Add Metal (Apple Silicon) build variants

.gitattributes

@@ -90,3 +90,5 @@ build/torch210-cxx11-cpu-x86_64-linux/_rmsnorm_cpu_b3d66c6.abi3.so filter=lfs di
 build/torch210-cxx11-xpu20253-x86_64-linux/_rmsnorm_xpu_b3d66c6.abi3.so filter=lfs diff=lfs merge=lfs -text
 build/torch29-cxx11-cpu-x86_64-linux/_rmsnorm_cpu_b3d66c6.abi3.so filter=lfs diff=lfs merge=lfs -text
 build/torch29-cxx11-xpu20252-x86_64-linux/_rmsnorm_xpu_b3d66c6.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch210-metal-aarch64-darwin/_fused_rms_norm_metal_feea0b8_dirty.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch29-metal-aarch64-darwin/_fused_rms_norm_metal_feea0b8_dirty.abi3.so filter=lfs diff=lfs merge=lfs -text

build/torch210-metal-aarch64-darwin/__init__.py

@@ -0,0 +1,8 @@
+from ._custom_ops import fused_add_rms_norm, rms_norm
+from ._ops import ops
+
+__all__ = [
+    "fused_add_rms_norm",
+    "ops",
+    "rms_norm",
+]

build/torch210-metal-aarch64-darwin/_custom_ops.py

@@ -0,0 +1,21 @@
+import torch
+
+from ._ops import ops
+
+
+def rms_norm(
+    out: torch.Tensor,
+    input: torch.Tensor,
+    weight: torch.Tensor,
+    epsilon: float,
+) -> None:
+    ops.rms_norm(out, input, weight, epsilon)
+
+
+def fused_add_rms_norm(
+    input: torch.Tensor,
+    residual: torch.Tensor,
+    weight: torch.Tensor,
+    epsilon: float,
+) -> None:
+    ops.fused_add_rms_norm(input, residual, weight, epsilon)

build/torch210-metal-aarch64-darwin/_fused_rms_norm_metal_feea0b8_dirty.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d5345b64749f1f73168207650def55ba33d51a79a0a8f7b623278b66de97a9a
+size 118520

build/torch210-metal-aarch64-darwin/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _fused_rms_norm_metal_feea0b8_dirty
+ops = torch.ops._fused_rms_norm_metal_feea0b8_dirty
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_fused_rms_norm_metal_feea0b8_dirty::{op_name}"

build/torch210-metal-aarch64-darwin/fused_rms_norm/__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-metal-aarch64-darwin/metadata.json

@@ -0,0 +1,3 @@
+{
+  "python-depends": []
+}

build/torch29-metal-aarch64-darwin/__init__.py

@@ -0,0 +1,8 @@
+from ._custom_ops import fused_add_rms_norm, rms_norm
+from ._ops import ops
+
+__all__ = [
+    "fused_add_rms_norm",
+    "ops",
+    "rms_norm",
+]

build/torch29-metal-aarch64-darwin/_custom_ops.py

@@ -0,0 +1,21 @@
+import torch
+
+from ._ops import ops
+
+
+def rms_norm(
+    out: torch.Tensor,
+    input: torch.Tensor,
+    weight: torch.Tensor,
+    epsilon: float,
+) -> None:
+    ops.rms_norm(out, input, weight, epsilon)
+
+
+def fused_add_rms_norm(
+    input: torch.Tensor,
+    residual: torch.Tensor,
+    weight: torch.Tensor,
+    epsilon: float,
+) -> None:
+    ops.fused_add_rms_norm(input, residual, weight, epsilon)

build/torch29-metal-aarch64-darwin/_fused_rms_norm_metal_feea0b8_dirty.abi3.so

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d19e27f2032ab9c814ec64815688ff5d7101aab790b8a2e4a70c99d99a2d7919
+size 117800

build/torch29-metal-aarch64-darwin/_ops.py

@@ -0,0 +1,9 @@
+import torch
+from . import _fused_rms_norm_metal_feea0b8_dirty
+ops = torch.ops._fused_rms_norm_metal_feea0b8_dirty
+
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_fused_rms_norm_metal_feea0b8_dirty::{op_name}"

build/torch29-metal-aarch64-darwin/fused_rms_norm/__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/torch29-metal-aarch64-darwin/metadata.json

@@ -0,0 +1,3 @@
+{
+  "python-depends": []
+}