Build uploaded using `kernels`.

.gitattributes

@@ -40,3 +40,9 @@ build/torch28-cxx11-cpu-x86_64-linux/_rmsnorm_a7a4369.abi3.so filter=lfs diff=lf
 build/torch28-cxx11-xpu20251-x86_64-linux/_rmsnorm_a7a4369.abi3.so filter=lfs diff=lfs merge=lfs -text
 build/torch29-cxx11-cpu-x86_64-linux/_rmsnorm_a7a4369.abi3.so filter=lfs diff=lfs merge=lfs -text
 build/torch29-cxx11-xpu20252-x86_64-linux/_rmsnorm_a7a4369.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch210-cxx11-cpu-x86_64-linux/_rmsnorm_fb26d8c.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch210-cxx11-xpu20253-x86_64-linux/_rmsnorm_fb26d8c.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch28-cxx11-cpu-x86_64-linux/_rmsnorm_fb26d8c.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch28-cxx11-xpu20251-x86_64-linux/_rmsnorm_fb26d8c.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch29-cxx11-cpu-x86_64-linux/_rmsnorm_fb26d8c.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch29-cxx11-xpu20252-x86_64-linux/_rmsnorm_fb26d8c.abi3.so filter=lfs diff=lfs merge=lfs -text

build/torch210-cxx11-cpu-x86_64-linux/__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/torch210-cxx11-cpu-x86_64-linux/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _rmsnorm_fb26d8c
+ops = torch.ops._rmsnorm_fb26d8c
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_rmsnorm_fb26d8c::{op_name}"

build/torch210-cxx11-cpu-x86_64-linux/_rmsnorm_fb26d8c.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f8a1c744d46b5b0b6455825653741008b06242630ae9946f0205ac2c055dbc7e
+size 326352

build/torch210-cxx11-cpu-x86_64-linux/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/torch210-cxx11-cpu-x86_64-linux/metadata.json

@@ -0,0 +1 @@
+{"python-depends":[]}

build/torch210-cxx11-cpu-x86_64-linux/rmsnorm/__init__.py

@@ -0,0 +1,26 @@
+import ctypes
+import sys
+
+import importlib
+from pathlib import Path
+from types import ModuleType
+
+def _import_from_path(file_path: Path) -> ModuleType:
+    # We cannot use the module name as-is, after adding it to `sys.modules`,
+    # it would also be used for other imports. So, we make a module name that
+    # depends on the path for it to be unique using the hex-encoded hash of
+    # the path.
+    path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value)
+    module_name = path_hash
+    spec = importlib.util.spec_from_file_location(module_name, file_path)
+    if spec is None:
+        raise ImportError(f"Cannot load spec for {module_name} from {file_path}")
+    module = importlib.util.module_from_spec(spec)
+    if module is None:
+        raise ImportError(f"Cannot load module {module_name} from spec")
+    sys.modules[module_name] = module
+    spec.loader.exec_module(module)  # type: ignore
+    return module
+
+
+globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py")))

build/torch210-cxx11-xpu20253-x86_64-linux/__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/torch210-cxx11-xpu20253-x86_64-linux/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _rmsnorm_fb26d8c
+ops = torch.ops._rmsnorm_fb26d8c
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_rmsnorm_fb26d8c::{op_name}"

build/torch210-cxx11-xpu20253-x86_64-linux/_rmsnorm_fb26d8c.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4be94737423cc4d02f4be83f38144614d71ccd8672d96699f0b10136dd541847
+size 104941392

build/torch210-cxx11-xpu20253-x86_64-linux/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/torch210-cxx11-xpu20253-x86_64-linux/metadata.json

@@ -0,0 +1 @@
+{"python-depends":[]}

build/torch210-cxx11-xpu20253-x86_64-linux/rmsnorm/__init__.py

@@ -0,0 +1,26 @@
+import ctypes
+import sys
+
+import importlib
+from pathlib import Path
+from types import ModuleType
+
+def _import_from_path(file_path: Path) -> ModuleType:
+    # We cannot use the module name as-is, after adding it to `sys.modules`,
+    # it would also be used for other imports. So, we make a module name that
+    # depends on the path for it to be unique using the hex-encoded hash of
+    # the path.
+    path_hash = "{:x}".format(ctypes.c_size_t(hash(file_path.absolute())).value)
+    module_name = path_hash
+    spec = importlib.util.spec_from_file_location(module_name, file_path)
+    if spec is None:
+        raise ImportError(f"Cannot load spec for {module_name} from {file_path}")
+    module = importlib.util.module_from_spec(spec)
+    if module is None:
+        raise ImportError(f"Cannot load module {module_name} from spec")
+    sys.modules[module_name] = module
+    spec.loader.exec_module(module)  # type: ignore
+    return module
+
+
+globals().update(vars(_import_from_path(Path(__file__).parent.parent / "__init__.py")))

build/torch28-cxx11-cpu-x86_64-linux/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _rmsnorm_a7a4369
-ops = torch.ops._rmsnorm_a7a4369
+from . import _rmsnorm_fb26d8c
+ops = torch.ops._rmsnorm_fb26d8c
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_rmsnorm_a7a4369::{op_name}"
+    return f"_rmsnorm_fb26d8c::{op_name}"

build/torch28-cxx11-cpu-x86_64-linux/{_rmsnorm_a7a4369.abi3.so → _rmsnorm_fb26d8c.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:55621bcd39646f99f979ef03f823fcb35afceead03961675835a80a2c5b6d134
+oid sha256:055fc9c5e82e48e503963bac3da30001e128774d8d9a333680b8aacab0650644
 size 324616

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

@@ -1,9 +1,9 @@
 import torch
-from . import _rmsnorm_a7a4369
-ops = torch.ops._rmsnorm_a7a4369
+from . import _rmsnorm_fb26d8c
+ops = torch.ops._rmsnorm_fb26d8c
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_rmsnorm_a7a4369::{op_name}"
+    return f"_rmsnorm_fb26d8c::{op_name}"

build/torch28-cxx11-xpu20251-x86_64-linux/{_rmsnorm_a7a4369.abi3.so → _rmsnorm_fb26d8c.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f265f642541565b741228747e4503426aadfaa9cdb428b1a571fa695fd726812
+oid sha256:c0ba9e0355977f76b16f6346377026ffde2977c613ee9b5633083d6f95f4e07c
 size 103861336

build/torch29-cxx11-cpu-x86_64-linux/_ops.py

@@ -1,9 +1,9 @@
 import torch
-from . import _rmsnorm_a7a4369
-ops = torch.ops._rmsnorm_a7a4369
+from . import _rmsnorm_fb26d8c
+ops = torch.ops._rmsnorm_fb26d8c
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_rmsnorm_a7a4369::{op_name}"
+    return f"_rmsnorm_fb26d8c::{op_name}"

build/torch29-cxx11-cpu-x86_64-linux/{_rmsnorm_a7a4369.abi3.so → _rmsnorm_fb26d8c.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1ec310cc2696fcc0ef846f6e940e1eec2ba7a50848b69e59a7834f3ddee5927c
+oid sha256:97de49bd6f5edb8a54394a123362f30026a119e4f8ccf796884f108c343ec562
 size 324592

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

@@ -1,9 +1,9 @@
 import torch
-from . import _rmsnorm_a7a4369
-ops = torch.ops._rmsnorm_a7a4369
+from . import _rmsnorm_fb26d8c
+ops = torch.ops._rmsnorm_fb26d8c
 
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_rmsnorm_a7a4369::{op_name}"
+    return f"_rmsnorm_fb26d8c::{op_name}"

build/torch29-cxx11-xpu20252-x86_64-linux/{_rmsnorm_a7a4369.abi3.so → _rmsnorm_fb26d8c.abi3.so}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f13902ee157cb167ad294a906252f5a349ee13c8afe0d71b51a895cf6944cbfe
+oid sha256:47dcb713294a6eca6d920f2e9aba27be280d75ac2d356845232008210d1df17a
 size 102340240