Kernels:

kernels-community
/

layer-norm

Trusted publisher

Kernel card Files Files and versions

xet

Community

kernels-bot commited on 15 days ago

Commit

a15c370

verified ·

1 Parent(s): 4ac6ece

Uploaded using `kernel-builder`.

Browse files

Files changed (37) hide show

.gitattributes +9 -0
build/torch210-cxx11-cu126-aarch64-linux/{_layer_norm_cuda_86f75d9.abi3.so → _layer_norm_cuda_73ccd0c.abi3.so} +1 -1
build/torch210-cxx11-cu126-aarch64-linux/_ops.py +3 -3
build/torch210-cxx11-cu126-aarch64-linux/metadata.json +1 -1
build/torch210-cxx11-cu128-aarch64-linux/{_layer_norm_cuda_86f75d9.abi3.so → _layer_norm_cuda_73ccd0c.abi3.so} +1 -1
build/torch210-cxx11-cu128-aarch64-linux/_ops.py +3 -3
build/torch210-cxx11-cu128-aarch64-linux/metadata.json +1 -1
build/torch210-cxx11-cu130-aarch64-linux/{_layer_norm_cuda_86f75d9.abi3.so → _layer_norm_cuda_73ccd0c.abi3.so} +1 -1
build/torch210-cxx11-cu130-aarch64-linux/_ops.py +3 -3
build/torch210-cxx11-cu130-aarch64-linux/metadata.json +1 -1
build/torch211-cxx11-cu126-aarch64-linux/{_layer_norm_cuda_86f75d9.abi3.so → _layer_norm_cuda_73ccd0c.abi3.so} +1 -1
build/torch211-cxx11-cu126-aarch64-linux/_ops.py +3 -3
build/torch211-cxx11-cu126-aarch64-linux/metadata.json +1 -1
build/torch211-cxx11-cu128-aarch64-linux/{_layer_norm_cuda_86f75d9.abi3.so → _layer_norm_cuda_73ccd0c.abi3.so} +1 -1
build/torch211-cxx11-cu128-aarch64-linux/_ops.py +3 -3
build/torch211-cxx11-cu128-aarch64-linux/metadata.json +1 -1
build/torch211-cxx11-cu130-aarch64-linux/{_layer_norm_cuda_86f75d9.abi3.so → _layer_norm_cuda_73ccd0c.abi3.so} +1 -1
build/torch211-cxx11-cu130-aarch64-linux/_ops.py +3 -3
build/torch211-cxx11-cu130-aarch64-linux/metadata.json +1 -1
build/torch212-cxx11-cu126-aarch64-linux/__init__.py +26 -0
build/torch212-cxx11-cu126-aarch64-linux/_layer_norm_cuda_73ccd0c.abi3.so +3 -0
build/torch212-cxx11-cu126-aarch64-linux/_ops.py +9 -0
build/torch212-cxx11-cu126-aarch64-linux/layer_norm/__init__.py +26 -0
build/torch212-cxx11-cu126-aarch64-linux/layers.py +51 -0
build/torch212-cxx11-cu126-aarch64-linux/metadata.json +15 -0
build/torch212-cxx11-cu130-aarch64-linux/__init__.py +26 -0
build/torch212-cxx11-cu130-aarch64-linux/_layer_norm_cuda_73ccd0c.abi3.so +3 -0
build/torch212-cxx11-cu130-aarch64-linux/_ops.py +9 -0
build/torch212-cxx11-cu130-aarch64-linux/layer_norm/__init__.py +26 -0
build/torch212-cxx11-cu130-aarch64-linux/layers.py +51 -0
build/torch212-cxx11-cu130-aarch64-linux/metadata.json +17 -0
build/torch212-cxx11-cu132-aarch64-linux/__init__.py +26 -0
build/torch212-cxx11-cu132-aarch64-linux/_layer_norm_cuda_73ccd0c.abi3.so +3 -0
build/torch212-cxx11-cu132-aarch64-linux/_ops.py +9 -0
build/torch212-cxx11-cu132-aarch64-linux/layer_norm/__init__.py +26 -0
build/torch212-cxx11-cu132-aarch64-linux/layers.py +51 -0
build/torch212-cxx11-cu132-aarch64-linux/metadata.json +17 -0

.gitattributes CHANGED Viewed

@@ -87,3 +87,12 @@ build/torch210-cxx11-cu130-aarch64-linux/_layer_norm_cuda_86f75d9.abi3.so filter
 build/torch211-cxx11-cu126-aarch64-linux/_layer_norm_cuda_86f75d9.abi3.so filter=lfs diff=lfs merge=lfs -text
 build/torch211-cxx11-cu128-aarch64-linux/_layer_norm_cuda_86f75d9.abi3.so filter=lfs diff=lfs merge=lfs -text
 build/torch211-cxx11-cu130-aarch64-linux/_layer_norm_cuda_86f75d9.abi3.so filter=lfs diff=lfs merge=lfs -text

 build/torch211-cxx11-cu126-aarch64-linux/_layer_norm_cuda_86f75d9.abi3.so filter=lfs diff=lfs merge=lfs -text
 build/torch211-cxx11-cu128-aarch64-linux/_layer_norm_cuda_86f75d9.abi3.so filter=lfs diff=lfs merge=lfs -text
 build/torch211-cxx11-cu130-aarch64-linux/_layer_norm_cuda_86f75d9.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch210-cxx11-cu126-aarch64-linux/_layer_norm_cuda_73ccd0c.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch210-cxx11-cu128-aarch64-linux/_layer_norm_cuda_73ccd0c.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch210-cxx11-cu130-aarch64-linux/_layer_norm_cuda_73ccd0c.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch211-cxx11-cu126-aarch64-linux/_layer_norm_cuda_73ccd0c.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch211-cxx11-cu128-aarch64-linux/_layer_norm_cuda_73ccd0c.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch211-cxx11-cu130-aarch64-linux/_layer_norm_cuda_73ccd0c.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch212-cxx11-cu126-aarch64-linux/_layer_norm_cuda_73ccd0c.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch212-cxx11-cu130-aarch64-linux/_layer_norm_cuda_73ccd0c.abi3.so filter=lfs diff=lfs merge=lfs -text
+build/torch212-cxx11-cu132-aarch64-linux/_layer_norm_cuda_73ccd0c.abi3.so filter=lfs diff=lfs merge=lfs -text

build/torch210-cxx11-cu126-aarch64-linux/{_layer_norm_cuda_86f75d9.abi3.so → _layer_norm_cuda_73ccd0c.abi3.so} RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3630532f4059b550f0b5a9a1a1792a83d86aac981e483ea188abd1c9b86c2bf7
 size 711710472

 version https://git-lfs.github.com/spec/v1
+oid sha256:dc6cbb986c0ac1160fd75d6db0033b133350ede8d51c815cd6c821d7e2c512a1
 size 711710472

build/torch210-cxx11-cu126-aarch64-linux/_ops.py CHANGED Viewed

@@ -1,9 +1,9 @@
 import torch
-from . import _layer_norm_cuda_86f75d9
-ops = torch.ops._layer_norm_cuda_86f75d9
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_layer_norm_cuda_86f75d9::{op_name}"

 import torch
+from . import _layer_norm_cuda_73ccd0c
+ops = torch.ops._layer_norm_cuda_73ccd0c
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
+    return f"_layer_norm_cuda_73ccd0c::{op_name}"

build/torch210-cxx11-cu126-aarch64-linux/metadata.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "layer-norm",
-  "id": "_layer_norm_cuda_86f75d9",
   "version": 1,
   "license": "BSD-3-Clause",
   "python-depends": [],

 {
   "name": "layer-norm",
+  "id": "_layer_norm_cuda_73ccd0c",
   "version": 1,
   "license": "BSD-3-Clause",
   "python-depends": [],

build/torch210-cxx11-cu128-aarch64-linux/{_layer_norm_cuda_86f75d9.abi3.so → _layer_norm_cuda_73ccd0c.abi3.so} RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f1d2b68faa8308f8d7cedb97f32891d526867aff31c319d2d772985f28fdb11a
 size 1231083200

 version https://git-lfs.github.com/spec/v1
+oid sha256:b32a3daff6960337d42eb4b5484b2fc628e773616e3361bcce21d656d477096d
 size 1231083200

build/torch210-cxx11-cu128-aarch64-linux/_ops.py CHANGED Viewed

@@ -1,9 +1,9 @@
 import torch
-from . import _layer_norm_cuda_86f75d9
-ops = torch.ops._layer_norm_cuda_86f75d9
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_layer_norm_cuda_86f75d9::{op_name}"

 import torch
+from . import _layer_norm_cuda_73ccd0c
+ops = torch.ops._layer_norm_cuda_73ccd0c
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
+    return f"_layer_norm_cuda_73ccd0c::{op_name}"

build/torch210-cxx11-cu128-aarch64-linux/metadata.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "layer-norm",
-  "id": "_layer_norm_cuda_86f75d9",
   "version": 1,
   "license": "BSD-3-Clause",
   "python-depends": [],

 {
   "name": "layer-norm",
+  "id": "_layer_norm_cuda_73ccd0c",
   "version": 1,
   "license": "BSD-3-Clause",
   "python-depends": [],

build/torch210-cxx11-cu130-aarch64-linux/{_layer_norm_cuda_86f75d9.abi3.so → _layer_norm_cuda_73ccd0c.abi3.so} RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:587033c6e216fe050479a8d91017b2407db2e1fc0b9d7cd2dcdad15cc17e89f2
 size 1235994200

 version https://git-lfs.github.com/spec/v1
+oid sha256:f1be4d1ef49363641003ad084d112971753c08b1c8ce6755ce073d7e6fce171c
 size 1235994200

build/torch210-cxx11-cu130-aarch64-linux/_ops.py CHANGED Viewed

@@ -1,9 +1,9 @@
 import torch
-from . import _layer_norm_cuda_86f75d9
-ops = torch.ops._layer_norm_cuda_86f75d9
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_layer_norm_cuda_86f75d9::{op_name}"

 import torch
+from . import _layer_norm_cuda_73ccd0c
+ops = torch.ops._layer_norm_cuda_73ccd0c
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
+    return f"_layer_norm_cuda_73ccd0c::{op_name}"

build/torch210-cxx11-cu130-aarch64-linux/metadata.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "layer-norm",
-  "id": "_layer_norm_cuda_86f75d9",
   "version": 1,
   "license": "BSD-3-Clause",
   "python-depends": [],

 {
   "name": "layer-norm",
+  "id": "_layer_norm_cuda_73ccd0c",
   "version": 1,
   "license": "BSD-3-Clause",
   "python-depends": [],

build/torch211-cxx11-cu126-aarch64-linux/{_layer_norm_cuda_86f75d9.abi3.so → _layer_norm_cuda_73ccd0c.abi3.so} RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9128d9ff130842868f25b77fa50d748d39bfbc97894e87838bcfc03d94ca17f4
 size 711706784

 version https://git-lfs.github.com/spec/v1
+oid sha256:9ebbc069d60e09aea141d01bef3f1a14b81d315d2e944f78b99ee794a370f199
 size 711706784

build/torch211-cxx11-cu126-aarch64-linux/_ops.py CHANGED Viewed

@@ -1,9 +1,9 @@
 import torch
-from . import _layer_norm_cuda_86f75d9
-ops = torch.ops._layer_norm_cuda_86f75d9
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_layer_norm_cuda_86f75d9::{op_name}"

 import torch
+from . import _layer_norm_cuda_73ccd0c
+ops = torch.ops._layer_norm_cuda_73ccd0c
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
+    return f"_layer_norm_cuda_73ccd0c::{op_name}"

build/torch211-cxx11-cu126-aarch64-linux/metadata.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "layer-norm",
-  "id": "_layer_norm_cuda_86f75d9",
   "version": 1,
   "license": "BSD-3-Clause",
   "python-depends": [],

 {
   "name": "layer-norm",
+  "id": "_layer_norm_cuda_73ccd0c",
   "version": 1,
   "license": "BSD-3-Clause",
   "python-depends": [],

build/torch211-cxx11-cu128-aarch64-linux/{_layer_norm_cuda_86f75d9.abi3.so → _layer_norm_cuda_73ccd0c.abi3.so} RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da98484519347044a3c9a090d6d3f55c446592d8f36fbb1d7505478538f18853
 size 1231079512

 version https://git-lfs.github.com/spec/v1
+oid sha256:b2ca8ce06883b52cc6fc635f080a8a1e025a2cf87a9ad151df8525d2d1842c87
 size 1231079512

build/torch211-cxx11-cu128-aarch64-linux/_ops.py CHANGED Viewed

@@ -1,9 +1,9 @@
 import torch
-from . import _layer_norm_cuda_86f75d9
-ops = torch.ops._layer_norm_cuda_86f75d9
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_layer_norm_cuda_86f75d9::{op_name}"

 import torch
+from . import _layer_norm_cuda_73ccd0c
+ops = torch.ops._layer_norm_cuda_73ccd0c
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
+    return f"_layer_norm_cuda_73ccd0c::{op_name}"

build/torch211-cxx11-cu128-aarch64-linux/metadata.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "layer-norm",
-  "id": "_layer_norm_cuda_86f75d9",
   "version": 1,
   "license": "BSD-3-Clause",
   "python-depends": [],

 {
   "name": "layer-norm",
+  "id": "_layer_norm_cuda_73ccd0c",
   "version": 1,
   "license": "BSD-3-Clause",
   "python-depends": [],

build/torch211-cxx11-cu130-aarch64-linux/{_layer_norm_cuda_86f75d9.abi3.so → _layer_norm_cuda_73ccd0c.abi3.so} RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0d71263c674c3bfd317b033d5bea90d309e66d068164beeb6ceb037a3693b4c0
 size 1235990520

 version https://git-lfs.github.com/spec/v1
+oid sha256:d526240fb5ecde4f704be1bd671e339a0da2f7861af43d1b87535c920d841d81
 size 1235990520

build/torch211-cxx11-cu130-aarch64-linux/_ops.py CHANGED Viewed

@@ -1,9 +1,9 @@
 import torch
-from . import _layer_norm_cuda_86f75d9
-ops = torch.ops._layer_norm_cuda_86f75d9
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
-    return f"_layer_norm_cuda_86f75d9::{op_name}"

 import torch
+from . import _layer_norm_cuda_73ccd0c
+ops = torch.ops._layer_norm_cuda_73ccd0c
 def add_op_namespace_prefix(op_name: str):
     """
     Prefix op by namespace.
     """
+    return f"_layer_norm_cuda_73ccd0c::{op_name}"

build/torch211-cxx11-cu130-aarch64-linux/metadata.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "layer-norm",
-  "id": "_layer_norm_cuda_86f75d9",
   "version": 1,
   "license": "BSD-3-Clause",
   "python-depends": [],

 {
   "name": "layer-norm",
+  "id": "_layer_norm_cuda_73ccd0c",
   "version": 1,
   "license": "BSD-3-Clause",
   "python-depends": [],

build/torch212-cxx11-cu126-aarch64-linux/__init__.py ADDED Viewed

	@@ -0,0 +1,26 @@

+import torch
+import torch.nn as nn
+from ._ops import ops
+from . import layers
+def dropout_add_ln_fwd(input, gamma, beta, rowscale, colscale, x0_subset, z_subset, dropout_p, epsilon, rowscale_const, z_numrows, gen, residual_in_fp32, is_rms_norm):
+    return ops.dropout_add_ln_fwd(input, gamma, beta, rowscale, colscale, x0_subset, z_subset, dropout_p, epsilon, rowscale_const, z_numrows, gen, residual_in_fp32, is_rms_norm)
+def dropout_add_ln_bwd(dz, dx, x, mu, rsigma, gamma, rowscale, colscale, x0_subset, z_subset, dropout_p, rowscale_const, x0_numrows, has_residual, is_rms_norm):
+    return ops.dropout_add_ln_bwd(dz, dx, x, mu, rsigma, gamma, rowscale, colscale, x0_subset, z_subset, dropout_p, rowscale_const, x0_numrows, has_residual, is_rms_norm)
+def dropout_add_ln_parallel_residual_fwd(input, gamma0, beta0, gamma1, beta1, dropout_p, epsilon, gen, residual_in_fp32, is_rms_norm):
+    return ops.dropout_add_ln_parallel_residual_fwd(input, gamma0, beta0, gamma1, beta1, dropout_p, epsilon, gen, residual_in_fp32, is_rms_norm)
+def dropout_add_ln_parallel_residual_bwd(dz0, dz1, dx, x, mu, rsigma, gamma0, gamma1, dropout_p, has_x1, has_residual, is_rms_norm):
+    return ops.dropout_add_ln_parallel_residual_bwd(dz0, dz1, dx, x, mu, rsigma, gamma0, gamma1, dropout_p, has_x1, has_residual, is_rms_norm)
+__all__ = [
+    "layers",
+    "dropout_add_ln_fwd",
+    "dropout_add_ln_bwd",
+    "dropout_add_ln_parallel_residual_fwd",
+    "dropout_add_ln_parallel_residual_bwd",
+]

build/torch212-cxx11-cu126-aarch64-linux/_layer_norm_cuda_73ccd0c.abi3.so ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:4053b1e60f9aa2d0c36871fb49638a2644dee850719716382159abf3749e997e
+size 711709952

build/torch212-cxx11-cu126-aarch64-linux/_ops.py ADDED Viewed

	@@ -0,0 +1,9 @@

+import torch
+from . import _layer_norm_cuda_73ccd0c
+ops = torch.ops._layer_norm_cuda_73ccd0c
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_layer_norm_cuda_73ccd0c::{op_name}"

build/torch212-cxx11-cu126-aarch64-linux/layer_norm/__init__.py ADDED Viewed

	@@ -0,0 +1,26 @@

+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/torch212-cxx11-cu126-aarch64-linux/layers.py ADDED Viewed

	@@ -0,0 +1,51 @@

+import torch
+import torch.nn as nn
+from ._ops import ops
+class LayerNorm(nn.Module):
+    weight: torch.Tensor
+    variance_epsilon: float
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        output = ops.dropout_add_ln_fwd(
+            hidden_states.view(-1, hidden_states.shape[-1]),
+            gamma = self.weight,
+            beta = None,
+            rowscale = None,
+            colscale = None,
+            x0_subset = None,
+            z_subset = None,
+            dropout_p = 0,
+            epsilon = self.variance_epsilon,
+            rowscale_const = 1.0,
+            z_numrows = hidden_states.shape[1],
+            gen = None,
+            residual_in_fp32 = False,
+            is_rms_norm = False,
+        )
+        return output[0].view(hidden_states.shape)
+class LlamaRMSNorm(nn.Module):
+    weight: torch.Tensor
+    variance_epsilon: float
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        output = ops.dropout_add_ln_fwd(
+            hidden_states.view(-1, hidden_states.shape[-1]),
+            gamma = self.weight,
+            beta = None,
+            rowscale = None,
+            colscale = None,
+            x0_subset = None,
+            z_subset = None,
+            dropout_p = 0,
+            epsilon = self.variance_epsilon,
+            rowscale_const = 1.0,
+            z_numrows = hidden_states.shape[1],
+            gen = None,
+            residual_in_fp32 = False,
+            is_rms_norm = True,
+        )
+        return output[0].view(hidden_states.shape)

build/torch212-cxx11-cu126-aarch64-linux/metadata.json ADDED Viewed

	@@ -0,0 +1,15 @@

+{
+  "name": "layer-norm",
+  "id": "_layer_norm_cuda_73ccd0c",
+  "version": 1,
+  "license": "BSD-3-Clause",
+  "python-depends": [],
+  "backend": {
+    "type": "cuda",
+    "archs": [
+      "8.0",
+      "8.9",
+      "9.0"
+    ]
+  }
+}

build/torch212-cxx11-cu130-aarch64-linux/__init__.py ADDED Viewed

	@@ -0,0 +1,26 @@

+import torch
+import torch.nn as nn
+from ._ops import ops
+from . import layers
+def dropout_add_ln_fwd(input, gamma, beta, rowscale, colscale, x0_subset, z_subset, dropout_p, epsilon, rowscale_const, z_numrows, gen, residual_in_fp32, is_rms_norm):
+    return ops.dropout_add_ln_fwd(input, gamma, beta, rowscale, colscale, x0_subset, z_subset, dropout_p, epsilon, rowscale_const, z_numrows, gen, residual_in_fp32, is_rms_norm)
+def dropout_add_ln_bwd(dz, dx, x, mu, rsigma, gamma, rowscale, colscale, x0_subset, z_subset, dropout_p, rowscale_const, x0_numrows, has_residual, is_rms_norm):
+    return ops.dropout_add_ln_bwd(dz, dx, x, mu, rsigma, gamma, rowscale, colscale, x0_subset, z_subset, dropout_p, rowscale_const, x0_numrows, has_residual, is_rms_norm)
+def dropout_add_ln_parallel_residual_fwd(input, gamma0, beta0, gamma1, beta1, dropout_p, epsilon, gen, residual_in_fp32, is_rms_norm):
+    return ops.dropout_add_ln_parallel_residual_fwd(input, gamma0, beta0, gamma1, beta1, dropout_p, epsilon, gen, residual_in_fp32, is_rms_norm)
+def dropout_add_ln_parallel_residual_bwd(dz0, dz1, dx, x, mu, rsigma, gamma0, gamma1, dropout_p, has_x1, has_residual, is_rms_norm):
+    return ops.dropout_add_ln_parallel_residual_bwd(dz0, dz1, dx, x, mu, rsigma, gamma0, gamma1, dropout_p, has_x1, has_residual, is_rms_norm)
+__all__ = [
+    "layers",
+    "dropout_add_ln_fwd",
+    "dropout_add_ln_bwd",
+    "dropout_add_ln_parallel_residual_fwd",
+    "dropout_add_ln_parallel_residual_bwd",
+]

build/torch212-cxx11-cu130-aarch64-linux/_layer_norm_cuda_73ccd0c.abi3.so ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:df4db4bae71fab28e0e88250091baaad4b1359ce1cf5f74afb4eaaaf1b8feb03
+size 1235993656

build/torch212-cxx11-cu130-aarch64-linux/_ops.py ADDED Viewed

	@@ -0,0 +1,9 @@

+import torch
+from . import _layer_norm_cuda_73ccd0c
+ops = torch.ops._layer_norm_cuda_73ccd0c
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_layer_norm_cuda_73ccd0c::{op_name}"

build/torch212-cxx11-cu130-aarch64-linux/layer_norm/__init__.py ADDED Viewed

	@@ -0,0 +1,26 @@

+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/torch212-cxx11-cu130-aarch64-linux/layers.py ADDED Viewed

	@@ -0,0 +1,51 @@

+import torch
+import torch.nn as nn
+from ._ops import ops
+class LayerNorm(nn.Module):
+    weight: torch.Tensor
+    variance_epsilon: float
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        output = ops.dropout_add_ln_fwd(
+            hidden_states.view(-1, hidden_states.shape[-1]),
+            gamma = self.weight,
+            beta = None,
+            rowscale = None,
+            colscale = None,
+            x0_subset = None,
+            z_subset = None,
+            dropout_p = 0,
+            epsilon = self.variance_epsilon,
+            rowscale_const = 1.0,
+            z_numrows = hidden_states.shape[1],
+            gen = None,
+            residual_in_fp32 = False,
+            is_rms_norm = False,
+        )
+        return output[0].view(hidden_states.shape)
+class LlamaRMSNorm(nn.Module):
+    weight: torch.Tensor
+    variance_epsilon: float
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        output = ops.dropout_add_ln_fwd(
+            hidden_states.view(-1, hidden_states.shape[-1]),
+            gamma = self.weight,
+            beta = None,
+            rowscale = None,
+            colscale = None,
+            x0_subset = None,
+            z_subset = None,
+            dropout_p = 0,
+            epsilon = self.variance_epsilon,
+            rowscale_const = 1.0,
+            z_numrows = hidden_states.shape[1],
+            gen = None,
+            residual_in_fp32 = False,
+            is_rms_norm = True,
+        )
+        return output[0].view(hidden_states.shape)

build/torch212-cxx11-cu130-aarch64-linux/metadata.json ADDED Viewed

	@@ -0,0 +1,17 @@

+{
+  "name": "layer-norm",
+  "id": "_layer_norm_cuda_73ccd0c",
+  "version": 1,
+  "license": "BSD-3-Clause",
+  "python-depends": [],
+  "backend": {
+    "type": "cuda",
+    "archs": [
+      "10.0",
+      "12.0",
+      "8.0",
+      "8.9",
+      "9.0"
+    ]
+  }
+}

build/torch212-cxx11-cu132-aarch64-linux/__init__.py ADDED Viewed

	@@ -0,0 +1,26 @@

+import torch
+import torch.nn as nn
+from ._ops import ops
+from . import layers
+def dropout_add_ln_fwd(input, gamma, beta, rowscale, colscale, x0_subset, z_subset, dropout_p, epsilon, rowscale_const, z_numrows, gen, residual_in_fp32, is_rms_norm):
+    return ops.dropout_add_ln_fwd(input, gamma, beta, rowscale, colscale, x0_subset, z_subset, dropout_p, epsilon, rowscale_const, z_numrows, gen, residual_in_fp32, is_rms_norm)
+def dropout_add_ln_bwd(dz, dx, x, mu, rsigma, gamma, rowscale, colscale, x0_subset, z_subset, dropout_p, rowscale_const, x0_numrows, has_residual, is_rms_norm):
+    return ops.dropout_add_ln_bwd(dz, dx, x, mu, rsigma, gamma, rowscale, colscale, x0_subset, z_subset, dropout_p, rowscale_const, x0_numrows, has_residual, is_rms_norm)
+def dropout_add_ln_parallel_residual_fwd(input, gamma0, beta0, gamma1, beta1, dropout_p, epsilon, gen, residual_in_fp32, is_rms_norm):
+    return ops.dropout_add_ln_parallel_residual_fwd(input, gamma0, beta0, gamma1, beta1, dropout_p, epsilon, gen, residual_in_fp32, is_rms_norm)
+def dropout_add_ln_parallel_residual_bwd(dz0, dz1, dx, x, mu, rsigma, gamma0, gamma1, dropout_p, has_x1, has_residual, is_rms_norm):
+    return ops.dropout_add_ln_parallel_residual_bwd(dz0, dz1, dx, x, mu, rsigma, gamma0, gamma1, dropout_p, has_x1, has_residual, is_rms_norm)
+__all__ = [
+    "layers",
+    "dropout_add_ln_fwd",
+    "dropout_add_ln_bwd",
+    "dropout_add_ln_parallel_residual_fwd",
+    "dropout_add_ln_parallel_residual_bwd",
+]

build/torch212-cxx11-cu132-aarch64-linux/_layer_norm_cuda_73ccd0c.abi3.so ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:9e3e2f8cb06b921f94b179521d5f773a41fc5817c5667d3a405cf111f04feaf2
+size 1219551712

build/torch212-cxx11-cu132-aarch64-linux/_ops.py ADDED Viewed

	@@ -0,0 +1,9 @@

+import torch
+from . import _layer_norm_cuda_73ccd0c
+ops = torch.ops._layer_norm_cuda_73ccd0c
+def add_op_namespace_prefix(op_name: str):
+    """
+    Prefix op by namespace.
+    """
+    return f"_layer_norm_cuda_73ccd0c::{op_name}"

build/torch212-cxx11-cu132-aarch64-linux/layer_norm/__init__.py ADDED Viewed

	@@ -0,0 +1,26 @@

+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/torch212-cxx11-cu132-aarch64-linux/layers.py ADDED Viewed

	@@ -0,0 +1,51 @@

+import torch
+import torch.nn as nn
+from ._ops import ops
+class LayerNorm(nn.Module):
+    weight: torch.Tensor
+    variance_epsilon: float
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        output = ops.dropout_add_ln_fwd(
+            hidden_states.view(-1, hidden_states.shape[-1]),
+            gamma = self.weight,
+            beta = None,
+            rowscale = None,
+            colscale = None,
+            x0_subset = None,
+            z_subset = None,
+            dropout_p = 0,
+            epsilon = self.variance_epsilon,
+            rowscale_const = 1.0,
+            z_numrows = hidden_states.shape[1],
+            gen = None,
+            residual_in_fp32 = False,
+            is_rms_norm = False,
+        )
+        return output[0].view(hidden_states.shape)
+class LlamaRMSNorm(nn.Module):
+    weight: torch.Tensor
+    variance_epsilon: float
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        output = ops.dropout_add_ln_fwd(
+            hidden_states.view(-1, hidden_states.shape[-1]),
+            gamma = self.weight,
+            beta = None,
+            rowscale = None,
+            colscale = None,
+            x0_subset = None,
+            z_subset = None,
+            dropout_p = 0,
+            epsilon = self.variance_epsilon,
+            rowscale_const = 1.0,
+            z_numrows = hidden_states.shape[1],
+            gen = None,
+            residual_in_fp32 = False,
+            is_rms_norm = True,
+        )
+        return output[0].view(hidden_states.shape)

build/torch212-cxx11-cu132-aarch64-linux/metadata.json ADDED Viewed

	@@ -0,0 +1,17 @@

+{
+  "name": "layer-norm",
+  "id": "_layer_norm_cuda_73ccd0c",
+  "version": 1,
+  "license": "BSD-3-Clause",
+  "python-depends": [],
+  "backend": {
+    "type": "cuda",
+    "archs": [
+      "10.0",
+      "12.0",
+      "8.0",
+      "8.9",
+      "9.0"
+    ]
+  }
+}