Kernels:

kernels-community
/

rmsnorm

Trusted publisher

Kernel card Files Files and versions

xet

Community

danieldk HF Staff commited on 20 days ago

Commit

d8a5f38

1 Parent(s): d679cae

Remove builds incompatible with kernels >= 0.14

Browse files

Files changed (42) hide show

build/torch210-xpu20253-x86_64-windows/__init__.py +0 -27
build/torch210-xpu20253-x86_64-windows/_ops.py +0 -9
build/torch210-xpu20253-x86_64-windows/_rmsnorm_xpu_1e2f028.pyd +0 -3
build/torch210-xpu20253-x86_64-windows/layers.py +0 -59
build/torch210-xpu20253-x86_64-windows/metadata.json +0 -8
build/torch210-xpu20253-x86_64-windows/rmsnorm/__init__.py +0 -26
build/torch27-cxx11-xpu20250-x86_64-linux/rmsnorm/__init__.py +0 -14
build/torch27-cxx11-xpu20250-x86_64-linux/rmsnorm/__pycache__/__init__.cpython-313.pyc +0 -0
build/torch27-cxx11-xpu20250-x86_64-linux/rmsnorm/__pycache__/_ops.cpython-313.pyc +0 -0
build/torch27-cxx11-xpu20250-x86_64-linux/rmsnorm/__pycache__/layers.cpython-313.pyc +0 -0
build/torch27-cxx11-xpu20250-x86_64-linux/rmsnorm/_ops.py +0 -9
build/torch27-cxx11-xpu20250-x86_64-linux/rmsnorm/_rmsnorm_0d12ee5.abi3.so +0 -3
build/torch27-cxx11-xpu20250-x86_64-linux/rmsnorm/layers.py +0 -36
build/torch28-cxx11-cpu-x86_64-linux/__init__.py +0 -27
build/torch28-cxx11-cpu-x86_64-linux/_ops.py +0 -9
build/torch28-cxx11-cpu-x86_64-linux/_rmsnorm_235cde1.abi3.so +0 -3
build/torch28-cxx11-cpu-x86_64-linux/layers.py +0 -59
build/torch28-cxx11-cpu-x86_64-linux/metadata.json +0 -4
build/torch28-cxx11-cpu-x86_64-linux/rmsnorm/__init__.py +0 -26
build/torch28-cxx11-xpu20251-x86_64-linux/__init__.py +0 -27
build/torch28-cxx11-xpu20251-x86_64-linux/_ops.py +0 -9
build/torch28-cxx11-xpu20251-x86_64-linux/_rmsnorm_235cde1.abi3.so +0 -3
build/torch28-cxx11-xpu20251-x86_64-linux/layers.py +0 -59
build/torch28-cxx11-xpu20251-x86_64-linux/metadata.json +0 -4
build/torch28-cxx11-xpu20251-x86_64-linux/rmsnorm/__init__.py +0 -26
build/torch29-cxx11-cpu-x86_64-linux/__init__.py +0 -27
build/torch29-cxx11-cpu-x86_64-linux/_ops.py +0 -9
build/torch29-cxx11-cpu-x86_64-linux/_rmsnorm_cpu_b3d66c6.abi3.so +0 -3
build/torch29-cxx11-cpu-x86_64-linux/layers.py +0 -59
build/torch29-cxx11-cpu-x86_64-linux/metadata.json +0 -5
build/torch29-cxx11-cpu-x86_64-linux/rmsnorm/__init__.py +0 -26
build/torch29-cxx11-xpu20252-x86_64-linux/__init__.py +0 -27
build/torch29-cxx11-xpu20252-x86_64-linux/_ops.py +0 -9
build/torch29-cxx11-xpu20252-x86_64-linux/_rmsnorm_xpu_cec90b8.abi3.so +0 -3
build/torch29-cxx11-xpu20252-x86_64-linux/layers.py +0 -59
build/torch29-cxx11-xpu20252-x86_64-linux/metadata.json +0 -8
build/torch29-cxx11-xpu20252-x86_64-linux/rmsnorm/__init__.py +0 -26
build/torch29-xpu20252-x86_64-windows/metadata.json +0 -4
build/torch29-xpu20252-x86_64-windows/rmsnorm/__init__.py +0 -27
build/torch29-xpu20252-x86_64-windows/rmsnorm/_ops.py +0 -9
build/torch29-xpu20252-x86_64-windows/rmsnorm/_rmsnorm_96c9886.pyd +0 -3
build/torch29-xpu20252-x86_64-windows/rmsnorm/layers.py +0 -59

build/torch210-xpu20253-x86_64-windows/__init__.py DELETED Viewed

@@ -1,27 +0,0 @@
-from . import layers
-from ._ops import ops
-def apply_rms_norm(input, weight, eps):
-    # ops.apply_rms_norm returns [output, rstd]
-    return ops.apply_rms_norm(
-            input,
-            weight,
-            eps,
-    )[0]
-def apply_rms_norm_backward(grad_output, input, weight, output, rstd, eps, input_requires_grad=True, weight_requires_grad=True):
-    return ops.apply_rms_norm_backward(
-            grad_output,
-            input,
-            weight,
-            output,
-            rstd,
-            eps,
-            input_requires_grad,
-            weight_requires_grad
-    )
-__all__ = ["layers", "apply_rms_norm_forward", "apply_rms_norm_backward"]

build/torch210-xpu20253-x86_64-windows/_ops.py DELETED Viewed

@@ -1,9 +0,0 @@
-import torch
-from . import _rmsnorm_xpu_1e2f028
-ops = torch.ops._rmsnorm_xpu_1e2f028
-def add_op_namespace_prefix(op_name: str):
-    """
-    Prefix op by namespace.
-    """
-    return f"_rmsnorm_xpu_1e2f028::{op_name}"

build/torch210-xpu20253-x86_64-windows/_rmsnorm_xpu_1e2f028.pyd DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:3406c8ace46216f6b49350810fa2a022ddfda89b1975ede4e556301a136f9d49
-size 2363904

build/torch210-xpu20253-x86_64-windows/layers.py DELETED Viewed

@@ -1,59 +0,0 @@
-import torch
-from ._ops import ops
-class RMSNormFunction(torch.autograd.Function):
-    @staticmethod
-    def forward(ctx, hidden_states, weight, variance_epsilon):
-        ctx.variance_epsilon = variance_epsilon
-        output, rstd = ops.apply_rms_norm(hidden_states, weight, variance_epsilon)
-        ctx.save_for_backward(hidden_states, weight, output, rstd)
-        return output
-    @staticmethod
-    def backward(ctx, grad_output):
-        hidden_states, weight, output, rstd = ctx.saved_tensors
-        grads = ops.apply_rms_norm_backward(
-            grad_output,
-            hidden_states,
-            weight,
-            output,
-            rstd,
-            ctx.variance_epsilon,
-            ctx.needs_input_grad[0],
-            ctx.needs_input_grad[1]
-        )
-        return grads[0], grads[1], None
-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 RMSNormFunction.apply(
-            hidden_states,
-            self.weight,
-            self.variance_epsilon,
-        )
-__all__ = ["RMSNorm"]

build/torch210-xpu20253-x86_64-windows/metadata.json DELETED Viewed

@@ -1,8 +0,0 @@
-{
-  "version": 1,
-  "license": "Apache-2.0",
-  "python-depends": [],
-  "backend": {
-    "type": "xpu"
-  }
-}

build/torch210-xpu20253-x86_64-windows/rmsnorm/__init__.py DELETED Viewed

@@ -1,26 +0,0 @@
-import ctypes
-import importlib.util
-import sys
-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/torch27-cxx11-xpu20250-x86_64-linux/rmsnorm/__init__.py DELETED Viewed

@@ -1,14 +0,0 @@
-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/__pycache__/__init__.cpython-313.pyc DELETED Viewed

Binary file (491 Bytes)

build/torch27-cxx11-xpu20250-x86_64-linux/rmsnorm/__pycache__/_ops.cpython-313.pyc DELETED Viewed

Binary file (520 Bytes)

build/torch27-cxx11-xpu20250-x86_64-linux/rmsnorm/__pycache__/layers.cpython-313.pyc DELETED Viewed

Binary file (1.68 kB)

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

@@ -1,9 +0,0 @@
-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 DELETED Viewed

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

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

@@ -1,36 +0,0 @@
-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-cpu-x86_64-linux/__init__.py DELETED Viewed

@@ -1,27 +0,0 @@
-from . import layers
-from ._ops import ops
-def apply_rms_norm(input, weight, eps):
-    # ops.apply_rms_norm returns [output, rstd]
-    return ops.apply_rms_norm(
-            input,
-            weight,
-            eps,
-    )[0]
-def apply_rms_norm_backward(grad_output, input, weight, output, rstd, eps, input_requires_grad=True, weight_requires_grad=True):
-    return ops.apply_rms_norm_backward(
-            grad_output,
-            input,
-            weight,
-            output,
-            rstd,
-            eps,
-            input_requires_grad,
-            weight_requires_grad
-    )
-__all__ = ["layers", "apply_rms_norm_forward", "apply_rms_norm_backward"]

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

@@ -1,9 +0,0 @@
-import torch
-from . import _rmsnorm_235cde1
-ops = torch.ops._rmsnorm_235cde1
-def add_op_namespace_prefix(op_name: str):
-    """
-    Prefix op by namespace.
-    """
-    return f"_rmsnorm_235cde1::{op_name}"

build/torch28-cxx11-cpu-x86_64-linux/_rmsnorm_235cde1.abi3.so DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:16c92de9cefabeeadc60ffff87189a1e66ecb9ea19b343570ac55e9d9c7d98fe
-size 156648

build/torch28-cxx11-cpu-x86_64-linux/layers.py DELETED Viewed

@@ -1,59 +0,0 @@
-import torch
-from ._ops import ops
-class RMSNormFunction(torch.autograd.Function):
-    @staticmethod
-    def forward(ctx, hidden_states, weight, variance_epsilon):
-        ctx.variance_epsilon = variance_epsilon
-        output, rstd = ops.apply_rms_norm(hidden_states, weight, variance_epsilon)
-        ctx.save_for_backward(hidden_states, weight, output, rstd)
-        return output
-    @staticmethod
-    def backward(ctx, grad_output):
-        hidden_states, weight, output, rstd = ctx.saved_tensors
-        grads = ops.apply_rms_norm_backward(
-            grad_output,
-            hidden_states,
-            weight,
-            output,
-            rstd,
-            ctx.variance_epsilon,
-            ctx.needs_input_grad[0],
-            ctx.needs_input_grad[1]
-        )
-        return grads[0], grads[1], None
-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 RMSNormFunction.apply(
-            hidden_states,
-            self.weight,
-            self.variance_epsilon,
-        )
-__all__ = ["RMSNorm"]

build/torch28-cxx11-cpu-x86_64-linux/metadata.json DELETED Viewed

@@ -1,4 +0,0 @@
-{
-  "version": 1,
-  "python-depends": []
-}

build/torch28-cxx11-cpu-x86_64-linux/rmsnorm/__init__.py DELETED Viewed

@@ -1,26 +0,0 @@
-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-xpu20251-x86_64-linux/__init__.py DELETED Viewed

@@ -1,27 +0,0 @@
-from . import layers
-from ._ops import ops
-def apply_rms_norm(input, weight, eps):
-    # ops.apply_rms_norm returns [output, rstd]
-    return ops.apply_rms_norm(
-            input,
-            weight,
-            eps,
-    )[0]
-def apply_rms_norm_backward(grad_output, input, weight, output, rstd, eps, input_requires_grad=True, weight_requires_grad=True):
-    return ops.apply_rms_norm_backward(
-            grad_output,
-            input,
-            weight,
-            output,
-            rstd,
-            eps,
-            input_requires_grad,
-            weight_requires_grad
-    )
-__all__ = ["layers", "apply_rms_norm_forward", "apply_rms_norm_backward"]

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

@@ -1,9 +0,0 @@
-import torch
-from . import _rmsnorm_235cde1
-ops = torch.ops._rmsnorm_235cde1
-def add_op_namespace_prefix(op_name: str):
-    """
-    Prefix op by namespace.
-    """
-    return f"_rmsnorm_235cde1::{op_name}"

build/torch28-cxx11-xpu20251-x86_64-linux/_rmsnorm_235cde1.abi3.so DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:77c4b43d63dc74b210633da81630023a6d6e359a7a1115bff55da9f4436053d9
-size 103700632

build/torch28-cxx11-xpu20251-x86_64-linux/layers.py DELETED Viewed

@@ -1,59 +0,0 @@
-import torch
-from ._ops import ops
-class RMSNormFunction(torch.autograd.Function):
-    @staticmethod
-    def forward(ctx, hidden_states, weight, variance_epsilon):
-        ctx.variance_epsilon = variance_epsilon
-        output, rstd = ops.apply_rms_norm(hidden_states, weight, variance_epsilon)
-        ctx.save_for_backward(hidden_states, weight, output, rstd)
-        return output
-    @staticmethod
-    def backward(ctx, grad_output):
-        hidden_states, weight, output, rstd = ctx.saved_tensors
-        grads = ops.apply_rms_norm_backward(
-            grad_output,
-            hidden_states,
-            weight,
-            output,
-            rstd,
-            ctx.variance_epsilon,
-            ctx.needs_input_grad[0],
-            ctx.needs_input_grad[1]
-        )
-        return grads[0], grads[1], None
-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 RMSNormFunction.apply(
-            hidden_states,
-            self.weight,
-            self.variance_epsilon,
-        )
-__all__ = ["RMSNorm"]

build/torch28-cxx11-xpu20251-x86_64-linux/metadata.json DELETED Viewed

@@ -1,4 +0,0 @@
-{
-  "version": 1,
-  "python-depends": []
-}

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

@@ -1,26 +0,0 @@
-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/torch29-cxx11-cpu-x86_64-linux/__init__.py DELETED Viewed

@@ -1,27 +0,0 @@
-from . import layers
-from ._ops import ops
-def apply_rms_norm(input, weight, eps):
-    # ops.apply_rms_norm returns [output, rstd]
-    return ops.apply_rms_norm(
-            input,
-            weight,
-            eps,
-    )[0]
-def apply_rms_norm_backward(grad_output, input, weight, output, rstd, eps, input_requires_grad=True, weight_requires_grad=True):
-    return ops.apply_rms_norm_backward(
-            grad_output,
-            input,
-            weight,
-            output,
-            rstd,
-            eps,
-            input_requires_grad,
-            weight_requires_grad
-    )
-__all__ = ["layers", "apply_rms_norm_forward", "apply_rms_norm_backward"]

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

@@ -1,9 +0,0 @@
-import torch
-from . import _rmsnorm_cpu_b3d66c6
-ops = torch.ops._rmsnorm_cpu_b3d66c6
-def add_op_namespace_prefix(op_name: str):
-    """
-    Prefix op by namespace.
-    """
-    return f"_rmsnorm_cpu_b3d66c6::{op_name}"

build/torch29-cxx11-cpu-x86_64-linux/_rmsnorm_cpu_b3d66c6.abi3.so DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:cf3b3a68445d97357b4c08dd07ed0d197d18c9e7449ad62172dd55dfc49e7d08
-size 1999776

build/torch29-cxx11-cpu-x86_64-linux/layers.py DELETED Viewed

@@ -1,59 +0,0 @@
-import torch
-from ._ops import ops
-class RMSNormFunction(torch.autograd.Function):
-    @staticmethod
-    def forward(ctx, hidden_states, weight, variance_epsilon):
-        ctx.variance_epsilon = variance_epsilon
-        output, rstd = ops.apply_rms_norm(hidden_states, weight, variance_epsilon)
-        ctx.save_for_backward(hidden_states, weight, output, rstd)
-        return output
-    @staticmethod
-    def backward(ctx, grad_output):
-        hidden_states, weight, output, rstd = ctx.saved_tensors
-        grads = ops.apply_rms_norm_backward(
-            grad_output,
-            hidden_states,
-            weight,
-            output,
-            rstd,
-            ctx.variance_epsilon,
-            ctx.needs_input_grad[0],
-            ctx.needs_input_grad[1]
-        )
-        return grads[0], grads[1], None
-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 RMSNormFunction.apply(
-            hidden_states,
-            self.weight,
-            self.variance_epsilon,
-        )
-__all__ = ["RMSNorm"]

build/torch29-cxx11-cpu-x86_64-linux/metadata.json DELETED Viewed

@@ -1,5 +0,0 @@
-{
-  "version": 1,
-  "license": "Apache-2.0",
-  "python-depends": []
-}

build/torch29-cxx11-cpu-x86_64-linux/rmsnorm/__init__.py DELETED Viewed

@@ -1,26 +0,0 @@
-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/torch29-cxx11-xpu20252-x86_64-linux/__init__.py DELETED Viewed

@@ -1,27 +0,0 @@
-from . import layers
-from ._ops import ops
-def apply_rms_norm(input, weight, eps):
-    # ops.apply_rms_norm returns [output, rstd]
-    return ops.apply_rms_norm(
-            input,
-            weight,
-            eps,
-    )[0]
-def apply_rms_norm_backward(grad_output, input, weight, output, rstd, eps, input_requires_grad=True, weight_requires_grad=True):
-    return ops.apply_rms_norm_backward(
-            grad_output,
-            input,
-            weight,
-            output,
-            rstd,
-            eps,
-            input_requires_grad,
-            weight_requires_grad
-    )
-__all__ = ["layers", "apply_rms_norm_forward", "apply_rms_norm_backward"]

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

@@ -1,9 +0,0 @@
-import torch
-from . import _rmsnorm_xpu_cec90b8
-ops = torch.ops._rmsnorm_xpu_cec90b8
-def add_op_namespace_prefix(op_name: str):
-    """
-    Prefix op by namespace.
-    """
-    return f"_rmsnorm_xpu_cec90b8::{op_name}"

build/torch29-cxx11-xpu20252-x86_64-linux/_rmsnorm_xpu_cec90b8.abi3.so DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:b5fb0e95d3b6be17bd03833abcf461bb10d9c62fbf1336d9226dce0950dce1fa
-size 102179544

build/torch29-cxx11-xpu20252-x86_64-linux/layers.py DELETED Viewed

@@ -1,59 +0,0 @@
-import torch
-from ._ops import ops
-class RMSNormFunction(torch.autograd.Function):
-    @staticmethod
-    def forward(ctx, hidden_states, weight, variance_epsilon):
-        ctx.variance_epsilon = variance_epsilon
-        output, rstd = ops.apply_rms_norm(hidden_states, weight, variance_epsilon)
-        ctx.save_for_backward(hidden_states, weight, output, rstd)
-        return output
-    @staticmethod
-    def backward(ctx, grad_output):
-        hidden_states, weight, output, rstd = ctx.saved_tensors
-        grads = ops.apply_rms_norm_backward(
-            grad_output,
-            hidden_states,
-            weight,
-            output,
-            rstd,
-            ctx.variance_epsilon,
-            ctx.needs_input_grad[0],
-            ctx.needs_input_grad[1]
-        )
-        return grads[0], grads[1], None
-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 RMSNormFunction.apply(
-            hidden_states,
-            self.weight,
-            self.variance_epsilon,
-        )
-__all__ = ["RMSNorm"]

build/torch29-cxx11-xpu20252-x86_64-linux/metadata.json DELETED Viewed

@@ -1,8 +0,0 @@
-{
-  "version": 1,
-  "license": "Apache-2.0",
-  "python-depends": [],
-  "backend": {
-    "type": "xpu"
-  }
-}

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

@@ -1,26 +0,0 @@
-import ctypes
-import importlib.util
-import sys
-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/torch29-xpu20252-x86_64-windows/metadata.json DELETED Viewed

@@ -1,4 +0,0 @@
-{
-  "version": 1,
-  "python-depends": []
-}

build/torch29-xpu20252-x86_64-windows/rmsnorm/__init__.py DELETED Viewed

@@ -1,27 +0,0 @@
-from . import layers
-from ._ops import ops
-def apply_rms_norm(input, weight, eps):
-    # ops.apply_rms_norm returns [output, rstd]
-    return ops.apply_rms_norm(
-            input,
-            weight,
-            eps,
-    )[0]
-def apply_rms_norm_backward(grad_output, input, weight, output, rstd, eps, input_requires_grad=True, weight_requires_grad=True):
-    return ops.apply_rms_norm_backward(
-            grad_output,
-            input,
-            weight,
-            output,
-            rstd,
-            eps,
-            input_requires_grad,
-            weight_requires_grad
-    )
-__all__ = ["layers", "apply_rms_norm_forward", "apply_rms_norm_backward"]

build/torch29-xpu20252-x86_64-windows/rmsnorm/_ops.py DELETED Viewed

@@ -1,9 +0,0 @@
-import torch
-from . import _rmsnorm_96c9886
-ops = torch.ops._rmsnorm_96c9886
-def add_op_namespace_prefix(op_name: str):
-    """
-    Prefix op by namespace.
-    """
-    return f"_rmsnorm_96c9886::{op_name}"

build/torch29-xpu20252-x86_64-windows/rmsnorm/_rmsnorm_96c9886.pyd DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:c0cfb67260dcf293c71463a698f1531e9d86fb497f9dcf86c296d612ffa4c142
-size 2379264

build/torch29-xpu20252-x86_64-windows/rmsnorm/layers.py DELETED Viewed

@@ -1,59 +0,0 @@
-import torch
-from ._ops import ops
-class RMSNormFunction(torch.autograd.Function):
-    @staticmethod
-    def forward(ctx, hidden_states, weight, variance_epsilon):
-        ctx.variance_epsilon = variance_epsilon
-        output, rstd = ops.apply_rms_norm(hidden_states, weight, variance_epsilon)
-        ctx.save_for_backward(hidden_states, weight, output, rstd)
-        return output
-    @staticmethod
-    def backward(ctx, grad_output):
-        hidden_states, weight, output, rstd = ctx.saved_tensors
-        grads = ops.apply_rms_norm_backward(
-            grad_output,
-            hidden_states,
-            weight,
-            output,
-            rstd,
-            ctx.variance_epsilon,
-            ctx.needs_input_grad[0],
-            ctx.needs_input_grad[1]
-        )
-        return grads[0], grads[1], None
-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 RMSNormFunction.apply(
-            hidden_states,
-            self.weight,
-            self.variance_epsilon,
-        )
-__all__ = ["RMSNorm"]