm4_moe_lite.py b/src/transformers/models/glm4_moe_lite/modeling_glm4_moe_lite.py
index bff71001ca1f..bfb6e40434d5 100644
--- a/src/transformers/models/glm4_moe_lite/modeling_glm4_moe_lite.py
+++ b/src/transformers/models/glm4_moe_lite/modeling_glm4_moe_lite.py
@@ -367,6 +367,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Glm4MoeLiteMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/glm4v_moe/modeling_glm4v_moe.py b/src/transformers/models/glm4v_moe/modeling_glm4v_moe.py
index 8f1c1577b4e8..37ea3f885f7d 100644
--- a/src/transformers/models/glm4v_moe/modeling_glm4v_moe.py
+++ b/src/transformers/models/glm4v_moe/modeling_glm4v_moe.py
@@ -314,6 +314,7 @@ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
         return hidden_states
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Glm4vMoeTextMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/glm5_next/modeling_glm5_next.py b/src/transformers/models/glm5_next/modeling_glm5_next.py
index ba0551f62ad8..08df37e518d9 100644
--- a/src/transformers/models/glm5_next/modeling_glm5_next.py
+++ b/src/transformers/models/glm5_next/modeling_glm5_next.py
@@ -82,6 +82,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Glm5NextTextMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/glm_moe_dsa/modeling_glm_moe_dsa.py b/src/transformers/models/glm_moe_dsa/modeling_glm_moe_dsa.py
index 5d7ba1a8f3df..6e4872e4f2e0 100644
--- a/src/transformers/models/glm_moe_dsa/modeling_glm_moe_dsa.py
+++ b/src/transformers/models/glm_moe_dsa/modeling_glm_moe_dsa.py
@@ -480,6 +480,7 @@ def forward(
         return attn_output, attn_weights, topk_indices
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class GlmMoeDsaMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/granite/modeling_granite.py b/src/transformers/models/granite/modeling_granite.py
index ba4affd2c93b..a5cec34c913b 100644
--- a/src/transformers/models/granite/modeling_granite.py
+++ b/src/transformers/models/granite/modeling_granite.py
@@ -200,6 +200,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class GraniteMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/granite4_vision/modeling_granite4_vision.py b/src/transformers/models/granite4_vision/modeling_granite4_vision.py
index f84d6593237c..080ec479c2df 100644
--- a/src/transformers/models/granite4_vision/modeling_granite4_vision.py
+++ b/src/transformers/models/granite4_vision/modeling_granite4_vision.py
@@ -428,6 +428,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Granite4VisionTextMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/granite_swa/modeling_granite_swa.py b/src/transformers/models/granite_swa/modeling_granite_swa.py
index 3ff773877342..b384a68f52c1 100644
--- a/src/transformers/models/granite_swa/modeling_granite_swa.py
+++ b/src/transformers/models/granite_swa/modeling_granite_swa.py
@@ -217,6 +217,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class GraniteSWAMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/helium/modeling_helium.py b/src/transformers/models/helium/modeling_helium.py
index 1403cd7def6f..548454e31945 100644
--- a/src/transformers/models/helium/modeling_helium.py
+++ b/src/transformers/models/helium/modeling_helium.py
@@ -27,6 +27,7 @@
 from ...activations import ACT2FN
 from ...cache_utils import Cache, DynamicCache
 from ...generation import GenerationMixin
+from ...integrations import use_kernel_forward_from_hub
 from ...masking_utils import create_causal_mask
 from ...modeling_layers import (
     GenericForSequenceClassification,
@@ -118,6 +119,7 @@ def forward(self, x, position_ids):
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class HeliumMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/higgs_audio_v2/modeling_higgs_audio_v2.py b/src/transformers/models/higgs_audio_v2/modeling_higgs_audio_v2.py
index 12a21f886a9f..7a4077e12c44 100644
--- a/src/transformers/models/higgs_audio_v2/modeling_higgs_audio_v2.py
+++ b/src/transformers/models/higgs_audio_v2/modeling_higgs_audio_v2.py
@@ -45,6 +45,7 @@
 logger = logging.get_logger(__name__)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class HiggsAudioV2MLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/hrm_text/modeling_hrm_text.py b/src/transformers/models/hrm_text/modeling_hrm_text.py
index 315e997c4fa8..8324d6018202 100644
--- a/src/transformers/models/hrm_text/modeling_hrm_text.py
+++ b/src/transformers/models/hrm_text/modeling_hrm_text.py
@@ -67,6 +67,7 @@ def extra_repr(self):
         return f"eps={self.eps}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class HrmTextMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/hunyuan_v1_dense/modeling_hunyuan_v1_dense.py b/src/transformers/models/hunyuan_v1_dense/modeling_hunyuan_v1_dense.py
index 7dccbbb44311..354e6151b850 100644
--- a/src/transformers/models/hunyuan_v1_dense/modeling_hunyuan_v1_dense.py
+++ b/src/transformers/models/hunyuan_v1_dense/modeling_hunyuan_v1_dense.py
@@ -64,6 +64,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class HunYuanDenseV1MLP(nn.Module):
     def __init__(self, config: HunYuanDenseV1Config, layer_idx=None, is_shared_mlp=False):
         super().__init__()
diff --git a/src/transformers/models/hunyuan_v1_moe/modeling_hunyuan_v1_moe.py b/src/transformers/models/hunyuan_v1_moe/modeling_hunyuan_v1_moe.py
index 60c69c1999f3..6e2b72a629f3 100644
--- a/src/transformers/models/hunyuan_v1_moe/modeling_hunyuan_v1_moe.py
+++ b/src/transformers/models/hunyuan_v1_moe/modeling_hunyuan_v1_moe.py
@@ -63,6 +63,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class HunYuanMoEV1MLP(nn.Module):
     def __init__(self, config: HunYuanMoEV1Config):
         super().__init__()
diff --git a/src/transformers/models/hunyuan_vl/modeling_hunyuan_vl.py b/src/transformers/models/hunyuan_vl/modeling_hunyuan_vl.py
index 116f36c138b2..3d49df932cbe 100644
--- a/src/transformers/models/hunyuan_vl/modeling_hunyuan_vl.py
+++ b/src/transformers/models/hunyuan_vl/modeling_hunyuan_vl.py
@@ -577,6 +577,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class HunYuanVLMLP(nn.Module):
     def __init__(self, config: HunYuanVLConfig, layer_idx=None, is_shared_mlp=False):
         super().__init__()
diff --git a/src/transformers/models/hy_v3/modeling_hy_v3.py b/src/transformers/models/hy_v3/modeling_hy_v3.py
index 5f9958d0fcb0..8381b9b78a30 100644
--- a/src/transformers/models/hy_v3/modeling_hy_v3.py
+++ b/src/transformers/models/hy_v3/modeling_hy_v3.py
@@ -120,6 +120,7 @@ def forward(self, x, position_ids):
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class HYV3MLP(nn.Module):
     def __init__(self, config: HYV3Config, intermediate_size: int | None = None):
         super().__init__()
diff --git a/src/transformers/models/hyperclovax/modeling_hyperclovax.py b/src/transformers/models/hyperclovax/modeling_hyperclovax.py
index b869e984b315..a4127440ad3a 100644
--- a/src/transformers/models/hyperclovax/modeling_hyperclovax.py
+++ b/src/transformers/models/hyperclovax/modeling_hyperclovax.py
@@ -258,6 +258,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class HyperCLOVAXMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/inkling/modular_inkling.py b/src/transformers/models/inkling/modular_inkling.py
index 89276fa3128c..1cdc7ac80282 100644
--- a/src/transformers/models/inkling/modular_inkling.py
+++ b/src/transformers/models/inkling/modular_inkling.py
@@ -40,7 +40,7 @@
     logging,
     torch_compilable_check,
 )
-from ...utils.generic import merge_with_config_defaults
+from ...utils.generic import merge_with_config_defaults, no_inherit_decorator
 from ...utils.output_capturing import capture_outputs
 from ..gemma3.modeling_gemma3 import (
     Gemma3CausalLMOutputWithPast,
@@ -500,6 +500,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@no_inherit_decorator
 class InklingMLP(Gemma3MLP):
     def __init__(self, config: InklingTextConfig):
         super().__init__(config)
diff --git a/src/transformers/models/jamba/modeling_jamba.py b/src/transformers/models/jamba/modeling_jamba.py
index 70fc8e44b31f..50407f7fc03d 100755
--- a/src/transformers/models/jamba/modeling_jamba.py
+++ b/src/transformers/models/jamba/modeling_jamba.py
@@ -611,6 +611,7 @@ def forward(
         return contextualized_states
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class JambaMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/laguna/modeling_laguna.py b/src/transformers/models/laguna/modeling_laguna.py
index edaf222a815a..7f7271769172 100644
--- a/src/transformers/models/laguna/modeling_laguna.py
+++ b/src/transformers/models/laguna/modeling_laguna.py
@@ -138,6 +138,7 @@ def forward(self, x, position_ids, layer_type):
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class LagunaMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/llama/modeling_llama.py b/src/transformers/models/llama/modeling_llama.py
index ceaea60027fa..b8f1e1cfd737 100644
--- a/src/transformers/models/llama/modeling_llama.py
+++ b/src/transformers/models/llama/modeling_llama.py
@@ -160,6 +160,7 @@ def apply_rotary_pos_emb(q, k, cos, sin, unsqueeze_dim=1):
     return q_embed, k_embed
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class LlamaMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/longcat_flash/modeling_longcat_flash.py b/src/transformers/models/longcat_flash/modeling_longcat_flash.py
index 1e071ca9a838..61e37db03a2c 100644
--- a/src/transformers/models/longcat_flash/modeling_longcat_flash.py
+++ b/src/transformers/models/longcat_flash/modeling_longcat_flash.py
@@ -124,6 +124,7 @@ def forward(self, x, position_ids):
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class LongcatFlashMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/mellum/modeling_mellum.py b/src/transformers/models/mellum/modeling_mellum.py
index c4be60ecf723..7cb39219068c 100644
--- a/src/transformers/models/mellum/modeling_mellum.py
+++ b/src/transformers/models/mellum/modeling_mellum.py
@@ -355,6 +355,7 @@ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
         return final_hidden_states.reshape(batch_size, sequence_length, hidden_dim)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class MellumMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/mimo_v2_flash/modeling_mimo_v2_flash.py b/src/transformers/models/mimo_v2_flash/modeling_mimo_v2_flash.py
index e8494170fc49..fc93085b3cd1 100644
--- a/src/transformers/models/mimo_v2_flash/modeling_mimo_v2_flash.py
+++ b/src/transformers/models/mimo_v2_flash/modeling_mimo_v2_flash.py
@@ -237,6 +237,7 @@ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
         return self.experts(hidden_states, topk_indices, topk_weights).view(*orig_shape)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class MiMoV2FlashMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/minicpm3/modeling_minicpm3.py b/src/transformers/models/minicpm3/modeling_minicpm3.py
index 028cd61c3e88..fdd7738861e0 100644
--- a/src/transformers/models/minicpm3/modeling_minicpm3.py
+++ b/src/transformers/models/minicpm3/modeling_minicpm3.py
@@ -134,6 +134,7 @@ def forward(self, x, position_ids):
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class MiniCPM3MLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/ministral/modeling_ministral.py b/src/transformers/models/ministral/modeling_ministral.py
index febf238729ba..06e69c77f4d0 100644
--- a/src/transformers/models/ministral/modeling_ministral.py
+++ b/src/transformers/models/ministral/modeling_ministral.py
@@ -47,6 +47,7 @@
 from .configuration_ministral import MinistralConfig
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class MinistralMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/ministral3/modeling_ministral3.py b/src/transformers/models/ministral3/modeling_ministral3.py
index 7bb5fdb0cc62..ea585954a761 100644
--- a/src/transformers/models/ministral3/modeling_ministral3.py
+++ b/src/transformers/models/ministral3/modeling_ministral3.py
@@ -173,6 +173,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Ministral3MLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/mistral/modeling_mistral.py b/src/transformers/models/mistral/modeling_mistral.py
index 6e3333148323..d9f2acf01e8b 100644
--- a/src/transformers/models/mistral/modeling_mistral.py
+++ b/src/transformers/models/mistral/modeling_mistral.py
@@ -32,6 +32,7 @@
 from .configuration_mistral import MistralConfig
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class MistralMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/mistral4/modeling_mistral4.py b/src/transformers/models/mistral4/modeling_mistral4.py
index db5f842e535b..1f7e6e187af1 100644
--- a/src/transformers/models/mistral4/modeling_mistral4.py
+++ b/src/transformers/models/mistral4/modeling_mistral4.py
@@ -125,6 +125,7 @@ def forward(self, x, position_ids):
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Mistral4MLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/moonshine_streaming/modeling_moonshine_streaming.py b/src/transformers/models/moonshine_streaming/modeling_moonshine_streaming.py
index 5f14849afbbd..26667ab38187 100644
--- a/src/transformers/models/moonshine_streaming/modeling_moonshine_streaming.py
+++ b/src/transformers/models/moonshine_streaming/modeling_moonshine_streaming.py
@@ -28,6 +28,7 @@
 from ...activations import ACT2FN
 from ...cache_utils import Cache, DynamicCache, EncoderDecoderCache
 from ...generation import GenerationMixin
+from ...integrations import use_kernel_forward_from_hub
 from ...masking_utils import create_bidirectional_mask, create_causal_mask
 from ...modeling_flash_attention_utils import FlashAttentionKwargs
 from ...modeling_layers import GradientCheckpointingLayer
@@ -431,6 +432,7 @@ def forward(
         return MoonshineStreamingEncoderModelOutput(last_hidden_state=hidden_states, attention_mask=attention_mask)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class MoonshinMoonshineStreamingDecoderMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/muse_glimmer/configuration_muse_glimmer.py b/src/transformers/models/muse_glimmer/configuration_muse_glimmer.py
index 19cb5a57eb43..af3d52ab5ad3 100644
--- a/src/transformers/models/muse_glimmer/configuration_muse_glimmer.py
+++ b/src/transformers/models/muse_glimmer/configuration_muse_glimmer.py
@@ -105,6 +105,7 @@ class MuseGlimmerTextConfig(PreTrainedConfig):
         "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
         "norm": (["hidden_states"], ["hidden_states"]),
     }
+    attribute_map = {"hidden_act": "hidden_activation"}
 
     vocab_size: int = 202_048
     hidden_size: int = 6656
diff --git a/src/transformers/models/muse_glimmer/modeling_muse_glimmer.py b/src/transformers/models/muse_glimmer/modeling_muse_glimmer.py
index 2bd5278248f1..095f9475f7b2 100644
--- a/src/transformers/models/muse_glimmer/modeling_muse_glimmer.py
+++ b/src/transformers/models/muse_glimmer/modeling_muse_glimmer.py
@@ -140,6 +140,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.eps}"
 
 
+@use_kernel_forward_from_hub("GeGLUMLP", condition=lambda module: module.config.hidden_act == "gelu_pytorch_tanh")
 class MuseGlimmerTextMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
@@ -149,7 +150,7 @@ def __init__(self, config):
         self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
-        self.act_fn = ACT2FN[config.hidden_activation]
+        self.act_fn = ACT2FN[config.hidden_act]
 
     def forward(self, x):
         down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
diff --git a/src/transformers/models/muse_glimmer_assistant/modeling_muse_glimmer_assistant.py b/src/transformers/models/muse_glimmer_assistant/modeling_muse_glimmer_assistant.py
index 71e6c71f761d..0c58948558a4 100644
--- a/src/transformers/models/muse_glimmer_assistant/modeling_muse_glimmer_assistant.py
+++ b/src/transformers/models/muse_glimmer_assistant/modeling_muse_glimmer_assistant.py
@@ -217,6 +217,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class MuseGlimmerAssistantMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/nomic_bert/modeling_nomic_bert.py b/src/transformers/models/nomic_bert/modeling_nomic_bert.py
index dbc40d6b196c..50d5b5179a5f 100644
--- a/src/transformers/models/nomic_bert/modeling_nomic_bert.py
+++ b/src/transformers/models/nomic_bert/modeling_nomic_bert.py
@@ -256,6 +256,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class NomicBertMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/nomic_bert/modular_nomic_bert.py b/src/transformers/models/nomic_bert/modular_nomic_bert.py
index bd27e8034b62..6d5656ec870e 100644
--- a/src/transformers/models/nomic_bert/modular_nomic_bert.py
+++ b/src/transformers/models/nomic_bert/modular_nomic_bert.py
@@ -19,6 +19,7 @@
 from torch.nn import CrossEntropyLoss
 
 from ...configuration_utils import PreTrainedConfig
+from ...integrations import use_kernel_forward_from_hub
 from ...masking_utils import create_bidirectional_mask
 from ...modeling_outputs import (
     BaseModelOutputWithPooling,
@@ -117,6 +118,7 @@ def __init__(self, config):
         self.o_proj = nn.Linear(config.num_attention_heads * self.head_dim, config.hidden_size, bias=False)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class NomicBertMLP(GemmaMLP):
     pass
 
diff --git a/src/transformers/models/olmo/modeling_olmo.py b/src/transformers/models/olmo/modeling_olmo.py
index dd2f1ed261c3..97f6318f5c77 100644
--- a/src/transformers/models/olmo/modeling_olmo.py
+++ b/src/transformers/models/olmo/modeling_olmo.py
@@ -60,6 +60,7 @@ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
         )
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class OlmoMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/olmo2/modeling_olmo2.py b/src/transformers/models/olmo2/modeling_olmo2.py
index 3a18aebc13b3..38f633866390 100644
--- a/src/transformers/models/olmo2/modeling_olmo2.py
+++ b/src/transformers/models/olmo2/modeling_olmo2.py
@@ -266,6 +266,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Olmo2MLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/olmo3/modeling_olmo3.py b/src/transformers/models/olmo3/modeling_olmo3.py
index 85a4317f683f..88f50d8dfb76 100644
--- a/src/transformers/models/olmo3/modeling_olmo3.py
+++ b/src/transformers/models/olmo3/modeling_olmo3.py
@@ -210,6 +210,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Olmo3MLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/olmo_hybrid/modeling_olmo_hybrid.py b/src/transformers/models/olmo_hybrid/modeling_olmo_hybrid.py
index c1dbf7320c77..d380d072e4f9 100644
--- a/src/transformers/models/olmo_hybrid/modeling_olmo_hybrid.py
+++ b/src/transformers/models/olmo_hybrid/modeling_olmo_hybrid.py
@@ -752,6 +752,7 @@ def forward(
         return output
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class OlmoHybridMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/olmoe/modeling_olmoe.py b/src/transformers/models/olmoe/modeling_olmoe.py
index ab732636f8d3..ce66b0ec68c4 100644
--- a/src/transformers/models/olmoe/modeling_olmoe.py
+++ b/src/transformers/models/olmoe/modeling_olmoe.py
@@ -118,6 +118,7 @@ def forward(self, x, position_ids):
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class OlmoeMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/olmoe/modular_olmoe.py b/src/transformers/models/olmoe/modular_olmoe.py
index adadcf0c9e5d..b78f6f6d425f 100644
--- a/src/transformers/models/olmoe/modular_olmoe.py
+++ b/src/transformers/models/olmoe/modular_olmoe.py
@@ -19,6 +19,7 @@
 from ... import initialization as init
 from ...cache_utils import Cache, DynamicCache
 from ...generation import GenerationMixin
+from ...integrations import use_kernel_forward_from_hub
 from ...masking_utils import create_causal_mask
 from ...modeling_outputs import MoeModelOutputWithPast
 from ...modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
@@ -51,6 +52,7 @@ class OlmoeRotaryEmbedding(LlamaRotaryEmbedding):
     pass
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class OlmoeMLP(GemmaMLP):
     pass
 
diff --git a/src/transformers/models/ovis2/modeling_ovis2.py b/src/transformers/models/ovis2/modeling_ovis2.py
index e489ccf5b446..798d661ff098 100644
--- a/src/transformers/models/ovis2/modeling_ovis2.py
+++ b/src/transformers/models/ovis2/modeling_ovis2.py
@@ -124,6 +124,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Ovis2VisionMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
@@ -254,6 +255,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Ovis2MLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/paddleocr_vl/modeling_paddleocr_vl.py b/src/transformers/models/paddleocr_vl/modeling_paddleocr_vl.py
index 50109e0c6dcd..8897fbff0815 100644
--- a/src/transformers/models/paddleocr_vl/modeling_paddleocr_vl.py
+++ b/src/transformers/models/paddleocr_vl/modeling_paddleocr_vl.py
@@ -174,6 +174,7 @@ def forward(self, x, position_ids):
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class PaddleOCRMLP(nn.Module):
     def __init__(self, config: PaddleOCRTextConfig):
         super().__init__()
diff --git a/src/transformers/models/pe_audio/modeling_pe_audio.py b/src/transformers/models/pe_audio/modeling_pe_audio.py
index c08e02ba5a4a..2750f4bb7043 100644
--- a/src/transformers/models/pe_audio/modeling_pe_audio.py
+++ b/src/transformers/models/pe_audio/modeling_pe_audio.py
@@ -431,6 +431,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class PeAudioEncoderMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/pe_audio_video/modeling_pe_audio_video.py b/src/transformers/models/pe_audio_video/modeling_pe_audio_video.py
index b2a5d96df50a..236e08c741f9 100644
--- a/src/transformers/models/pe_audio_video/modeling_pe_audio_video.py
+++ b/src/transformers/models/pe_audio_video/modeling_pe_audio_video.py
@@ -368,6 +368,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class PeAudioVideoEncoderMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/pe_video/modeling_pe_video.py b/src/transformers/models/pe_video/modeling_pe_video.py
index c1726eb8aa0d..e5134e0cb223 100644
--- a/src/transformers/models/pe_video/modeling_pe_video.py
+++ b/src/transformers/models/pe_video/modeling_pe_video.py
@@ -355,6 +355,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class PeVideoEncoderMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/qwen2/modeling_qwen2.py b/src/transformers/models/qwen2/modeling_qwen2.py
index b4fed98fafae..4ad77081a407 100644
--- a/src/transformers/models/qwen2/modeling_qwen2.py
+++ b/src/transformers/models/qwen2/modeling_qwen2.py
@@ -32,6 +32,7 @@
 from .configuration_qwen2 import Qwen2Config
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Qwen2MLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/qwen2_moe/modeling_qwen2_moe.py b/src/transformers/models/qwen2_moe/modeling_qwen2_moe.py
index bdbeeab12271..4f3527f3daf9 100644
--- a/src/transformers/models/qwen2_moe/modeling_qwen2_moe.py
+++ b/src/transformers/models/qwen2_moe/modeling_qwen2_moe.py
@@ -127,6 +127,7 @@ def forward(self, x, position_ids):
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Qwen2MoeMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/qwen2_moe/modular_qwen2_moe.py b/src/transformers/models/qwen2_moe/modular_qwen2_moe.py
index dc620da2782a..72b52747e097 100644
--- a/src/transformers/models/qwen2_moe/modular_qwen2_moe.py
+++ b/src/transformers/models/qwen2_moe/modular_qwen2_moe.py
@@ -25,6 +25,7 @@
 from ...activations import ACT2FN
 from ...cache_utils import Cache, DynamicCache
 from ...generation import GenerationMixin
+from ...integrations import use_kernel_forward_from_hub
 from ...masking_utils import create_causal_mask, create_sliding_window_causal_mask
 from ...modeling_layers import (
     GenericForQuestionAnswering,
@@ -56,6 +57,7 @@ class Qwen2MoeRotaryEmbedding(Gemma2RotaryEmbedding):
     pass
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Qwen2MoeMLP(GemmaMLP):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/qwen3/modeling_qwen3.py b/src/transformers/models/qwen3/modeling_qwen3.py
index 5e474983ab92..a1aab54c85dc 100644
--- a/src/transformers/models/qwen3/modeling_qwen3.py
+++ b/src/transformers/models/qwen3/modeling_qwen3.py
@@ -67,6 +67,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Qwen3MLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/qwen3/modular_qwen3.py b/src/transformers/models/qwen3/modular_qwen3.py
index ea6ddb2f7499..12164c6e5793 100644
--- a/src/transformers/models/qwen3/modular_qwen3.py
+++ b/src/transformers/models/qwen3/modular_qwen3.py
@@ -18,6 +18,7 @@
 import torch
 
 from ...cache_utils import Cache
+from ...integrations import use_kernel_forward_from_hub
 from ...modeling_flash_attention_utils import FlashAttentionKwargs
 from ...modeling_outputs import CausalLMOutputWithPast
 from ...modeling_utils import ALL_ATTENTION_FUNCTIONS
@@ -49,6 +50,7 @@ class Qwen3RMSNorm(Qwen2RMSNorm):
     pass
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Qwen3MLP(GemmaMLP):
     pass
 
diff --git a/src/transformers/models/qwen3_5/modeling_qwen3_5.py b/src/transformers/models/qwen3_5/modeling_qwen3_5.py
index 06c19237c170..a5ed4f96c0eb 100644
--- a/src/transformers/models/qwen3_5/modeling_qwen3_5.py
+++ b/src/transformers/models/qwen3_5/modeling_qwen3_5.py
@@ -777,6 +777,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Qwen3_5MLP(nn.Module):
     def __init__(self, config: Qwen3_5Config, intermediate_size: int):
         super().__init__()
diff --git a/src/transformers/models/qwen3_5_moe/modeling_qwen3_5_moe.py b/src/transformers/models/qwen3_5_moe/modeling_qwen3_5_moe.py
index e538a653f210..33a9bdbbe0b6 100644
--- a/src/transformers/models/qwen3_5_moe/modeling_qwen3_5_moe.py
+++ b/src/transformers/models/qwen3_5_moe/modeling_qwen3_5_moe.py
@@ -777,6 +777,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Qwen3_5MoeMLP(nn.Module):
     def __init__(self, config: Qwen3_5MoeConfig, intermediate_size: int):
         super().__init__()
diff --git a/src/transformers/models/qwen3_moe/modeling_qwen3_moe.py b/src/transformers/models/qwen3_moe/modeling_qwen3_moe.py
index a4b13c545601..0a2181931c6a 100644
--- a/src/transformers/models/qwen3_moe/modeling_qwen3_moe.py
+++ b/src/transformers/models/qwen3_moe/modeling_qwen3_moe.py
@@ -190,6 +190,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Qwen3MoeMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/qwen3_next/modeling_qwen3_next.py b/src/transformers/models/qwen3_next/modeling_qwen3_next.py
index 6a7e6dc0791b..ea86bc03d8a1 100644
--- a/src/transformers/models/qwen3_next/modeling_qwen3_next.py
+++ b/src/transformers/models/qwen3_next/modeling_qwen3_next.py
@@ -767,6 +767,7 @@ def forward(
         return output
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Qwen3NextMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/qwen3_omni_moe/modeling_qwen3_omni_moe.py b/src/transformers/models/qwen3_omni_moe/modeling_qwen3_omni_moe.py
index 6ec317e1debd..c64e2edb8ee7 100644
--- a/src/transformers/models/qwen3_omni_moe/modeling_qwen3_omni_moe.py
+++ b/src/transformers/models/qwen3_omni_moe/modeling_qwen3_omni_moe.py
@@ -1473,6 +1473,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Qwen3OmniMoeThinkerTextMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
@@ -2376,6 +2377,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Qwen3OmniMoeMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
@@ -2687,6 +2689,7 @@ class Qwen3OmniMoeTalkerRotaryEmbedding(Qwen3OmniMoeThinkerTextRotaryEmbedding):
     pass
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Qwen3OmniMoeTalkerTextMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
@@ -3391,6 +3394,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Qwen3OmniMoeCode2WavMlp(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/qwen3_vl/modeling_qwen3_vl.py b/src/transformers/models/qwen3_vl/modeling_qwen3_vl.py
index 7d964d080792..2da4fa4dee35 100644
--- a/src/transformers/models/qwen3_vl/modeling_qwen3_vl.py
+++ b/src/transformers/models/qwen3_vl/modeling_qwen3_vl.py
@@ -506,6 +506,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Qwen3VLTextMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/qwen3_vl_moe/modeling_qwen3_vl_moe.py b/src/transformers/models/qwen3_vl_moe/modeling_qwen3_vl_moe.py
index 42bb1d65346b..299594dd2e74 100644
--- a/src/transformers/models/qwen3_vl_moe/modeling_qwen3_vl_moe.py
+++ b/src/transformers/models/qwen3_vl_moe/modeling_qwen3_vl_moe.py
@@ -302,6 +302,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Qwen3VLMoeTextMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/qwen4_exp/modeling_qwen4_exp.py b/src/transformers/models/qwen4_exp/modeling_qwen4_exp.py
index 4273886cc4ab..9935fd896fd6 100644
--- a/src/transformers/models/qwen4_exp/modeling_qwen4_exp.py
+++ b/src/transformers/models/qwen4_exp/modeling_qwen4_exp.py
@@ -912,6 +912,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class Qwen4ExpTextMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/smollm3/modeling_smollm3.py b/src/transformers/models/smollm3/modeling_smollm3.py
index fcf2eac3556e..9a381adaca49 100644
--- a/src/transformers/models/smollm3/modeling_smollm3.py
+++ b/src/transformers/models/smollm3/modeling_smollm3.py
@@ -268,6 +268,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class SmolLM3MLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/solar_open/modeling_solar_open.py b/src/transformers/models/solar_open/modeling_solar_open.py
index e685d8848892..f871cab9bea4 100644
--- a/src/transformers/models/solar_open/modeling_solar_open.py
+++ b/src/transformers/models/solar_open/modeling_solar_open.py
@@ -83,6 +83,7 @@ def forward(
         return hidden_states
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class SolarOpenMLP(nn.Module):
     def __init__(self, config, intermediate_size=None):
         super().__init__()
diff --git a/src/transformers/models/step3p7/modular_step3p7.py b/src/transformers/models/step3p7/modular_step3p7.py
index 3c714f671b03..89461460c786 100644
--- a/src/transformers/models/step3p7/modular_step3p7.py
+++ b/src/transformers/models/step3p7/modular_step3p7.py
@@ -782,6 +782,7 @@ class Step3p7RMSNorm(MiniMaxM3VLRMSNorm):
     pass
 
 
+@no_inherit_decorator
 class Step3p7MLP(DeepseekV4MLP):
     def __init__(self, config, layer_idx, is_shared_expert=False):
         super().__init__(config)
diff --git a/src/transformers/models/t5gemma/configuration_t5gemma.py b/src/transformers/models/t5gemma/configuration_t5gemma.py
index 9de40c832259..245d619c08ba 100644
--- a/src/transformers/models/t5gemma/configuration_t5gemma.py
+++ b/src/transformers/models/t5gemma/configuration_t5gemma.py
@@ -64,6 +64,7 @@ class T5GemmaModuleConfig(PreTrainedConfig):
         "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
         "norm": (["hidden_states"], ["hidden_states"]),
     }
+    attribute_map = {"hidden_act": "hidden_activation"}
 
     vocab_size: int = 256000
     hidden_size: int = 2304
diff --git a/src/transformers/models/t5gemma/modeling_t5gemma.py b/src/transformers/models/t5gemma/modeling_t5gemma.py
index ddbacedb7782..e1b5c231a2d6 100644
--- a/src/transformers/models/t5gemma/modeling_t5gemma.py
+++ b/src/transformers/models/t5gemma/modeling_t5gemma.py
@@ -87,7 +87,7 @@ def __init__(self, config):
         self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
-        self.act_fn = ACT2FN[config.hidden_activation]
+        self.act_fn = ACT2FN[config.hidden_act]
         self.dropout = nn.Dropout(config.dropout_rate)
 
     def forward(self, x):
diff --git a/src/transformers/models/t5gemma/modular_t5gemma.py b/src/transformers/models/t5gemma/modular_t5gemma.py
index 6f0c399830e5..c686e19a6c7f 100644
--- a/src/transformers/models/t5gemma/modular_t5gemma.py
+++ b/src/transformers/models/t5gemma/modular_t5gemma.py
@@ -48,7 +48,7 @@
     can_return_tuple,
     logging,
 )
-from ...utils.generic import merge_with_config_defaults
+from ...utils.generic import merge_with_config_defaults, no_inherit_decorator
 from ...utils.output_capturing import OutputRecorder, capture_outputs
 from ..gemma2.configuration_gemma2 import Gemma2Config
 from ..gemma2.modeling_gemma2 import (
@@ -153,6 +153,7 @@ class T5GemmaRMSNorm(Gemma2RMSNorm):
     pass
 
 
+@no_inherit_decorator
 class T5GemmaMLP(Gemma2MLP):
     def __init__(self, config):
         super().__init__(config)
diff --git a/src/transformers/models/t5gemma2/configuration_t5gemma2.py b/src/transformers/models/t5gemma2/configuration_t5gemma2.py
index ebddcb7810f2..fdf08a82317b 100644
--- a/src/transformers/models/t5gemma2/configuration_t5gemma2.py
+++ b/src/transformers/models/t5gemma2/configuration_t5gemma2.py
@@ -60,6 +60,7 @@ class T5Gemma2TextConfig(PreTrainedConfig):
         "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
         "norm": (["hidden_states"], ["hidden_states"]),
     }
+    attribute_map = {"hidden_act": "hidden_activation"}
 
     vocab_size: int = 262_208
     hidden_size: int = 2304
@@ -244,6 +245,7 @@ class T5Gemma2DecoderConfig(PreTrainedConfig):
         "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
         "norm": (["hidden_states"], ["hidden_states"]),
     }
+    attribute_map = {"hidden_act": "hidden_activation"}
 
     vocab_size: int = 262_208
     hidden_size: int = 2304
diff --git a/src/transformers/models/t5gemma2/modeling_t5gemma2.py b/src/transformers/models/t5gemma2/modeling_t5gemma2.py
index d54e95d76a59..6164c3b7e22d 100644
--- a/src/transformers/models/t5gemma2/modeling_t5gemma2.py
+++ b/src/transformers/models/t5gemma2/modeling_t5gemma2.py
@@ -81,7 +81,7 @@ def __init__(self, config: T5Gemma2TextConfig):
         self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
-        self.act_fn = ACT2FN[config.hidden_activation]
+        self.act_fn = ACT2FN[config.hidden_act]
         self.dropout = nn.Dropout(config.dropout_rate)
 
     def forward(self, x):
diff --git a/src/transformers/models/t5gemma2/modular_t5gemma2.py b/src/transformers/models/t5gemma2/modular_t5gemma2.py
index 131106686fca..4f6ada91a2dc 100644
--- a/src/transformers/models/t5gemma2/modular_t5gemma2.py
+++ b/src/transformers/models/t5gemma2/modular_t5gemma2.py
@@ -45,7 +45,7 @@
     logging,
     torch_compilable_check,
 )
-from ...utils.generic import merge_with_config_defaults
+from ...utils.generic import merge_with_config_defaults, no_inherit_decorator
 from ...utils.output_capturing import OutputRecorder, capture_outputs
 from ..auto import AutoModel
 from ..gemma3.configuration_gemma3 import Gemma3Config, Gemma3TextConfig
@@ -231,6 +231,7 @@ class T5Gemma2RMSNorm(Gemma3RMSNorm):
     pass
 
 
+@no_inherit_decorator
 class T5Gemma2MLP(Gemma3MLP):
     def __init__(self, config: T5Gemma2TextConfig):
         super().__init__(config)
diff --git a/src/transformers/models/vaultgemma/configuration_vaultgemma.py b/src/transformers/models/vaultgemma/configuration_vaultgemma.py
index a60b7e8edc0c..8c646db44a0b 100644
--- a/src/transformers/models/vaultgemma/configuration_vaultgemma.py
+++ b/src/transformers/models/vaultgemma/configuration_vaultgemma.py
@@ -63,6 +63,7 @@ class VaultGemmaConfig(PreTrainedConfig):
         "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
         "norm": (["hidden_states"], ["hidden_states"]),
     }
+    attribute_map = {"hidden_act": "hidden_activation"}
 
     vocab_size: int = 256000
     hidden_size: int = 2304
diff --git a/src/transformers/models/vaultgemma/modeling_vaultgemma.py b/src/transformers/models/vaultgemma/modeling_vaultgemma.py
index ad84fd75c125..4eb81e8632a0 100644
--- a/src/transformers/models/vaultgemma/modeling_vaultgemma.py
+++ b/src/transformers/models/vaultgemma/modeling_vaultgemma.py
@@ -63,6 +63,7 @@ def extra_repr(self):
         return f"{tuple(self.weight.shape)}, eps={self.eps}"
 
 
+@use_kernel_forward_from_hub("GeGLUMLP", condition=lambda module: module.config.hidden_act == "gelu_pytorch_tanh")
 class VaultGemmaMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
@@ -72,7 +73,7 @@ def __init__(self, config):
         self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
         self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
-        self.act_fn = ACT2FN[config.hidden_activation]
+        self.act_fn = ACT2FN[config.hidden_act]
 
     def forward(self, x):
         down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
diff --git a/src/transformers/models/vibevoice/modeling_vibevoice.py b/src/transformers/models/vibevoice/modeling_vibevoice.py
index 77afa1d29ca3..16d7b8a17fd3 100644
--- a/src/transformers/models/vibevoice/modeling_vibevoice.py
+++ b/src/transformers/models/vibevoice/modeling_vibevoice.py
@@ -118,6 +118,7 @@ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
         return hidden_states
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class VibeVoiceMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/voxtral_realtime/modeling_voxtral_realtime.py b/src/transformers/models/voxtral_realtime/modeling_voxtral_realtime.py
index e3abf0c60b9b..152eff8257a7 100644
--- a/src/transformers/models/voxtral_realtime/modeling_voxtral_realtime.py
+++ b/src/transformers/models/voxtral_realtime/modeling_voxtral_realtime.py
@@ -396,6 +396,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class VoxtralRealtimeMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
@@ -681,6 +682,7 @@ def forward(
         return attn_output, attn_weights
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class VoxtralRealtimeTextMLP(nn.Module):
     def __init__(self, config):
         super().__init__()
diff --git a/src/transformers/models/youtu/modeling_youtu.py b/src/transformers/models/youtu/modeling_youtu.py
index 37f573e32a5d..4c7ca03d1f6e 100644
--- a/src/transformers/models/youtu/modeling_youtu.py
+++ b/src/transformers/models/youtu/modeling_youtu.py
@@ -127,6 +127,7 @@ def forward(self, x, position_ids):
         return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
 
 
+@use_kernel_forward_from_hub("SwiGLUMLP", condition=lambda module: module.config.hidden_act == "silu")
 class YoutuMLP(nn.Module):
     def __init__(self, config):
         super().__init__()