Build uploaded using `kernels`.

.gitattributes

@@ -33,3 +33,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+build/torch27-cxx11-xpu20250-x86_64-linux/rmsnorm/_rmsnorm_0d12ee5.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch28-cxx11-xpu20251-x86_64-linux/rmsnorm/_rmsnorm_0d12ee5.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch29-cxx11-xpu20252-x86_64-linux/rmsnorm/_rmsnorm_0d12ee5.abi3.so filter=lfs diff=lfs merge=lfs -text

build/torch27-cxx11-xpu20250-x86_64-linux/rmsnorm/__init__.py

@@ -0,0 +1,14 @@
+from . import layers
+
+from ._ops import ops
+
+
+def apply_rms_norm(input, weight, eps):
+    return ops.apply_rms_norm(
+            input,
+            weight,
+            eps,
+    )
+
+__all__ = ["layers", "apply_rms_norm"]
+

build/torch27-cxx11-xpu20250-x86_64-linux/rmsnorm/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _rmsnorm_0d12ee5
+ops = torch.ops._rmsnorm_0d12ee5
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_rmsnorm_0d12ee5::{op_name}"

build/torch27-cxx11-xpu20250-x86_64-linux/rmsnorm/_rmsnorm_0d12ee5.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:79eb24cb07a24a3f829ce1d210bd0cbd79badd0cc236710a84e83c15575ddf04
+size 100963504

build/torch27-cxx11-xpu20250-x86_64-linux/rmsnorm/layers.py

@@ -0,0 +1,36 @@
+import torch
+from ._ops import ops
+
+class RMSNorm(torch.nn.Module):
+    """
+    RMSNorm module that uses the optimized LigerRMSNormFunction.
+    
+    Args:
+        hidden_size (int): The size of the hidden dimension.
+        eps (float, optional): The epsilon value for numerical stability. Defaults to 1e-6.
+        offset (float, optional): Offset value to shift the weight tensor. Defaults to 0.0.
+        casting_mode (str, optional): The casting mode to use. Defaults to "llama".
+        in_place (bool, optional): Whether to modify dY in-place to store dX during backward. Defaults to True.
+    """
+    
+
+    weight: torch.Tensor
+    variance_epsilon: float
+    
+    def forward(self, hidden_states):
+        """
+        Apply RMS normalization to the input tensor.
+        
+        Args:
+            hidden_states (torch.Tensor): Input tensor of shape (B, T, H) or (BxT, H)
+            
+        Returns:
+            torch.Tensor: Normalized tensor of the same shape as input
+        """
+        return ops.apply_rms_norm(
+            hidden_states,
+            self.weight, 
+            self.variance_epsilon,
+        )
+    
+__all__ = ["RMSNorm"]

build/torch28-cxx11-xpu20251-x86_64-linux/rmsnorm/__init__.py

@@ -0,0 +1,14 @@
+from . import layers
+
+from ._ops import ops
+
+
+def apply_rms_norm(input, weight, eps):
+    return ops.apply_rms_norm(
+            input,
+            weight,
+            eps,
+    )
+
+__all__ = ["layers", "apply_rms_norm"]
+

build/torch28-cxx11-xpu20251-x86_64-linux/rmsnorm/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _rmsnorm_0d12ee5
+ops = torch.ops._rmsnorm_0d12ee5
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_rmsnorm_0d12ee5::{op_name}"

build/torch28-cxx11-xpu20251-x86_64-linux/rmsnorm/_rmsnorm_0d12ee5.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f0b10d6efd62c187242ff110bfb3a2278bb076c053d50aa63354f34d1353a931
+size 103943040

build/torch28-cxx11-xpu20251-x86_64-linux/rmsnorm/layers.py

@@ -0,0 +1,36 @@
+import torch
+from ._ops import ops
+
+class RMSNorm(torch.nn.Module):
+    """
+    RMSNorm module that uses the optimized LigerRMSNormFunction.
+    
+    Args:
+        hidden_size (int): The size of the hidden dimension.
+        eps (float, optional): The epsilon value for numerical stability. Defaults to 1e-6.
+        offset (float, optional): Offset value to shift the weight tensor. Defaults to 0.0.
+        casting_mode (str, optional): The casting mode to use. Defaults to "llama".
+        in_place (bool, optional): Whether to modify dY in-place to store dX during backward. Defaults to True.
+    """
+    
+
+    weight: torch.Tensor
+    variance_epsilon: float
+    
+    def forward(self, hidden_states):
+        """
+        Apply RMS normalization to the input tensor.
+        
+        Args:
+            hidden_states (torch.Tensor): Input tensor of shape (B, T, H) or (BxT, H)
+            
+        Returns:
+            torch.Tensor: Normalized tensor of the same shape as input
+        """
+        return ops.apply_rms_norm(
+            hidden_states,
+            self.weight, 
+            self.variance_epsilon,
+        )
+    
+__all__ = ["RMSNorm"]

build/torch29-cxx11-xpu20252-x86_64-linux/rmsnorm/__init__.py

@@ -0,0 +1,14 @@
+from . import layers
+
+from ._ops import ops
+
+
+def apply_rms_norm(input, weight, eps):
+    return ops.apply_rms_norm(
+            input,
+            weight,
+            eps,
+    )
+
+__all__ = ["layers", "apply_rms_norm"]
+

build/torch29-cxx11-xpu20252-x86_64-linux/rmsnorm/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _rmsnorm_0d12ee5
+ops = torch.ops._rmsnorm_0d12ee5
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_rmsnorm_0d12ee5::{op_name}"

build/torch29-cxx11-xpu20252-x86_64-linux/rmsnorm/_rmsnorm_0d12ee5.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ee351d25752fbbd5e05c23a5a15445f1576ec58978fe0e8126f649384cf998aa
+size 102416296

build/torch29-cxx11-xpu20252-x86_64-linux/rmsnorm/layers.py

@@ -0,0 +1,36 @@
+import torch
+from ._ops import ops
+
+class RMSNorm(torch.nn.Module):
+    """
+    RMSNorm module that uses the optimized LigerRMSNormFunction.
+    
+    Args:
+        hidden_size (int): The size of the hidden dimension.
+        eps (float, optional): The epsilon value for numerical stability. Defaults to 1e-6.
+        offset (float, optional): Offset value to shift the weight tensor. Defaults to 0.0.
+        casting_mode (str, optional): The casting mode to use. Defaults to "llama".
+        in_place (bool, optional): Whether to modify dY in-place to store dX during backward. Defaults to True.
+    """
+    
+
+    weight: torch.Tensor
+    variance_epsilon: float
+    
+    def forward(self, hidden_states):
+        """
+        Apply RMS normalization to the input tensor.
+        
+        Args:
+            hidden_states (torch.Tensor): Input tensor of shape (B, T, H) or (BxT, H)
+            
+        Returns:
+            torch.Tensor: Normalized tensor of the same shape as input
+        """
+        return ops.apply_rms_norm(
+            hidden_states,
+            self.weight, 
+            self.variance_epsilon,
+        )
+    
+__all__ = ["RMSNorm"]