diff --git a/.gitattributes b/.gitattributes index 568ff7ef239098618af6fb02e82813104af43e6b..6304df8bee7a61fd5a1eb66cfb1f9573d65f7b95 100644 --- a/.gitattributes +++ b/.gitattributes @@ -189,3 +189,21 @@ build/torch214-cu132-aarch64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs d build/torch215-cu130-aarch64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text build/torch215-cu132-aarch64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text build/torch215-cu134-aarch64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch210-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch210-cu128-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch210-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch211-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch211-cu128-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch211-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch212-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch212-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch212-cu132-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch213-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch213-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch213-cu132-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch214-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch214-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch214-cu132-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch215-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch215-cu132-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text +build/torch215-cu134-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so filter=lfs diff=lfs merge=lfs -text diff --git a/build/torch210-cu126-x86_64-linux/__init__.py b/build/torch210-cu126-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch210-cu126-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch210-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch210-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..04ae2fea6db609fe61e0b2d2973630049502fb05 --- /dev/null +++ b/build/torch210-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b41efa41f217e01a39b29d378cd7971e3438e170524a9dda7fc0be370c4bdbdf +size 710174400 diff --git a/build/torch210-cu126-x86_64-linux/_ops.py b/build/torch210-cu126-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch210-cu126-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch210-cu126-x86_64-linux/layers.py b/build/torch210-cu126-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch210-cu126-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch210-cu126-x86_64-linux/metadata.json b/build/torch210-cu126-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..e066522f495e1fc4d2f5b115b78ded7f33bda1dd --- /dev/null +++ b/build/torch210-cu126-x86_64-linux/metadata.json @@ -0,0 +1,38 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "tB76QfIX4Bo5sp03jNeXHjQ44XBSSp3af8C+NwxL298=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch210-cu126-x86_64-linux/metadata.json.sigstore b/build/torch210-cu126-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..5f5f823e2b20128c8ba9e671acbb904e25116b96 --- /dev/null +++ b/build/torch210-cu126-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"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"}, "tlogEntries":[{"logIndex":"3323910490", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429854", "inclusionPromise":{"signedEntryTimestamp":"MEUCIEx2HHK6efHffnNNvZ4t7HuKFAPBlCyh4wWPnd3iqZ/DAiEAkVIT71f+IaCQ+Kol6rFXlgANvNozeGvQt0YvCjY0Oy0="}, "inclusionProof":{"logIndex":"3202006228", "rootHash":"f3QiEGLmU2w3ZRmUxrNVS+02+a0eNYnPKbEFdqPaQ9c=", "treeSize":"3202006273", "hashes":["eT0iFaW1Z6Z+EAStVVpLteBJoRp6+Rr8qEiahJpHUnY=", "4fik5m0CLgFfpGjvPm7S7v/A+Ce1yYZmwDKKUXDYjmQ=", "4o1OTuN5SctgDJvRIXWlZYWc1qEmkfryFNBMYOSgaPE=", "VZkr9H/SYd+U31lCjb9BUHnSjbDHrTV5fRFwSsZ2Azo=", "Z1gi/dtiu2X6cUOydgrfMulQTFlJjHWbNI6NCykowZg=", "i200dFTcfIMzEChsXlC3RGQMfxAFSgHJ26soWfilDHI=", "658royWROErMI7L0cEzj81XVD3qIGPJkw7ZuVw3QUn4=", "FGjXvLSP9qEiwQ2lFVpuvnjCGrsiBpmm/3uhLsnNyzc=", "Vdu9ifsb+Qx6T2LY8i+l+QmRV7xI4wOC+bfFyDwIqNo=", "HNHkzoVPlf7PeYWJfmldl+iyKsPisuDob18VZbjG0gQ=", "WftaMH77XKBBrFMruQHHBU93T7033Ku32s/G5YGAB0U=", "tULoUiaZSfDmGtIHtVvxUCMXG2cjuSWTemce1ht0cGc=", "hxy8AssNYKl9eGTImYKqOoyG+UOhD95if4qQf6/guWY=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202006273\nf3QiEGLmU2w3ZRmUxrNVS+02+a0eNYnPKbEFdqPaQ9c=\n\n— rekor.sigstore.dev wNI9ajBEAiBPJSyj+Chg10m7RuaBnJFbCDuB1w56Yq9zLH59exXMQgIgZoXa4QCcnuTePTegFw5gIjRWwwf23teFa8zKGxJdoXE=\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"ZP/K8LRIhmUmpz6UzTgTMf/hSQG51V6+eZq5FKFwzxM="}, "signature":"MEUCIQDVxlbasLMDmP1o+XG1EX4S1R2nu2+glO6ohPvJwpJN5wIgEDydk/ZtQdzoX5vQl507EudnUNAFy0krntriQstS5c8="}} \ No newline at end of file diff --git a/build/torch210-cu126-x86_64-linux/symbols.json b/build/torch210-cu126-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch210-cu126-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch210-cu128-x86_64-linux/__init__.py b/build/torch210-cu128-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch210-cu128-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch210-cu128-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch210-cu128-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..d4e908f18520b672ebd7e5489a54a13788522cd2 --- /dev/null +++ b/build/torch210-cu128-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a460532301b1930ae79817e78e032c82c104531f6945f24d2a7b2d69a24d6ab +size 1229449320 diff --git a/build/torch210-cu128-x86_64-linux/_ops.py b/build/torch210-cu128-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch210-cu128-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch210-cu128-x86_64-linux/layers.py b/build/torch210-cu128-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch210-cu128-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch210-cu128-x86_64-linux/metadata.json b/build/torch210-cu128-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..b410f9ada933a06ca0084db9b4b248bea07827c8 --- /dev/null +++ b/build/torch210-cu128-x86_64-linux/metadata.json @@ -0,0 +1,40 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "10.0", + "12.0", + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "akYFMjAbGTCueYF+eOAyyCwQRTH2lF8k0qey1pok1qs=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch210-cu128-x86_64-linux/metadata.json.sigstore b/build/torch210-cu128-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..b09dfd0ccdaad322d4fc06c8441be9ceaf2b0123 --- /dev/null +++ b/build/torch210-cu128-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"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"}, "tlogEntries":[{"logIndex":"3323910691", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429855", "inclusionPromise":{"signedEntryTimestamp":"MEUCIQCAC5fjANQx1f13us0AFAd/rSAOtm9hiZd59SxF+y+rwQIgPaMMu54iMJPRNOPp59ootLwXqppLLv1iHN5884ACIIc="}, "inclusionProof":{"logIndex":"3202006429", "rootHash":"pJcQAyE3OMyOjshCYkKUi/EuBOothkOsxeGvrSaXKvs=", "treeSize":"3202006798", "hashes":["NYalebDdT2eC+8HqDmZNpAvzFUpXQirz3N8YXlhOCVw=", "XPXYX2SGbuDn3AxZcBFPSvqxEqNTYvqXFPCzkeRkB8A=", "W+aT7ATPNSD7FG6/GYE5QJY3x+i6+0RUgP3EORppQq0=", "V3UDQNYaxOd9+bPi6U2qKpEGJYhV3lnF4lNu2J6kVus=", "5ofPSD1QAzW/Y3uPnTVrQ7IC0EJvB1pTOY0CuXiFYNI=", "4iRu+HvOpbqq1vredJtuSP6SeEK1D1tCluufJ5NS9LA=", "039BIYeEIE0/ieeI0a17Z6lDpZpciHRHGCzLdNqeYbY=", "mzIKQi3HKkIYiSBEddNG7V5phjsxRtAeEs7/PJPyxfQ=", "8o0G5irp8mtd8zUxRl1imc4VkcX4pmpReOUEWaoHaW4=", "1V+eVZHK6aNsIhW9FaaWKo9adPzZtC/Nj3M/DA7ia/8=", "HNHkzoVPlf7PeYWJfmldl+iyKsPisuDob18VZbjG0gQ=", "WftaMH77XKBBrFMruQHHBU93T7033Ku32s/G5YGAB0U=", "tULoUiaZSfDmGtIHtVvxUCMXG2cjuSWTemce1ht0cGc=", "hxy8AssNYKl9eGTImYKqOoyG+UOhD95if4qQf6/guWY=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202006798\npJcQAyE3OMyOjshCYkKUi/EuBOothkOsxeGvrSaXKvs=\n\n— rekor.sigstore.dev wNI9ajBGAiEA8oHppmG0pZUYMCSnYUbaUE1oxVHvm9Cf29jAlo1Oem4CIQDqQv4LtOlGsnnoCBKqz+Vu9FgGI1qAgs7SvfAqrC5zjA==\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"AF+uS2Bj8fIzWIKcjYv9UZM9AgwXcQx7Lb7O/XxyKRo="}, "signature":"MEUCIQCz39gjQav2SXCT4RdYRPqEX4HB1eiKRg+SIpwtcbmLIAIgHMOXTARuwsy60EYvT4J3PG0MYutWJ/0jf7VcqCXuAB8="}} \ No newline at end of file diff --git a/build/torch210-cu128-x86_64-linux/symbols.json b/build/torch210-cu128-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch210-cu128-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch210-cu130-x86_64-linux/__init__.py b/build/torch210-cu130-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch210-cu130-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch210-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch210-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..2d770729b33f3547f93883275f6c810f7161c792 --- /dev/null +++ b/build/torch210-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ccf8424385ede384c61b34bca47da89606f73644590c53446551e81a5a91984e +size 1236411496 diff --git a/build/torch210-cu130-x86_64-linux/_ops.py b/build/torch210-cu130-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch210-cu130-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch210-cu130-x86_64-linux/layers.py b/build/torch210-cu130-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch210-cu130-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch210-cu130-x86_64-linux/metadata.json b/build/torch210-cu130-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..196b6392cc02bf4e3e0a0c251c03e80bbeb9ea97 --- /dev/null +++ b/build/torch210-cu130-x86_64-linux/metadata.json @@ -0,0 +1,40 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "10.0", + "12.0", + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "zPhCQ4Xt44TGGzS8pH2olgb3NkRZDFNEZVHoGlqRmE4=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch210-cu130-x86_64-linux/metadata.json.sigstore b/build/torch210-cu130-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..0b224c16d378939fbcd42495f97e97ee9891ef58 --- /dev/null +++ b/build/torch210-cu130-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"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"}, "tlogEntries":[{"logIndex":"3323911488", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429858", "inclusionPromise":{"signedEntryTimestamp":"MEUCIQDSLeBlQTnH41BH7zgkNkpNLy6md+UKGxDd7ASbd+ArEwIgR0XkZ0WXs6doKIbheYtcACe7vZ63venayx5rUvp2KsA="}, "inclusionProof":{"logIndex":"3202007226", "rootHash":"dz7Fj3Ak8QuYkxZYrn/c/Pal7HHCEJ2ElhOVbGEeYzs=", "treeSize":"3202007497", "hashes":["ayPoWoI8Wcvqt9NEyEEJzhMdgtVMNPyK/g1WeIFfxKg=", "I/O6NTaFnrOmvZx8fbZMxJuMC1lqUAI7j4Iu7CJmPhw=", "QLnKMXWCz9+i0oERCQsAge4Klc6f/fB4j8QveP7scvI=", "bsAIGae7PvbU2cakgLaKHkMbv505nfjyzbP5LegoSgI=", "genI/3eDN8dNM0y4IxmdvmH+fniAun5OZpY4KS63wwQ=", "xzAnJWo3BbIZ6RTTJKRdp3wKengh+RjXB8D6bzl4t+o=", "MMeBaEJtXBjvF/3n5hA6d+6hWCjgc7qb2Sm0tAoypVM=", "om9p9uF3qehFPYvnVLzgALCojJNaWfACWdYZaTLtYCU=", "I0XbTC6puzh8Zo67FfePqxYXSoSEAmAUxSCcWYA8wdM=", "ZoEKTEi+vNH82Rv/iMhp5QJLlu/Ocy2iHNm45P+eFP8=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202007497\ndz7Fj3Ak8QuYkxZYrn/c/Pal7HHCEJ2ElhOVbGEeYzs=\n\n— rekor.sigstore.dev wNI9ajBFAiEA0Ji+Is74M3N9zMRo6S7fFBKrpo/NxAl9NHxdnCMuVUYCIHQXOT2fGL9dHf9mFNE9gyqgW/bt3v7v4bSA5qpR33xm\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"z0jkCMO3buowvFFOkHtDaOor7NOt+mn5qgX9z2/k2nw="}, "signature":"MEUCIQCBzSudyjieI/pvZSTWxnGYw52pRczayyb7hGU+9blarQIgXd4kR9rlYOgLB4TzYzSQHs8whV2MaXC/rAeuQrxxR0w="}} \ No newline at end of file diff --git a/build/torch210-cu130-x86_64-linux/symbols.json b/build/torch210-cu130-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch210-cu130-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch211-cu126-x86_64-linux/__init__.py b/build/torch211-cu126-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch211-cu126-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch211-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch211-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..312359d18b0af48f043601259b16c8a7f08e8825 --- /dev/null +++ b/build/torch211-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5c076aaa1995e80647dcf3b0902a6fa104a48aea88007562974f2879c0bfb9ba +size 710163352 diff --git a/build/torch211-cu126-x86_64-linux/_ops.py b/build/torch211-cu126-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch211-cu126-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch211-cu126-x86_64-linux/layers.py b/build/torch211-cu126-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch211-cu126-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch211-cu126-x86_64-linux/metadata.json b/build/torch211-cu126-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..52d870b0043d2c7ca231e4a2d8933eebff965c92 --- /dev/null +++ b/build/torch211-cu126-x86_64-linux/metadata.json @@ -0,0 +1,38 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "XAdqqhmV6AZH3POwkCpvoQSkiuqIAHVil08oecC/ubo=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch211-cu126-x86_64-linux/metadata.json.sigstore b/build/torch211-cu126-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..11ed0f87260ac888a70041efe0b365d3118b5dd9 --- /dev/null +++ b/build/torch211-cu126-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"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"}, "tlogEntries":[{"logIndex":"3323912079", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429861", "inclusionPromise":{"signedEntryTimestamp":"MEQCIAYr0QLPDriv56jlouLgFeuQaxM0qPy+O5QQ9cA7wLDUAiBGmq1xwNWcVPQqzpLprhZBp9G5LgFHoYlrS2UQgjjo4A=="}, "inclusionProof":{"logIndex":"3202007817", "rootHash":"qoZWPfMiAztyCSmWQZDDhvZlsdq8c8pMYqhPvpYD64o=", "treeSize":"3202007965", "hashes":["j084dzqaXnXMIno4NslW1GncHx+s9w/PIAiD2DAJ9YY=", "crmQMVDeSna0CiFpCbN/kvk5uJe+JLKeV+LIEgYb+1A=", "P4FXxnhf6Cc7BAiTeawUSKlxl1UfpuVSVZn0o+2/8+I=", "NIUQFxjH3LEp/J8CPMpXE/2t8B2JkQHNX6jQ9iP/ArU=", "OkKCBL4qU9XoEgiDEXZfM8gXTg3T3GSEhvCIRmCk7Rc=", "hoqXnN+ieXI/iEX/r+yjvKUAc9rKKQ6hWfI22r7s8oQ=", "Q1myo2aSOkYvT9nsu2T2k/VpKnPs4liWgMktv7UEPXA=", "NTCyraBjbxJ7lBp4+y7mNFqeq+63G4EinS7a9gEhr5E=", "vNXZIDIN92DuTOoCNe6sd4l4PwAhMwyiCwr+yuDvfPA=", "TSVQvVv+3shZvGiCNI8HX2P3GT9fmKjN/alkJsMcML4=", "ZoEKTEi+vNH82Rv/iMhp5QJLlu/Ocy2iHNm45P+eFP8=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202007965\nqoZWPfMiAztyCSmWQZDDhvZlsdq8c8pMYqhPvpYD64o=\n\n— rekor.sigstore.dev wNI9ajBFAiEAqn/ItMqwK/iuJdNruWPmiWlben5fP0nfcIhXuXbWkzACIH8CRGK3vlIQBuqPMwIB2XcHGOlVIjFUo8jOtZsnw27K\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"d49fDyYWZfR0EdUSseXfeVGWVeqoScKkDyTsgX0HRx0="}, "signature":"MEUCIFZYPhr8R41AHaJLsiVIJ6ltqCMXTbaqqC3PH2jNXUAZAiEApv5pDCv5nlo0A5H7D6EWz3dgen5PSnVP02dGrxbk5Qo="}} \ No newline at end of file diff --git a/build/torch211-cu126-x86_64-linux/symbols.json b/build/torch211-cu126-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch211-cu126-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch211-cu128-x86_64-linux/__init__.py b/build/torch211-cu128-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch211-cu128-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch211-cu128-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch211-cu128-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..2370edc2c30784d0b8c33153729327702805a576 --- /dev/null +++ b/build/torch211-cu128-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8ed8a7cc5dee325388092a1832c6d31ad81dd8962d3e0b4daeadd869d24e6741 +size 1229438296 diff --git a/build/torch211-cu128-x86_64-linux/_ops.py b/build/torch211-cu128-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch211-cu128-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch211-cu128-x86_64-linux/layers.py b/build/torch211-cu128-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch211-cu128-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch211-cu128-x86_64-linux/metadata.json b/build/torch211-cu128-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..e930eb84ce4eee7df0a89ad2590674be4583261d --- /dev/null +++ b/build/torch211-cu128-x86_64-linux/metadata.json @@ -0,0 +1,40 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "10.0", + "12.0", + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "jtinzF3uMlOICSoYMsbTGtgd2JYtPgtNrq3YadJOZ0E=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch211-cu128-x86_64-linux/metadata.json.sigstore b/build/torch211-cu128-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..369ec02dc1cf82ce71cff8be2fb3e58606884bfb --- /dev/null +++ b/build/torch211-cu128-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"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"}, "tlogEntries":[{"logIndex":"3323912490", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429864", "inclusionPromise":{"signedEntryTimestamp":"MEUCIGlxCUt8NFtHDq0pg4DtBL/QbeQduXDLHCx2BSKVCtT9AiEA0jo14yXezJ7/HCT/n85f0+p8gChbp2PXsqJSYZgOWts="}, "inclusionProof":{"logIndex":"3202008228", "rootHash":"KDyfs9mdA3MS3pE7gntXPlcxtYJ05ZnqXYHC/nDx5SA=", "treeSize":"3202008277", "hashes":["kZQVIn9P+0UeChAKlr3vjZHNBx8tE7nVYfTvoduf4fA=", "2SCIJaQKaep4BpHehVWsz1mkTPsPPns6Zg50EGX59PY=", "3uNSWW1NPJEJd9g842iQgARh3j8ivGvFhfH31XfOjQY=", "9vy+cjJoG8ivp9Mb+6W8fYy/nZF9cJhqh+uaN2UfGo8=", "IgmPb66+vznm6GwShjp9BamARIukGpjYqvDUiv1ifdE=", "hgSMJ95g2LO8Q8uO8G5kN98EM0yyG8NUKwPbvvKwrOo=", "cha2vyFRSiq+4nCIZXKuFmrutUsvx6ZWdRjiX9UIAbA=", "BZPDUTn+BXiOtwXcB67BFwzimJ5P3VIex9nt0jlS1qM=", "nUjKQXfNArSNYDW3bx7hl44Y3U3GE0LwbQsIG0+tL70=", "ZoEKTEi+vNH82Rv/iMhp5QJLlu/Ocy2iHNm45P+eFP8=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202008277\nKDyfs9mdA3MS3pE7gntXPlcxtYJ05ZnqXYHC/nDx5SA=\n\n— rekor.sigstore.dev wNI9ajBFAiEAyT0+W02UqMALd+8nf7mMuoC9DpZVu7X7qbXJlQgfjdgCIEDPmOosjyePd9nxFwhr9GuDg1qqpP0e3aZ2xvUBSFrB\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"7Q0qEc53zvJ5QtVeLooqEPNNHIPG1nK3ccou368qY9Y="}, "signature":"MEYCIQCG35Xo2txnVTp/AZp9sldGn0LxTsNhajAcRS+qKEDg4wIhAKfkCit5eqUnDiOXbD0oFidcdu/VDBX1XcMW2TO+PkVB"}} \ No newline at end of file diff --git a/build/torch211-cu128-x86_64-linux/symbols.json b/build/torch211-cu128-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch211-cu128-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch211-cu130-x86_64-linux/__init__.py b/build/torch211-cu130-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch211-cu130-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch211-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch211-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..bc151664d1eb317c004225e13c2cf9e3ff71b3d3 --- /dev/null +++ b/build/torch211-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:103c7a60d592b794a5d1004307e686c07bbe6112dbf505860ed7534c709432aa +size 1236404560 diff --git a/build/torch211-cu130-x86_64-linux/_ops.py b/build/torch211-cu130-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch211-cu130-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch211-cu130-x86_64-linux/layers.py b/build/torch211-cu130-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch211-cu130-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch211-cu130-x86_64-linux/metadata.json b/build/torch211-cu130-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..54466784ecff080ffb636d974a00def429c38770 --- /dev/null +++ b/build/torch211-cu130-x86_64-linux/metadata.json @@ -0,0 +1,40 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "10.0", + "12.0", + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "EDx6YNWSt5Sl0QBDB+aGwHu+YRLb9QWGDtdTTHCUMqo=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch211-cu130-x86_64-linux/metadata.json.sigstore b/build/torch211-cu130-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..8418d5dedf562cfcf2bb28b6fb80286dbf0902d4 --- /dev/null +++ b/build/torch211-cu130-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"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"}, "tlogEntries":[{"logIndex":"3323912836", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429866", "inclusionPromise":{"signedEntryTimestamp":"MEYCIQD+zMkCD7mDh6uI9w+IteqDiURpf/z3LFUJStLbbf60UgIhAM6kBQhWhZAYlrT0Cd9bb7okRdX9/c6pFXh1dJJGMWhZ"}, "inclusionProof":{"logIndex":"3202008574", "rootHash":"077aofAE87FoyN+CSHcBdXsk22EbrS9KFrXW0bLneH4=", "treeSize":"3202008596", "hashes":["YnLi1OSwEE91Te8M0LLq1UjohSDGQnghmeckan0FVrs=", "skaWWqxE+GA1O5I5jFOxP5teWnhhRFjTdyUBCrrxoZc=", "7hdaEZ5TNZ5mK2indRTL4VzlTGU/q5gBFeh4F5CsHFo=", "jCu4ozARG8CXQNkaOgNTpW8G0EqaLKIeoqg3pl0u81c=", "PvAavV8U8ZRBrvKE4MucdqJ4J5DJLwmSfDZ/uVApKjo=", "Wkg7T9H/f69Yxes08gOBrnz3TzTPpBCUjRDQJ2qWXQo=", "4tJVN7N+y7V4Y5f1AzumrqFFvjCn4B0JT5CZQCpWl5Y=", "1U1z2hPMXcTqVjtbqbFGy2XK5hLyvE3ti3l3OkiCDts=", "UtfSLFsyV/BplzV4TWwngkp48e8ZJHdvE6Y2P80iWf0=", "2aXDmnNDvhkYeoKlUA0+L8Vn/R9Lc2+v6K9TpwGGQDg=", "nUjKQXfNArSNYDW3bx7hl44Y3U3GE0LwbQsIG0+tL70=", "ZoEKTEi+vNH82Rv/iMhp5QJLlu/Ocy2iHNm45P+eFP8=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202008596\n077aofAE87FoyN+CSHcBdXsk22EbrS9KFrXW0bLneH4=\n\n— rekor.sigstore.dev wNI9ajBGAiEAh1eteXkeVsbu90orL4mqGl5+Ej4UC3kuLOf1zplpHVsCIQC9Wu8n2kx1r2lpBkdTF3iYeUSi/qoL851bXPWXqM9PsQ==\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"718DJL75v17lcIW2vEf8W5psuXtrZl6LVqNn8XJJXqU="}, "signature":"MEYCIQDAW0i0QARzw5dB7lH0XHEuUCepyKIpH9dzk084aWrOgQIhAKfZyoG+33KQOIgZHuR+qf2gb/7+5WIAyPFGE+IyZVuX"}} \ No newline at end of file diff --git a/build/torch211-cu130-x86_64-linux/symbols.json b/build/torch211-cu130-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch211-cu130-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch212-cu126-x86_64-linux/__init__.py b/build/torch212-cu126-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch212-cu126-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch212-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch212-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..fe50a47ab3f4fd62a4bd6713b74fc7ae09485b44 --- /dev/null +++ b/build/torch212-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2d766ff0788001b308bafafece2fca33c3d27c022dcfc025581a6309080d411e +size 710169464 diff --git a/build/torch212-cu126-x86_64-linux/_ops.py b/build/torch212-cu126-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch212-cu126-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch212-cu126-x86_64-linux/layers.py b/build/torch212-cu126-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch212-cu126-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch212-cu126-x86_64-linux/metadata.json b/build/torch212-cu126-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..0577946d2b6d38e8c9e371b09e3339dd80b20554 --- /dev/null +++ b/build/torch212-cu126-x86_64-linux/metadata.json @@ -0,0 +1,38 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "LXZv8HiAAbMIuvr+zi/KM8PSfAItz8AlWBpjCQgNQR4=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch212-cu126-x86_64-linux/metadata.json.sigstore b/build/torch212-cu126-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..fd99b2e0bdf3234244af50a92055a1c727c20157 --- /dev/null +++ b/build/torch212-cu126-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"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"}, "tlogEntries":[{"logIndex":"3323913046", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429867", "inclusionPromise":{"signedEntryTimestamp":"MEYCIQD4FUjh7vcZngZqENFxctfzdLe7CZcvx3DdIeevkuSBIgIhAM3VEDabJEwwcVbHC43gqQpsbECl3HYS+FGE1pT1F+5f"}, "inclusionProof":{"logIndex":"3202008784", "rootHash":"9MFkmsDsxSsyRSyzM1ySVmNJuyjNDDVGkkkhCkZ8Yu4=", "treeSize":"3202008925", "hashes":["EfJpYgTkYL00MbtZ7Z7rcrF/Xnn0qn5LYZkTgA0uSVk=", "Y0xub18piJEJBhB77/CvVFEeq+VVPHSZNgiIVP8O7XA=", "FWGkWbROAaN9yxS+0UQqldgdzKU7CzBpAi8EErM7b0Q=", "iCdeUUPKUW8nWXOJiBumvmmVc5VDDjhkC+ZaMng2JK4=", "SHD/GkdoGj1g+HiJzL0ZFJgKc9OLbPCOsGeB359uQKg=", "0pOJcL1bN1cddqGIiR5UFe0vCwZ9+o0sbFF7T8uytd4=", "AZM9mb9X1VLaDYTHkeID/jWjlEdZSehzbNx8oT0rLIk=", "CrRAqHSkwZeM7gswNDjWw09K9RE0jhZluoBV2shhH+o=", "wJe3nfRKhUq9bsyxM9uKtrFO4BNIioKbkeeWj5WDr6o=", "Uq86hL9+1310xS3t28/TASNui3/8OgPnZwAAw1GmyOk=", "nUjKQXfNArSNYDW3bx7hl44Y3U3GE0LwbQsIG0+tL70=", "ZoEKTEi+vNH82Rv/iMhp5QJLlu/Ocy2iHNm45P+eFP8=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202008925\n9MFkmsDsxSsyRSyzM1ySVmNJuyjNDDVGkkkhCkZ8Yu4=\n\n— rekor.sigstore.dev wNI9ajBEAiAtTCFUMU73Vw7Wc4J7mCJLybP7nRYG2Eo8GO59/XhlSQIgHfEGjjW+4Oszx14+K7rS1ToPD3/IAQNbOxcO0Y8u+kk=\n"}}, "canonicalizedBody":"eyJhcGlWZXJzaW9uIjoiMC4wLjEiLCJraW5kIjoiaGFzaGVkcmVrb3JkIiwic3BlYyI6eyJkYXRhIjp7Imhhc2giOnsiYWxnb3JpdGhtIjoic2hhMjU2IiwidmFsdWUiOiI1NDc2NTZlMjA1NGMyZDVmODA1MmE4YzA0MDAxNDAzNDQyYmM4NDRmOTM3ZmQyMGNlY2RiODg3Y2U5ODI2ODk0In19LCJzaWduYXR1cmUiOnsiY29udGVudCI6Ik1FUUNJQWltVm1VdmpMMXFqVWV4QWh5dzlGdFpmSVJ2OHdTcVhCSk10dzhxTW4ySUFpQUpqVFBrajVmSVVTNUllQnUyUjljQUxmVnRDeFdvam03cFhUZFpoOS8rWnc9PSIsInB1YmxpY0tleSI6eyJjb250ZW50IjoiTFMwdExTMUNSVWRKVGlCRFJWSlVTVVpKUTBGVVJTMHRMUzB0Q2sxSlNVaE1SRU5EUW5KTFowRjNTVUpCWjBsVllWTk5Sa1JsU0d4WGQxUmxMekpEV1ZST2FUQjJOMDQ1Ulcwd2QwTm5XVWxMYjFwSmVtb3dSVUYzVFhjS1RucEZWazFDVFVkQk1WVkZRMmhOVFdNeWJHNWpNMUoyWTIxVmRWcEhWakpOVWpSM1NFRlpSRlpSVVVSRmVGWjZZVmRrZW1SSE9YbGFVekZ3WW01U2JBcGpiVEZzV2tkc2FHUkhWWGRJYUdOT1RXcFplRTFFUlRWTlZHTjRUVlJCTWxkb1kwNU5hbGw0VFVSRk5VMVVZM2xOVkVFeVYycEJRVTFHYTNkRmQxbElDa3R2V2tsNmFqQkRRVkZaU1V0dldrbDZhakJFUVZGalJGRm5RVVZyYTI5QlNuSlJlbU55YUU1NVFrZE9ibE5GYlZKS1MxZGFSM2hvTWtjdlFtVk9lRE1LU1ZneU1HOUJWRnAzWlVSRldFMDRVVEJJVUhWUU1raFRZVVZ0WWk5UFlXaGxRVTh4SzIxSFl6WkRRbXhTVkRsWlZVdFBRMEprUlhkbloxaE9UVUUwUndwQk1WVmtSSGRGUWk5M1VVVkJkMGxJWjBSQlZFSm5UbFpJVTFWRlJFUkJTMEpuWjNKQ1owVkdRbEZqUkVGNlFXWkNaMDVXU0ZOTlJVZEVRVmRuUWxSbUNqQXJibEJXYVZGU2JIWnRiekpQYTI5V1lVeEhUR2hvYTFCNlFuSkNaMDVXU0ZKRlFrRm1PRVZaVkVKbWFHd3hiMlJJVW5kamVtOTJUREprY0dSSGFERUtXV2sxYW1JeU1IWmhTRlp1V2pKc2RWb3lXbWhaTWxWMllUSldlV0p0Vm5OamVURnFZakl4ZEdSWE5YQmtTR3QyVEcxa2NHUkhhREZaYVRrellqTktjZ3BhYlhoMlpETk5kbGx1Vm5CaVIxRjFaVmRHZEdKRlFubGFWMXA2VERKb2JGbFhVbnBNTWpGb1lWYzBkMDlSV1V0TGQxbENRa0ZIUkhaNlFVSkJVVkZ5Q21GSVVqQmpTRTAyVEhrNU1HSXlkR3hpYVRWb1dUTlNjR0l5TlhwTWJXUndaRWRvTVZsdVZucGFXRXBxWWpJMU1GcFhOVEJNYlU1MllsUkJaa0puYjNJS1FtZEZSVUZaVHk5TlFVVkRRa0pHTTJJelNuSmFiWGgyWkRFNWEyRllUbmRaV0ZKcVlVUkJNa0puYjNKQ1owVkZRVmxQTDAxQlJVUkNRMmh0V2xSS2FRcFphbGt5V20xV2JFNXRUWGxaVkdNeldsZE9hbHBxUVRSWlZFRXhUVVJvYTFsNlZUTk9SRWswV2tSUk5FOUhWbXROUWsxSFEybHpSMEZSVVVKbk56aDNDa0ZSVVVWQ1ZVb3hZVmQ0YTAxRGMwZERhWE5IUVZGUlFtYzNPSGRCVVZWRlNGZG9NVm95WkhCaWJXUnRXVmRPYkV3eWRHeGpiVFZzWWtoTmRGa3lPWFFLWWxoV2RXRllValZOUWpCSFEybHpSMEZSVVVKbk56aDNRVkZaUlVRelNteGFiazEyWVVkV2FGcElUWFppVjBad1ltcEJOMEpuYjNKQ1owVkZRVmxQTHdwTlFVVkpRa013VFVzeWFEQmtTRUo2VDJrNGRtUkhPWEphVnpSMVdWZE9NR0ZYT1hWamVUVnVZVmhTYjJSWFNqRmpNbFo1V1RJNWRXUkhWblZrUXpWcUNtSXlNSGRpVVZsTFMzZFpRa0pCUjBSMmVrRkNRMUZTWmtSR01XOWtTRkozWTNwdmRrd3laSEJrUjJneFdXazFhbUl5TUhaaFNGWnVXakpzZFZveVdtZ0tXVEpWZG1FeVZubGliVlp6WTNreGFtSXlNWFJrVnpWd1pFaHJka3h0WkhCa1IyZ3hXV2s1TTJJelNuSmFiWGgyWkROTmRsbHVWbkJpUjFGMVpWZEdkQXBpUlVKNVdsZGFla3d5YUd4WlYxSjZUREl4YUdGWE5IZFBRVmxMUzNkWlFrSkJSMFIyZWtGQ1EyZFJjVVJEYUcxYVZFcHBXV3BaTWxwdFZteE9iVTE1Q2xsVVl6TmFWMDVxV21wQk5GbFVRVEZOUkdocldYcFZNMDVFU1RSYVJGRTBUMGRXYTAxQ2MwZERhWE5IUVZGUlFtYzNPSGRCVVhORlJGRjNUR015Vm5NS1dta3hiMkl6VGpCYVYxRjNVVUZaUzB0M1dVSkNRVWRFZG5wQlFrUkJVWGxFUkVKdlpFaFNkMk42YjNaTU1tUndaRWRvTVZscE5XcGlNakIyWVVoV2JncGFNbXgxV2pKYWFGa3lWWFpoTWxaNVltMVdjMk41TVdwaU1qRjBaRmMxY0dSSWEzZFBRVmxMUzNkWlFrSkJSMFIyZWtGQ1JGRlJjVVJEYUcxYVZFcHBDbGxxV1RKYWJWWnNUbTFOZVZsVVl6TmFWMDVxV21wQk5GbFVRVEZOUkdocldYcFZNMDVFU1RSYVJGRTBUMGRXYTAxQ09FZERhWE5IUVZGUlFtYzNPSGNLUVZFMFJVVlJkMUJqYlZadFkzazViMXBYUm10amVUbDBXVmRzZFUxQ2IwZERhWE5IUVZGUlFtYzNPSGRCVVRoRlJFRjNTMDFVUVROTlZGRXpUbFJWZVFwUFZFRjFRbWR2Y2tKblJVVkJXVTh2VFVGRlVVSkRRVTFJYldnd1pFaENlazlwT0haYU1td3dZVWhXYVV4dFRuWmlVemx2WkZka2JtRlhOVzVhYlVacUNscFVRVmxDWjI5eVFtZEZSVUZaVHk5TlFVVlNRa0Z2VFVORVNURk9la2wzVG5wUmVrMUhNRWREYVhOSFFWRlJRbWMzT0hkQlVrbEZXSGQ0WkdGSVVqQUtZMGhOTmt4NU9XNWhXRkp2WkZkSmRWa3lPWFJNTW1neFdqSmtjR0p0WkcxWlYwNXNUREowYkdOdE5XeGlTRTEwV1RJNWRHSllWblZoV0ZJMVRIazFiZ3BoV0ZKdlpGZEpkbVF5T1hsaE1scHpZak5rZWt3eVNqRmhWM2hyVEc1c2FHSlhlRUZqYlZadFkzazViMXBYUm10amVUbDBXVmRzZFUxRVowZERhWE5IQ2tGUlVVSm5OemgzUVZKTlJVdG5kMjlhYlZWNVdXMUpNazV0V214YVZGcHFUVzFGTTA0eVZtcFpNbGwzVDBkRmQwNVVRVFJhUjAweFRucFJlVTlIVVRBS1QwUm9iRnBFUVdoQ1oyOXlRbWRGUlVGWlR5OU5RVVZWUWtKTlRVVllaSFpqYlhSdFlrYzVNMWd5VW5Cak0wSm9aRWRPYjAxSFVVZERhWE5IUVZGUlFncG5OemgzUVZKVlJWWm5lRlZoU0ZJd1kwaE5Oa3g1T1c1aFdGSnZaRmRKZFZreU9YUk1NbWd4V2pKa2NHSnRaRzFaVjA1c1RESjBiR050Tld4aVNFMTBDbGt5T1hSaVdGWjFZVmhTTlV3eVJtcGtSMngyWW01TmRtTnVWblZqZVRoNlQxUkplazlVVlRST2VtczBUbE01YUdSSVVteGlXRUl3WTNrNGVFMUNXVWNLUTJselIwRlJVVUpuTnpoM1FWSlpSVU5CZDBkalNGWnBZa2RzYWsxRldVZERhWE5IUVZGUlFtYzNPSGRCVW1kRlQwRjNNbU50Vm5kaWVuQnZaRmRrYmdwaFZ6VnVXbTFHYWxwVE9YSmFXRXAxV2xkNGVreFhUblppVnpFeFltMXNNR1ZVY0hsYVYxazJZMjFXYldONU9XOWFWMFpyWTNrNWRGbFhiSFZOU1VkTENrSm5iM0pDWjBWRlFXUmFOVUZuVVVOQ1NIZEZaV2RDTkVGSVdVRXpWREIzWVhOaVNFVlVTbXBIVWpSamJWZGpNMEZ4U2t0WWNtcGxVRXN6TDJnMGNIa0taME00Y0Rkdk5FRkJRVWRvVmxOVkwwMTNRVUZDUVUxQlVucENSa0ZwUWxWVFJucElSR2h2VVhwclFXaEVWMkZGVFdOWlRFOVBUVGg2TjNacGNFSklTQXBpV0VOT1l6VnVjblZuU1doQlR6VXJXRTFsVHpWR1dYWkxWV2RtVm5WQ2MxUkxhM0pZT0RGNFdrbDNibkpPZFhWMVlrNTVNblF6VDAxQmIwZERRM0ZIQ2xOTk5EbENRVTFFUVRKblFVMUhWVU5OVVVSelUzSlZVVkYzWjJWM05FTnZNMXBCYVZFNFNsWjRjVWs0ZEc5VmNFYzFibWh0TVdocWEweFNOVnB4TDJFS09VNXZaSGc1THpVckt6Wm1ZWGt4WXk5cGIwTk5SbnBoUm5KTFIxVnpibmwwWWxOSFluaFpTM2N3TWxaUGNVWXphVXRoVDFSRVIxQnhhMU5SUkVkSFF3cFVjV3B5ZW1GTFNuQkVWVUV6YVVNNE1uZHlUVkozUFQwS0xTMHRMUzFGVGtRZ1EwVlNWRWxHU1VOQlZFVXRMUzB0TFFvPSJ9fX19"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"VHZW4gVMLV+AUqjAQAFANEK8hE+Tf9IM7NuIfOmCaJQ="}, "signature":"MEQCIAimVmUvjL1qjUexAhyw9FtZfIRv8wSqXBJMtw8qMn2IAiAJjTPkj5fIUS5IeBu2R9cALfVtCxWojm7pXTdZh9/+Zw=="}} \ No newline at end of file diff --git a/build/torch212-cu126-x86_64-linux/symbols.json b/build/torch212-cu126-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch212-cu126-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch212-cu130-x86_64-linux/__init__.py b/build/torch212-cu130-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch212-cu130-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch212-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch212-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..4dcc5582a4b6590464b0cf486dfde7345bdc23a1 --- /dev/null +++ b/build/torch212-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d4abdae879c1e5e1d965e8da7892101321331f0e17cdeb17c453fec37dd67d7 +size 1236410976 diff --git a/build/torch212-cu130-x86_64-linux/_ops.py b/build/torch212-cu130-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch212-cu130-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch212-cu130-x86_64-linux/layers.py b/build/torch212-cu130-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch212-cu130-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch212-cu130-x86_64-linux/metadata.json b/build/torch212-cu130-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..cc5e82e96834dcb36b3e3772f05677976d711b44 --- /dev/null +++ b/build/torch212-cu130-x86_64-linux/metadata.json @@ -0,0 +1,40 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "10.0", + "12.0", + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "DUq9roecHl4dll6Np4khATITMfDhfN6xfEU/7DfdZ9c=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch212-cu130-x86_64-linux/metadata.json.sigstore b/build/torch212-cu130-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..76eb43db23045f4ff7d2c536950b7ebe842fa3a2 --- /dev/null +++ b/build/torch212-cu130-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"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"}, "tlogEntries":[{"logIndex":"3323913431", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429868", "inclusionPromise":{"signedEntryTimestamp":"MEQCICbHDpWGe9Y+Od+bbiC8qB0UhptEibl6rocioELncSGtAiBcbwUec8f/lvJeOwmxGYYNhRw9nIdPvTybGBfFxXfG3w=="}, "inclusionProof":{"logIndex":"3202009169", "rootHash":"lPvYbylxLykfd2yNzGWoro0DW2T2dzbrtkEHCLPMGwI=", "treeSize":"3202009324", "hashes":["bZrjymjf7Ck/SrJvUPjanqcRYKCdJwr+ThxwnnMfybE=", "BCS8Ve0PTuBAReqOUyvcVLtZ/iNFYdkpJe2EEyF7Rck=", "AL8GEmfbVl96bssp8aFiwz6fBeQGN0W6G3H9PEZxKmc=", "Ugj9IeBEf/8RcbSUeHeGGoi6sUFxziUBUp5aejH+fkM=", "Y/1NSqdtV6tHhtFBnW92I3Nm/khdXT5Rouui+dxxsAE=", "3yfqm/et4CW9EwxVhATs0ZCOYFLf3zhKi6eMps61X5g=", "rbyxFIylrSPdlSI9h4u10fk3fCXhVvGIftlUB0eYd5U=", "NinK7eLOddAfFhQSIcawWkuJnGHi56v1YCMLuXX4occ=", "1C7/Gr9ThZu8wAXJvsC9tJ+9T5f/SwUdxqHmYEJ2s7s=", "ZoEKTEi+vNH82Rv/iMhp5QJLlu/Ocy2iHNm45P+eFP8=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202009324\nlPvYbylxLykfd2yNzGWoro0DW2T2dzbrtkEHCLPMGwI=\n\n— rekor.sigstore.dev wNI9ajBGAiEA5G0UA3JD1rIKuJbXjvuRsmwRArC+uKhs1EYs0/jaBhYCIQDkExdQd3GoVnqJocVmKrEsTcizXiA5Vnqua8NF7VM4mg==\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"u14jakL0xCznodtwnsQXBPDIKD8eisOUqFXr9gh87LI="}, "signature":"MEUCIQDYLbDUni0TuhWRS+LSLOP8r1uCl5mHYpHfqH+9qDfQKQIgGrfMsDbObqJnSteacEi66/7UKxLxpAWSsawhzR/0WA4="}} \ No newline at end of file diff --git a/build/torch212-cu130-x86_64-linux/symbols.json b/build/torch212-cu130-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch212-cu130-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch212-cu132-x86_64-linux/__init__.py b/build/torch212-cu132-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch212-cu132-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch212-cu132-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch212-cu132-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..4eec9d26ca7204211fdbc84a51a56e64afbee33d --- /dev/null +++ b/build/torch212-cu132-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa53d7d30afd95ba05ef2a5d1e76dc8c44aac70756d893b43809a471610e77a2 +size 1220019176 diff --git a/build/torch212-cu132-x86_64-linux/_ops.py b/build/torch212-cu132-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch212-cu132-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch212-cu132-x86_64-linux/layers.py b/build/torch212-cu132-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch212-cu132-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch212-cu132-x86_64-linux/metadata.json b/build/torch212-cu132-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..310be5e15e3b92c51ff38198bccd6d0e9691208e --- /dev/null +++ b/build/torch212-cu132-x86_64-linux/metadata.json @@ -0,0 +1,40 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "10.0", + "12.0", + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "qlPX0wr9lboF7ypdHnbcjESqxwdW2JO0OAmkcWEOd6I=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch212-cu132-x86_64-linux/metadata.json.sigstore b/build/torch212-cu132-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..727c5b280dc50679283bd098eec4b71f588e134d --- /dev/null +++ b/build/torch212-cu132-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"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"}, "tlogEntries":[{"logIndex":"3323913780", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429870", "inclusionPromise":{"signedEntryTimestamp":"MEUCICxQiRpyLpUevvaolApULPvNEkqlXuJaR/Kft8v2PoCkAiEA9FkUFnY7jT3eNlCAdi0WRnPQR5tD3HxMYEtMIe64V5U="}, "inclusionProof":{"logIndex":"3202009518", "rootHash":"d2VJyD6y3jlUOoRq1xUdavMuOLFP3JZInlZlQTWYybQ=", "treeSize":"3202009791", "hashes":["XxONpaOZegKsWPoBVInTjuRI/oMBCSJRSpFDlXhmfac=", "ylCsNBCBQAO1/6d+1MdVsodkm8b4kpnu3Mg7jock8sg=", "I/uvWQdxNsu/Lnlx3kLwUz+13ECXsZeI8OlBrHBYeJc=", "lqKEloPZWw/NRxr65aOzXa+AHKgbrkaszAJ2FF2mp6w=", "MIflXx5lZIYJomDN0TcU7PyNU2dqPHA35RFsE2sSkyU=", "Ci8ioM5HInAkgb6M2WpdnAfVTogvvWzBuJexQbhosWU=", "QCiNWX8Q1i6mXVfJXNrJi1S7qzssnfHZ8gYA7O6KXDc=", "63ODpqU/O/45ukxd6gCdTja+bOI4pqvTwHogr7G/hOA=", "pYv+kth4hKlg7ajGiOWIoasPq6k6Yp9vcfijwOnV9TA=", "7xvRSXaBDVVhKWMCCwejteut7yIKG5qVVLqYEAkafXM=", "1C7/Gr9ThZu8wAXJvsC9tJ+9T5f/SwUdxqHmYEJ2s7s=", "ZoEKTEi+vNH82Rv/iMhp5QJLlu/Ocy2iHNm45P+eFP8=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202009791\nd2VJyD6y3jlUOoRq1xUdavMuOLFP3JZInlZlQTWYybQ=\n\n— rekor.sigstore.dev wNI9ajBGAiEA5azdCINOxCFbT/fhiU6vKBBapCv8ToIi9pGtcmfs+xgCIQCRLTsNsFFE1nTXjEZSfFBE2dxkZ1gSAzyjT5dbV08eqw==\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"bNFRt6OZohkbGmABtPSJdMciUWtYmQqi9hfdQdpmYag="}, "signature":"MEUCIQC4TZp6KYOR3CzVJeW415wBGv/qA4yWkpyieaw/PhIuFQIgUCcQeyAnBSbI9b/h8JETZ/11eQ59nHgSybIiAU/rc9Q="}} \ No newline at end of file diff --git a/build/torch212-cu132-x86_64-linux/symbols.json b/build/torch212-cu132-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch212-cu132-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch213-cu126-x86_64-linux/__init__.py b/build/torch213-cu126-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch213-cu126-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch213-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch213-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..9bdfd7f26654502cf7974b00e40d4a370958ad28 --- /dev/null +++ b/build/torch213-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:daacd535bb4b32350aff458b1bf011df09930a906fa1a3fc38bce4522537ba0e +size 710169656 diff --git a/build/torch213-cu126-x86_64-linux/_ops.py b/build/torch213-cu126-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch213-cu126-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch213-cu126-x86_64-linux/layers.py b/build/torch213-cu126-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch213-cu126-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch213-cu126-x86_64-linux/metadata.json b/build/torch213-cu126-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..d7f0ffc86fd8ed02c254f48db9f94ab2cac817d1 --- /dev/null +++ b/build/torch213-cu126-x86_64-linux/metadata.json @@ -0,0 +1,38 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "2qzVNbtLMjUK/0WLG/AR3wmTCpBvoaP8OLzkUiU3ug4=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch213-cu126-x86_64-linux/metadata.json.sigstore b/build/torch213-cu126-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..712214e673171f66cc3071c63bb8dd1a006e7ec9 --- /dev/null +++ b/build/torch213-cu126-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"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"}, "tlogEntries":[{"logIndex":"3323914295", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429873", "inclusionPromise":{"signedEntryTimestamp":"MEYCIQDL4Ww77sAOALcQrb0xFgd39S5lHGvrbSjK4lx7AmJQlwIhAM2fMb85+RFcSaGnqa6Xno4dW+LxT+ARn0OPOk3LRTWF"}, "inclusionProof":{"logIndex":"3202010033", "rootHash":"eeAxgwrKEMns8zTMEBZ73sksP2GUfnTM/0KiaT4aZ+Y=", "treeSize":"3202010116", "hashes":["iCgaPwdiHmkMic6iR4PJr1yQrBlbpEORzr91IMAmjzs=", "Xsm48FkvuTIpg0W73RLTfSnVARCdpztyi0x45nlwJAA=", "omqpIVgWL72WtD63Dmj2ZwxyiuuFfZ+jmXqA4gy6mAM=", "p4nTfvNZFfmfWS4wxY0wbyqgboOFlnfOaD14GzOSVB8=", "AWFSCrijk93JEHindFYhn4qiCrhjwc6LW6r2YaVmkhc=", "jJR/AY3uLXAa0bmF601dakqhv9NgR5Z1WucsSVQMZ9Q=", "PNO0LC4CMV2eDvrK1i+1BonmlFmvlCVkNlz9aEdfCOg=", "c3oU6p+Uuv59HhQYYMgP0MMxu/d3NUt8Uqz9Gx4rD4s=", "/ei1YodWC0gMBPtr5AyuayiFQHLaUzHuRg+S9iP74Mc=", "dp/6soH6jce08sA4k0CH7/WBvVcUJGTSgShdG6q8VaI=", "1lxbusfQAQYXfV+YDeQ+aAtRehvwEgDBq/b4aerQS1s=", "1C7/Gr9ThZu8wAXJvsC9tJ+9T5f/SwUdxqHmYEJ2s7s=", "ZoEKTEi+vNH82Rv/iMhp5QJLlu/Ocy2iHNm45P+eFP8=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202010116\neeAxgwrKEMns8zTMEBZ73sksP2GUfnTM/0KiaT4aZ+Y=\n\n— rekor.sigstore.dev wNI9ajBEAiAGrpAUg9FKJEl95t2T91D7kGBPqMHaHwv/pKEv+AAllAIgJ45yLLSAA0w19TLzWc/WgEKjRO8JjtLxuv5JCDXw9Cw=\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"D4AsPvgrOt92lfwTveMKFyMlZZgXyIHZefsYA1xEOVw="}, "signature":"MEYCIQDGysczrArIXEWsyMQtl9dajPIB/JTVshlHk3JHAPR+4gIhAKMcHwdYhC7/Ot/x995MBhPyK++IiCFNXU9zHXJE0tIE"}} \ No newline at end of file diff --git a/build/torch213-cu126-x86_64-linux/symbols.json b/build/torch213-cu126-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch213-cu126-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch213-cu130-x86_64-linux/__init__.py b/build/torch213-cu130-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch213-cu130-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch213-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch213-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..fd3ea8a89549a5b66a3713770b0e9f1f057194f0 --- /dev/null +++ b/build/torch213-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9f8780a1f6453b1016c328422e6a9adac3635dd9e7edb60e8880e0d57d552079 +size 1236411168 diff --git a/build/torch213-cu130-x86_64-linux/_ops.py b/build/torch213-cu130-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch213-cu130-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch213-cu130-x86_64-linux/layers.py b/build/torch213-cu130-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch213-cu130-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch213-cu130-x86_64-linux/metadata.json b/build/torch213-cu130-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..326b64e51587336f8ecb8b7ab86162a7ebf65194 --- /dev/null +++ b/build/torch213-cu130-x86_64-linux/metadata.json @@ -0,0 +1,40 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "10.0", + "12.0", + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "n4eAofZFOxAWwyhCLmqa2sNjXdnn7bYOiIDg1X1VIHk=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch213-cu130-x86_64-linux/metadata.json.sigstore b/build/torch213-cu130-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..31050accb3220b00dbf9585a4f4ff03e7f4cd2ae --- /dev/null +++ b/build/torch213-cu130-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"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"}, "tlogEntries":[{"logIndex":"3323914839", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429876", "inclusionPromise":{"signedEntryTimestamp":"MEYCIQD7gbZ4a4Qli4HcHPdNG/ZhCwGDDFxKlnJjf52Encvk8gIhAMczzGvAB7/qeHbjLsqSGZCI+HzJlQHrj8opwKzv46Jj"}, "inclusionProof":{"logIndex":"3202010577", "rootHash":"WyOrhSabzu8yAYAgV+P4+oXDNZTgUYt7kF/+R9hfgbE=", "treeSize":"3202010813", "hashes":["M0tWgn3vzAHPliXJkjwuzmJ2yevwuVWGOqIXx0okTh0=", "zdrxMwhcYN8fOCikBl77oQnXzJvdKeheHP9jyX9WIKc=", "5RorJoBxCreyAmDTE6vLxxcL8oKEU7UpYhSM9JF7/7c=", "tAtYKbBNkuOvk3wbaKKDp078yO9oPlTosCGV8nrim6s=", "7TMK4jqcb2UX+VfbbOmF7lhsNK1xpUSaQgy1DZlLqkk=", "hR8QhBJiEvl+NyDqdHbNhcixJbPaRuEUMICuSnmwd8g=", "yCawQbMj3fs0ENQSG/k6Zuq3FUe//gsfcTmwGED8MOQ=", "LOmFX+QGLINux1Va8x/jszzSUD3XooUAto+EYT/atYA=", "h60OGJA0sumccKLf6MSF9mxxcd31VdGBVJgj8GVJ0ME=", "HZJtvVTTe0OknIEI7maVhEF5ZGnKHBrCPJhuqoeg2ak=", "fGTQy+lLjz9eRx33Gub5DrkD3KBXyFN90z+ISoaAjuc=", "1C7/Gr9ThZu8wAXJvsC9tJ+9T5f/SwUdxqHmYEJ2s7s=", "ZoEKTEi+vNH82Rv/iMhp5QJLlu/Ocy2iHNm45P+eFP8=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202010813\nWyOrhSabzu8yAYAgV+P4+oXDNZTgUYt7kF/+R9hfgbE=\n\n— rekor.sigstore.dev wNI9ajBEAiApJawhoqksUW3aJ5+O+qczxMH1pqlc2HVWPpe9BNgqzQIgV3o2hEhFpu6iDClpqM1N2SyDM9nsStWRwtt0RTHm/Mc=\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"8Kux7JZLswR1gVQxN33EA4mOcuusKYgj9nvn7q+sUl0="}, "signature":"MEYCIQDC4lmX4zXmTtmQzbvbvd/OQvBSVz3ksexzMPwmQbgO4wIhANYRlc6kbUWu5cXLB+N1ZWiiVJryDnIa4rQILqtG3UUa"}} \ No newline at end of file diff --git a/build/torch213-cu130-x86_64-linux/symbols.json b/build/torch213-cu130-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch213-cu130-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch213-cu132-x86_64-linux/__init__.py b/build/torch213-cu132-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch213-cu132-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch213-cu132-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch213-cu132-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..db8961f34fa0482e0e2582da4b0cbefde77061c8 --- /dev/null +++ b/build/torch213-cu132-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c0faa107f50bbf6345a47a38f0094b96083e4eb034ee3434bbf93f9e4e6388bb +size 1220019368 diff --git a/build/torch213-cu132-x86_64-linux/_ops.py b/build/torch213-cu132-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch213-cu132-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch213-cu132-x86_64-linux/layers.py b/build/torch213-cu132-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch213-cu132-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch213-cu132-x86_64-linux/metadata.json b/build/torch213-cu132-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..35fd6ae632827f06379baee950a08dc8f87e2f22 --- /dev/null +++ b/build/torch213-cu132-x86_64-linux/metadata.json @@ -0,0 +1,40 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "10.0", + "12.0", + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "wPqhB/ULv2NFpHo48AlLlgg+TrA07jQ0u/k/nk5jiLs=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch213-cu132-x86_64-linux/metadata.json.sigstore b/build/torch213-cu132-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..daca432e585a47b812a6b46255d418ffd3ee5a80 --- /dev/null +++ b/build/torch213-cu132-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"MIIHLDCCBrKgAwIBAgIUG9XCAE4BncTjmDAZ/RdYWoND4UcwCgYIKoZIzj0EAwMwNzEVMBMGA1UEChMMc2lnc3RvcmUuZGV2MR4wHAYDVQQDExVzaWdzdG9yZS1pbnRlcm1lZGlhdGUwHhcNMjYxMDE5MTcxMTE3WhcNMjYxMDE5MTcyMTE3WjAAMFkwEwYHKoZIzj0CAQYIKoZIzj0DAQcDQgAEjXlESzVx0ueveknJxyYWpxMynJYfx8cAbpXTQDgMGoS51aCUmxKkG/MqvkCmqE3mRlasTqBLyVHKfCUsWmKqA6OCBdEwggXNMA4GA1UdDwEB/wQEAwIHgDATBgNVHSUEDDAKBggrBgEFBQcDAzAfBgNVHSMEGDAWgBTf0+nPViQRlvmo2OkoVaLGLhhkPzBrBgNVHREBAf8EYTBfhl1odHRwczovL2dpdGh1Yi5jb20vaHVnZ2luZ2ZhY2Uva2VybmVscy1jb21tdW5pdHkvLmdpdGh1Yi93b3JrZmxvd3MvYnVpbGQueWFtbEByZWZzL2hlYWRzL21haW4wOQYKKwYBBAGDvzABAQQraHR0cHM6Ly90b2tlbi5hY3Rpb25zLmdpdGh1YnVzZXJjb250ZW50LmNvbTAfBgorBgEEAYO/MAECBBF3b3JrZmxvd19kaXNwYXRjaDA2BgorBgEEAYO/MAEDBChmZTJiYjY2ZmVlNmMyYTc3ZWNjZjA4YTA1MDhkYzU3NDI4ZDQ4OGVkMBMGCisGAQQBg78wAQQEBUJ1aWxkMCsGCisGAQQBg78wAQUEHWh1Z2dpbmdmYWNlL2tlcm5lbHMtY29tbXVuaXR5MB0GCisGAQQBg78wAQYED3JlZnMvaGVhZHMvbWFpbjA7BgorBgEEAYO/MAEIBC0MK2h0dHBzOi8vdG9rZW4uYWN0aW9ucy5naXRodWJ1c2VyY29udGVudC5jb20wbQYKKwYBBAGDvzABCQRfDF1odHRwczovL2dpdGh1Yi5jb20vaHVnZ2luZ2ZhY2Uva2VybmVscy1jb21tdW5pdHkvLmdpdGh1Yi93b3JrZmxvd3MvYnVpbGQueWFtbEByZWZzL2hlYWRzL21haW4wOAYKKwYBBAGDvzABCgQqDChmZTJiYjY2ZmVlNmMyYTc3ZWNjZjA4YTA1MDhkYzU3NDI4ZDQ4OGVkMBsGCisGAQQBg78wAQsEDQwLc2VsZi1ob3N0ZWQwQAYKKwYBBAGDvzABDAQyDDBodHRwczovL2dpdGh1Yi5jb20vaHVnZ2luZ2ZhY2Uva2VybmVscy1jb21tdW5pdHkwOAYKKwYBBAGDvzABDQQqDChmZTJiYjY2ZmVlNmMyYTc3ZWNjZjA4YTA1MDhkYzU3NDI4ZDQ4OGVkMB8GCisGAQQBg78wAQ4EEQwPcmVmcy9oZWFkcy9tYWluMBoGCisGAQQBg78wAQ8EDAwKMTA3MTQ3NTUyOTAuBgorBgEEAYO/MAEQBCAMHmh0dHBzOi8vZ2l0aHViLmNvbS9odWdnaW5nZmFjZTAYBgorBgEEAYO/MAERBAoMCDI1NzIwNzQzMG0GCisGAQQBg78wARIEXwxdaHR0cHM6Ly9naXRodWIuY29tL2h1Z2dpbmdmYWNlL2tlcm5lbHMtY29tbXVuaXR5Ly5naXRodWIvd29ya2Zsb3dzL2J1aWxkLnlhbWxAcmVmcy9oZWFkcy9tYWluMDgGCisGAQQBg78wARMEKgwoZmUyYmI2NmZlZTZjMmE3N2VjY2YwOGEwNTA4ZGM1NzQyOGQ0ODhlZDAhBgorBgEEAYO/MAEUBBMMEXdvcmtmbG93X2Rpc3BhdGNoMGQGCisGAQQBg78wARUEVgxUaHR0cHM6Ly9naXRodWIuY29tL2h1Z2dpbmdmYWNlL2tlcm5lbHMtY29tbXVuaXR5L2FjdGlvbnMvcnVucy8zOTIzOTU4Nzk4NS9hdHRlbXB0cy8xMBYGCisGAQQBg78wARYECAwGcHVibGljMEYGCisGAQQBg78wARgEOAw2cmVwbzpodWdnaW5nZmFjZS9rZXJuZWxzLWNvbW11bml0eTpyZWY6cmVmcy9oZWFkcy9tYWluMIGKBgorBgEEAdZ5AgQCBHwEegB4AHYA3T0wasbHETJjGR4cmWc3AqJKXrjePK3/h4pygC8p7o4AAAGhVSVqpAAABAMARzBFAiEAoMZLsn+l/vJE+X5E18bL+uK61UmXKkFz10/wDbtT2ooCIFUV1560QkJSt6r9W8OQO5eWbQOA7J/OSmZdAflI+i5qMAoGCCqGSM49BAMDA2gAMGUCMQDLf1Z/nYxTsAfSBgUXrTJs6AJbQpoXNe1Jk67xG2QziFjbaC5/A7panjT4ThQ6FkECMBX5cbA60xnRBu5bPG15YEH14AWuVzNCL02fNUmgcSmvo9TDr27dMlA+ZcM/wZ2tog=="}, "tlogEntries":[{"logIndex":"3323915453", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429879", "inclusionPromise":{"signedEntryTimestamp":"MEUCIFKKomOc0avZcffBkfzOdcNn2klwcCg4cC95aaLSThvkAiEA4dDaVGqI3CkPKi60fulUVWywxCqAJ/GhOXIvYxPmRrk="}, "inclusionProof":{"logIndex":"3202011191", "rootHash":"3hl1h6Rw5wSTWwO0inOvbA78Y8sUwegKyjzrHNMJPKs=", "treeSize":"3202011246", "hashes":["vlAU13pSKf3a4KiT+ytmkMiZso6AOkmcSzdhPfGmYvk=", "GSnOTMSC5+mjdhMsOcklW940KCjHKcMTN3F/df24dfA=", "vGuoRxQ4iG5qC9QxAumxkGuyCDvjItRmbh541JPz/Yw=", "tliictUw4WiroUvK2EBlbiiH0cb4gw7ZF/Y8ztgIzX4=", "+GHJnpDjYX8EhyzJKiPM7ZKcv89MnaIPKqMHhlp59XM=", "9UJCaNYZ1jL0QTuClFT2x6T7c++oJmb1JouLli7VkXE=", "ZK5gkC2iNioQjNp2wszwZXA9pUhUk5vrdczHwsoH0KQ=", "X1OwG1pW/NAf1zSALYc7dJD4MRPAS3+RElCmZVkRTNQ=", "ZoEKTEi+vNH82Rv/iMhp5QJLlu/Ocy2iHNm45P+eFP8=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202011246\n3hl1h6Rw5wSTWwO0inOvbA78Y8sUwegKyjzrHNMJPKs=\n\n— rekor.sigstore.dev wNI9ajBEAiARTjlXD1mCag0BfqwlL+Ptuq0+LZcX5aBWBZTo8/K0QAIgOwKJYFNh6PJ/A48q/9U/v2h3/hSgte2fF7Ojgt09GyM=\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"dE6AVGUB+4zo2GmnADmTLQGcsKioxwXBLNCcuNDYXes="}, "signature":"MEYCIQDILxGogejgsFQD9u/RxoMIvWPTSOjzHVJgtmHxf5T+4wIhAN5Z2uyobrOSXGY6143mCB2qzT+Vn/pLKtDukmlBt+PB"}} \ No newline at end of file diff --git a/build/torch213-cu132-x86_64-linux/symbols.json b/build/torch213-cu132-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch213-cu132-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch214-cu126-x86_64-linux/__init__.py b/build/torch214-cu126-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch214-cu126-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch214-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch214-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..433f1763d4ca5a062c50858dc5e36cb86a2f5885 --- /dev/null +++ b/build/torch214-cu126-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:72f14805155f041abdef8f6f1c412d563ea607aabf800eadbf71a5ea476b4249 +size 710178248 diff --git a/build/torch214-cu126-x86_64-linux/_ops.py b/build/torch214-cu126-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch214-cu126-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch214-cu126-x86_64-linux/layers.py b/build/torch214-cu126-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch214-cu126-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch214-cu126-x86_64-linux/metadata.json b/build/torch214-cu126-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..b3bbaf6f46b83cc5c7d0ae5b1ebbec00998470e4 --- /dev/null +++ b/build/torch214-cu126-x86_64-linux/metadata.json @@ -0,0 +1,38 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "cvFIBRVfBBq9749vHEEtVj6mB6q/gA6tv3Gl6kdrQkk=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch214-cu126-x86_64-linux/metadata.json.sigstore b/build/torch214-cu126-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..6c79b672d0daa2aed6e98e5e3da4b6818a75f85f --- /dev/null +++ b/build/torch214-cu126-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"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"}, "tlogEntries":[{"logIndex":"3323915903", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429881", "inclusionPromise":{"signedEntryTimestamp":"MEUCIAUN5dtsxo+2BAZI4Nm9EucMNFrxaRMTZTDR5xRAuN+2AiEA1QU0UNGTcJx1l7sC3hF8qea34x3Ii/EVxR3vHbs26rY="}, "inclusionProof":{"logIndex":"3202011641", "rootHash":"feq4D6qjGWgast4UrpzwXfrWJ7fTg9gvYXse0VF7Ro8=", "treeSize":"3202011767", "hashes":["xUTOJPGWqiWV9XEUDzL47OeYFtxPSyTuTBwwnId6Hs0=", "4D1HdXMz0Ejdd7Dq1fDBokiWm9IjSVqIRQi2fsw9eVc=", "WV8e9CnbmCA4Y+1bLKHEeUuIOOA9zIISXYJlc7lNwQo=", "pyuO7yraLxl8ZxYFOhYfiKG50kCHmp/eW30GPLmOU14=", "S0W2i9fDqrCPyKQoT3mOevx08pZ8bY5TgXmoZ79VK7g=", "023WBVz+oSxJjyDMfMU180pQbt+MTOM0I1l0NE1VhZo=", "Se4Q8Ax6tpAijsLbfP+4hOOc1Yw1TArU6ormQdyPvaM=", "KVtqLijTGphnGe2ztoNinTkoL49mjuMYZ+bMraHszjo=", "b84T50eZ/t6DEni+Rza9sW7urm+VSit9vB4GccnOG18=", "bffy4kwMaL6FfhnSuCMntiRmPZtgma22QGM6tRPbq3Y=", "X1OwG1pW/NAf1zSALYc7dJD4MRPAS3+RElCmZVkRTNQ=", "ZoEKTEi+vNH82Rv/iMhp5QJLlu/Ocy2iHNm45P+eFP8=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202011767\nfeq4D6qjGWgast4UrpzwXfrWJ7fTg9gvYXse0VF7Ro8=\n\n— rekor.sigstore.dev wNI9ajBEAiBhpLXBLa4FHBXLExYetBBSCTaijeYDJbMMtwJSenCJNwIgSCK92BfDrtIz6exqW7/YDFJNES1BHtLOTWv+RrBTXpo=\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"m3YKiW49SjSDtPIEXF8ejxUQyr31mgxzNCU574e2Rwo="}, "signature":"MEUCIAj1S2DgocQvVeYqK46TqHzLtYZManFQFdiYKwtrSrr/AiEAuCEAMWfgWylJfrMVDuVTiSeEhdULaN35vmx/yCBzm0I="}} \ No newline at end of file diff --git a/build/torch214-cu126-x86_64-linux/symbols.json b/build/torch214-cu126-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch214-cu126-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch214-cu130-x86_64-linux/__init__.py b/build/torch214-cu130-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch214-cu130-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch214-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch214-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..236f3b3b8c8ebb3a6b222d60b781425af4c9034b --- /dev/null +++ b/build/torch214-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:57a9f37f8d5c08eced72d2592066afbcc9d28d53a60f24e7979e118ed6421f63 +size 1236415472 diff --git a/build/torch214-cu130-x86_64-linux/_ops.py b/build/torch214-cu130-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch214-cu130-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch214-cu130-x86_64-linux/layers.py b/build/torch214-cu130-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch214-cu130-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch214-cu130-x86_64-linux/metadata.json b/build/torch214-cu130-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..fb17f035c6c753bd943e14abb143fac22f55b6b7 --- /dev/null +++ b/build/torch214-cu130-x86_64-linux/metadata.json @@ -0,0 +1,40 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "10.0", + "12.0", + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "V6nzf41cCOztctJZIGavvMnSjVOmDyTnl54RjtZCH2M=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch214-cu130-x86_64-linux/metadata.json.sigstore b/build/torch214-cu130-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..7d859d314fd535f929b573b66fef96b304bfa19b --- /dev/null +++ b/build/torch214-cu130-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"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"}, "tlogEntries":[{"logIndex":"3323916574", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429884", "inclusionPromise":{"signedEntryTimestamp":"MEQCIDpXHSewl6DAP6ICh1uhdOjeKfspRXNE321lFw8wnXkzAiBGjFlxZHwDPSscSV9pF6eC4gHHCRzJwGjC8tufRliNjg=="}, "inclusionProof":{"logIndex":"3202012312", "rootHash":"z5rIiYDt+DLRabzN4Cho4rHwzvogza7WOqhxzVbZb5w=", "treeSize":"3202012408", "hashes":["nm32+yxPl23Evjivj+Llewt+mmxtoqSuroKW+PvqAPw=", "T4cWheJaBnrvYeRx+X8dtT6qJoobP1bJ/nTB6rWJIL0=", "4mRqVPICTG6DfJm0FQOMhuuwGe7YZXd5KDZ9wB10K3k=", "Vf9ljKqhvm5Rm/pkMwgWa5eUiLhArzg+GRA9XdRoaR0=", "nKxTj5r4e4JamO1ruRNAiOV3trQ1ccqt+pM5RSJCLP0=", "JWs9v42fr9BeavfHogLk8/wbdm4pC3mn1Pf19RAPmqc=", "GNFqKaIiEZqpMD4ioiLPdQlT9WbKd8sz1tDvefAfE2Q=", "FnoedKG8T6/zvGZaWMzYb99PtQntP+HYkZfuzTGTh2k=", "gZ6OoUzN1d4HmEL4I8aWZQr9IojzGAmXgRZRvgGne6M=", "X1OwG1pW/NAf1zSALYc7dJD4MRPAS3+RElCmZVkRTNQ=", "ZoEKTEi+vNH82Rv/iMhp5QJLlu/Ocy2iHNm45P+eFP8=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202012408\nz5rIiYDt+DLRabzN4Cho4rHwzvogza7WOqhxzVbZb5w=\n\n— rekor.sigstore.dev wNI9ajBFAiEAiTADnjqQMpQsPBvPzzPB2moPYIyNR6nSlvichYWGzLMCIE0emJF7MJ7udhrszoLYwEFEypa9J/xP+U1O6ezjc1TK\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"XHRB3HKONnyU992cqS6GEa7rYNEKJTt/v2WoSveQWlg="}, "signature":"MEUCIEKLJCBnSLHDfJqeGW8HZhswP92qvh8GxMg+xfMOHiKuAiEAhhgkLrCUuewAmsgcFpTz89GKKyWmOq7xDlc77i/h8oE="}} \ No newline at end of file diff --git a/build/torch214-cu130-x86_64-linux/symbols.json b/build/torch214-cu130-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch214-cu130-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch214-cu132-x86_64-linux/__init__.py b/build/torch214-cu132-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch214-cu132-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch214-cu132-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch214-cu132-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..9d984f9a99b265a23a5c0e203b720950955ddaf9 --- /dev/null +++ b/build/torch214-cu132-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4367d93e7900ddc662da3bf2b7d6838ecb37d6aec52d6b762f9a658969ea13fa +size 1220023672 diff --git a/build/torch214-cu132-x86_64-linux/_ops.py b/build/torch214-cu132-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch214-cu132-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch214-cu132-x86_64-linux/layers.py b/build/torch214-cu132-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch214-cu132-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch214-cu132-x86_64-linux/metadata.json b/build/torch214-cu132-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..3f76fecd57ad2c05d9abd637152974882b86c28e --- /dev/null +++ b/build/torch214-cu132-x86_64-linux/metadata.json @@ -0,0 +1,40 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "10.0", + "12.0", + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "Q2fZPnkA3cZi2jvyt9aDjss31q7FLWt2L5pliWnqE/o=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch214-cu132-x86_64-linux/metadata.json.sigstore b/build/torch214-cu132-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..9574a1528ed652eb803d501eabd4c3281e530dc4 --- /dev/null +++ b/build/torch214-cu132-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"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"}, "tlogEntries":[{"logIndex":"3323916911", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429886", "inclusionPromise":{"signedEntryTimestamp":"MEYCIQDkWSitap3BGaOS44nWDHnsDOEXvBJfO2+Jp/0lSpNg2AIhAPtYXzWdyyRLKlIucdX0/ei1FPrjXu3S+HmGwQO0Q7Xx"}, "inclusionProof":{"logIndex":"3202012649", "rootHash":"vNNS3rWQT91M2MXMbdCQDUauSoGuYc2+xKDkA01jJKI=", "treeSize":"3202012752", "hashes":["L1GZqmPenq5AE0SPL1y1WGv28PJ+Ss2K2VHn7FWUOcY=", "gPCXJ5KmWaR3iNQwTHdZYKni0AF3JfeZ4aVVf8YlkH4=", "f9b2uAoZP45de7QAW5TiokyHp9Ys6LglFZWOPX9kl7A=", "qr+EL55c43RRgVCotbtIJTCmgWuX0ksYrWLYtPMB2Uc=", "0zv10niF1JgNr824HuW+0CSyWairb70o5m+Y26OVZ+g=", "J2lXtvpNLqajcjVk7fiixMyFk/LmQnhNa19c9OVidkE=", "3IrwFvqOesJCmT2h1gwE3TIIwk3vqmFEdEaALAFHfwI=", "94wLqREdgWYCXbEQU9lt0dlfuX/xMdYOKAft92fXKQU=", "sZ+915VwpG8n7NjFW29mAUwK7Uvx1ovtVylwe5G+Utg=", "rraFPUM27ROzsXP607mxCdIE3g9vtcBokp0K76thbCg=", "gZ6OoUzN1d4HmEL4I8aWZQr9IojzGAmXgRZRvgGne6M=", "X1OwG1pW/NAf1zSALYc7dJD4MRPAS3+RElCmZVkRTNQ=", "ZoEKTEi+vNH82Rv/iMhp5QJLlu/Ocy2iHNm45P+eFP8=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202012752\nvNNS3rWQT91M2MXMbdCQDUauSoGuYc2+xKDkA01jJKI=\n\n— rekor.sigstore.dev wNI9ajBFAiAh/0pUVt0TtY2Ey6qsiXR7z/91MiQwPrwZuOyS/4ItXAIhAI5Jiq8nrHJ5ADJDZOlqhQgGuVi869utzS0laK1l3PMF\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"/wKbspZtFSmG3S5M6tA278U2a0HHZq/iHv4SaSXnw98="}, "signature":"MEYCIQD6iG/3GDBnTRBKiKdonNSlQkgqkj4WF5eireYCL7NSDwIhANC98SU24DfxpiW/+Iv98rxf1p0iB6J7CYzhi7q1V9KZ"}} \ No newline at end of file diff --git a/build/torch214-cu132-x86_64-linux/symbols.json b/build/torch214-cu132-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch214-cu132-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch215-cu130-x86_64-linux/__init__.py b/build/torch215-cu130-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch215-cu130-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch215-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch215-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..a616ebcda2b4f80887e0af94234351a15f19b1ff --- /dev/null +++ b/build/torch215-cu130-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b6d47413367426ee8f38169a9a6973c6ac1aedb038f5e764a49e7d6a1ce90122 +size 1236415672 diff --git a/build/torch215-cu130-x86_64-linux/_ops.py b/build/torch215-cu130-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch215-cu130-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch215-cu130-x86_64-linux/layers.py b/build/torch215-cu130-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch215-cu130-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch215-cu130-x86_64-linux/metadata.json b/build/torch215-cu130-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..6fe5ff8711ce798cd649800bbb32c639120d70b2 --- /dev/null +++ b/build/torch215-cu130-x86_64-linux/metadata.json @@ -0,0 +1,40 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "10.0", + "12.0", + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "ttR0EzZ0Ju6POBaammlzxqwa7bA49edkpJ59ahzpASI=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch215-cu130-x86_64-linux/metadata.json.sigstore b/build/torch215-cu130-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..314b0ba7376723fb6e4818f305ed543141502c47 --- /dev/null +++ b/build/torch215-cu130-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"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"}, "tlogEntries":[{"logIndex":"3323917278", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429888", "inclusionPromise":{"signedEntryTimestamp":"MEUCIQCL5L3D0clSTlJaaqalPM0EuPO68PlhrZa9VNSBxCt6JAIgM3MDbpVe/9TI2oUWf4GLYFxxGzOfDzd1u+RoEK+7VyI="}, "inclusionProof":{"logIndex":"3202013016", "rootHash":"t9Tcli2scPJjUx2p7NT6Cjr1u2Rp2rWbZ5Pg95crabM=", "treeSize":"3202013094", "hashes":["hTy7HQnO82ecoQO2pIVAYQf/XFHrO/8pcYRtbSSAfdQ=", "TBgvPn7n/g1V/hfXwWoF1PtaRQ5Xs8jl3P9tpEc8XaM=", "y+Q3Rc3kY6S3U0cG1fLzeX2qC6fUHZPMnbLtr8fOzUc=", "NfuhkNOEzJ9Tpjbn+pyC/3TjQ4T3ynfEpuEZoP2NGH8=", "HT36QNN2dOOphxkFjJLjUD5y449saNIwq2niNSZqOhE=", "/18gnQRN6W9hrQHuJ+JFQgbUQ/MUxEupi6rHIN6U64w=", "9e7QV03xYVqszzxDQK5D0p3o9GIcBR6WGVf3L8oZaWU=", "pA3Ry4oOt5nxwzm8sjM6SH2pbjJHeZlo+cQ6eYZjW2Q=", "ZnI/7/PQXaCfP6/gzZexrgbl1s1BKGF1qHLhyQ/IaP8=", "B30I/uOM1qC2G4ieZ+Th/fy1x3ikLgykj7/kJt50TGg=", "gZ6OoUzN1d4HmEL4I8aWZQr9IojzGAmXgRZRvgGne6M=", "X1OwG1pW/NAf1zSALYc7dJD4MRPAS3+RElCmZVkRTNQ=", "ZoEKTEi+vNH82Rv/iMhp5QJLlu/Ocy2iHNm45P+eFP8=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202013094\nt9Tcli2scPJjUx2p7NT6Cjr1u2Rp2rWbZ5Pg95crabM=\n\n— rekor.sigstore.dev wNI9ajBFAiBdBqHFSG0TmCztMVztVLEE78JDze81X/pTf7ty1Kb8bAIhAKvbtdJOs8ctAT01PSmylu9T0LrrNS7KYc9aLutQn8e7\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"CoqHRSGZJqGVnVnmnc5/LZA11nC+cC5PhqpAn9eUeKk="}, "signature":"MEYCIQC7YQkZMIG708eFTsu8jTkUXeTeG0RnTeX/1iuSRMbdeQIhALxDHLylXqfT7uR79Zd4Dfb6j3EiAdc2qSldwBREgvSR"}} \ No newline at end of file diff --git a/build/torch215-cu130-x86_64-linux/symbols.json b/build/torch215-cu130-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch215-cu130-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch215-cu132-x86_64-linux/__init__.py b/build/torch215-cu132-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch215-cu132-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch215-cu132-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch215-cu132-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..e8237d0eaa4cff633cf8efa024ecd4e25d68fed3 --- /dev/null +++ b/build/torch215-cu132-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b12303d40e505fb72cff460806d2f3b06b1ab4544c300963c0d6a67f54f94496 +size 1220023872 diff --git a/build/torch215-cu132-x86_64-linux/_ops.py b/build/torch215-cu132-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch215-cu132-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch215-cu132-x86_64-linux/layers.py b/build/torch215-cu132-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch215-cu132-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch215-cu132-x86_64-linux/metadata.json b/build/torch215-cu132-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..87ae025f4797db76394c01de7738ba4b6c2480e7 --- /dev/null +++ b/build/torch215-cu132-x86_64-linux/metadata.json @@ -0,0 +1,40 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "10.0", + "12.0", + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "sSMD1A5QX7cs/0YIBtLzsGsatFRMMAljwNamf1T5RJY=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch215-cu132-x86_64-linux/metadata.json.sigstore b/build/torch215-cu132-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..96ceda2369b179a5c5aedf4d1cb22855734c5a73 --- /dev/null +++ b/build/torch215-cu132-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"MIIHKzCCBrKgAwIBAgIUIHCW75qfMha2Tua1DoBXtAS+TLEwCgYIKoZIzj0EAwMwNzEVMBMGA1UEChMMc2lnc3RvcmUuZGV2MR4wHAYDVQQDExVzaWdzdG9yZS1pbnRlcm1lZGlhdGUwHhcNMjYxMDE5MTcxMTI5WhcNMjYxMDE5MTcyMTI5WjAAMFkwEwYHKoZIzj0CAQYIKoZIzj0DAQcDQgAEkfiYZbX8dXXoWskfyZXpycfdbf9/7288zNEsSODmvUh05DsdpRwekG+Am3uroyWgWabkgq0AKdigDG30WmHlx6OCBdEwggXNMA4GA1UdDwEB/wQEAwIHgDATBgNVHSUEDDAKBggrBgEFBQcDAzAfBgNVHSMEGDAWgBTf0+nPViQRlvmo2OkoVaLGLhhkPzBrBgNVHREBAf8EYTBfhl1odHRwczovL2dpdGh1Yi5jb20vaHVnZ2luZ2ZhY2Uva2VybmVscy1jb21tdW5pdHkvLmdpdGh1Yi93b3JrZmxvd3MvYnVpbGQueWFtbEByZWZzL2hlYWRzL21haW4wOQYKKwYBBAGDvzABAQQraHR0cHM6Ly90b2tlbi5hY3Rpb25zLmdpdGh1YnVzZXJjb250ZW50LmNvbTAfBgorBgEEAYO/MAECBBF3b3JrZmxvd19kaXNwYXRjaDA2BgorBgEEAYO/MAEDBChmZTJiYjY2ZmVlNmMyYTc3ZWNjZjA4YTA1MDhkYzU3NDI4ZDQ4OGVkMBMGCisGAQQBg78wAQQEBUJ1aWxkMCsGCisGAQQBg78wAQUEHWh1Z2dpbmdmYWNlL2tlcm5lbHMtY29tbXVuaXR5MB0GCisGAQQBg78wAQYED3JlZnMvaGVhZHMvbWFpbjA7BgorBgEEAYO/MAEIBC0MK2h0dHBzOi8vdG9rZW4uYWN0aW9ucy5naXRodWJ1c2VyY29udGVudC5jb20wbQYKKwYBBAGDvzABCQRfDF1odHRwczovL2dpdGh1Yi5jb20vaHVnZ2luZ2ZhY2Uva2VybmVscy1jb21tdW5pdHkvLmdpdGh1Yi93b3JrZmxvd3MvYnVpbGQueWFtbEByZWZzL2hlYWRzL21haW4wOAYKKwYBBAGDvzABCgQqDChmZTJiYjY2ZmVlNmMyYTc3ZWNjZjA4YTA1MDhkYzU3NDI4ZDQ4OGVkMBsGCisGAQQBg78wAQsEDQwLc2VsZi1ob3N0ZWQwQAYKKwYBBAGDvzABDAQyDDBodHRwczovL2dpdGh1Yi5jb20vaHVnZ2luZ2ZhY2Uva2VybmVscy1jb21tdW5pdHkwOAYKKwYBBAGDvzABDQQqDChmZTJiYjY2ZmVlNmMyYTc3ZWNjZjA4YTA1MDhkYzU3NDI4ZDQ4OGVkMB8GCisGAQQBg78wAQ4EEQwPcmVmcy9oZWFkcy9tYWluMBoGCisGAQQBg78wAQ8EDAwKMTA3MTQ3NTUyOTAuBgorBgEEAYO/MAEQBCAMHmh0dHBzOi8vZ2l0aHViLmNvbS9odWdnaW5nZmFjZTAYBgorBgEEAYO/MAERBAoMCDI1NzIwNzQzMG0GCisGAQQBg78wARIEXwxdaHR0cHM6Ly9naXRodWIuY29tL2h1Z2dpbmdmYWNlL2tlcm5lbHMtY29tbXVuaXR5Ly5naXRodWIvd29ya2Zsb3dzL2J1aWxkLnlhbWxAcmVmcy9oZWFkcy9tYWluMDgGCisGAQQBg78wARMEKgwoZmUyYmI2NmZlZTZjMmE3N2VjY2YwOGEwNTA4ZGM1NzQyOGQ0ODhlZDAhBgorBgEEAYO/MAEUBBMMEXdvcmtmbG93X2Rpc3BhdGNoMGQGCisGAQQBg78wARUEVgxUaHR0cHM6Ly9naXRodWIuY29tL2h1Z2dpbmdmYWNlL2tlcm5lbHMtY29tbXVuaXR5L2FjdGlvbnMvcnVucy8zOTIzOTU4Nzk4NS9hdHRlbXB0cy8xMBYGCisGAQQBg78wARYECAwGcHVibGljMEYGCisGAQQBg78wARgEOAw2cmVwbzpodWdnaW5nZmFjZS9rZXJuZWxzLWNvbW11bml0eTpyZWY6cmVmcy9oZWFkcy9tYWluMIGKBgorBgEEAdZ5AgQCBHwEegB4AHYA3T0wasbHETJjGR4cmWc3AqJKXrjePK3/h4pygC8p7o4AAAGhVSWY1AAABAMARzBFAiB/o/bzpxEg5yJRWlxvEAGWmTWbrBOWOWMC07VoxL4YmQIhAPkjvEZ5ZLX2QDWdWHa1Q4iThKbm2UyNduSQAPCuNo+8MAoGCCqGSM49BAMDA2cAMGQCMC+yRRhvL4lZjk0F4GqKst29+Oobqc5HVuOYiEFeRHPDue97MHaezNLhw2EDqKVAfgIwOVnJgbuXmOsBSX/xCR9A+xV0L+O/PN6sBfLFpgH3y6k/75YAIBWFMPf/fpn6X9WT"}, "tlogEntries":[{"logIndex":"3323917484", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429889", "inclusionPromise":{"signedEntryTimestamp":"MEYCIQCs4SxieDYLeJll3wr/lxqHjwGiTmp8E2wUsBHNNIWqfQIhAK5HVZHLRaAtOdUaXd3X1/7jgx2pKoF0Eq2Bmvz3FrVU"}, "inclusionProof":{"logIndex":"3202013222", "rootHash":"SBmKvmuwI4TM8hnhRmb2Pt5GFPhSiyr/j3WAcUdp9EU=", "treeSize":"3202013302", "hashes":["ugy5r5rsOB4kGKdBT0LGNa67UPOzbwDa2X6vKPKnAwY=", "SpjHYL69QlK2bfUOapYf30AbdimC8Xe7INvaKzeQfY4=", "2x/WA6BizU56LiGKdPqh/w2SZdVqysaIkiJjAiDHVrc=", "RgSkyJbV4KpxZzcsMUvLZCeQTCsV4YJuvCwUToAfrU8=", "if7Ah5l8n1RxNpBA4USWz76LIb4AUX3aV+gKeobF/Ac=", "cuieRBpqGn+jYxOez7X46ihGIgPnNl18ycrIc46fIdY=", "r0CtKyVn70Qe8Q7BPedhYww2dM/bYiFJux9juN0NWEo=", "PHGbo2WyIyVQqY8QuKi7/3rJG+1gQJmkeq/V5JmmTpg=", "X1OwG1pW/NAf1zSALYc7dJD4MRPAS3+RElCmZVkRTNQ=", "ZoEKTEi+vNH82Rv/iMhp5QJLlu/Ocy2iHNm45P+eFP8=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202013302\nSBmKvmuwI4TM8hnhRmb2Pt5GFPhSiyr/j3WAcUdp9EU=\n\n— rekor.sigstore.dev wNI9ajBEAiASQ4efpr7L5GYoTashB7oAgMC3Splved/BmFDGFSz6jQIgIvc25e1P6Mj6aMlUlNZDDg5Y3dMtDFS4uyI1ijxNeSE=\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"MIICyDADAgEAMIICvwYJKoZIhvcNAQcCoIICsDCCAqwCAQMxDTALBglghkgBZQMEAgEwgbcGCyqGSIb3DQEJEAEEoIGnBIGkMIGhAgEBBgkrBgEEAYO/MAIwMTANBglghkgBZQMEAgEFAAQgmTvNc3GH3lxNLbhfOOvot1Z6w/iObtTT8+2fgzfRcoICFDapzhIINThwny2fZzG/GP16fuItGA8yMDI2MTAxOTE3MTEyOVowAwIBAaAypDAwLjEVMBMGA1UEChMMc2lnc3RvcmUuZGV2MRUwEwYDVQQDEwxzaWdzdG9yZS10c2GgADGCAdowggHWAgEBMFEwOTEVMBMGA1UEChMMc2lnc3RvcmUuZGV2MSAwHgYDVQQDExdzaWdzdG9yZS10c2Etc2VsZnNpZ25lZAIUOhNULwyQYe68wUMvy4qOiyojiwwwCwYJYIZIAWUDBAIBoIH8MBoGCSqGSIb3DQEJAzENBgsqhkiG9w0BCRABBDAcBgkqhkiG9w0BCQUxDxcNMjYxMDE5MTcxMTI5WjAvBgkqhkiG9w0BCQQxIgQgrDQ8X+rTncNPFewX0es7aWc9xWaH5sF0Dwo8IUepsn4wgY4GCyqGSIb3DQEJEAIvMX8wfTB7MHkEIIX5J7wHq2LKw7RDVsEO/IGyxog/2nq55thw2dE6zQW3MFUwPaQ7MDkxFTATBgNVBAoTDHNpZ3N0b3JlLmRldjEgMB4GA1UEAxMXc2lnc3RvcmUtdHNhLXNlbGZzaWduZWQCFDoTVC8MkGHuvMFDL8uKjosqI4sMMAoGCCqGSM49BAMCBGYwZAIwejfzbwQcQn9jMg3knd+CCQ/b1JIWVvmfeazx+sYU/aDuWU7JKuQX3Rjtnz7MkxJ6AjBcORuHqDT3gYT1HK1J96Q5H6h5T4KdeayNcfIo6FzleQXa3iS77hSbfxST2Cext88="}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"I5hjYdWyVK+5xmLM2BiyG9Z0AHJSozdcWkalsBc4iJI="}, "signature":"MEQCICrFF6KtRvGC/Hx3Ng98+8yApd4RAAVa5BKD0HcukSmsAiAyCqm2qi8sWuX7lNNaJpdb4ct9NMMBuEdyCiMm4fCRBA=="}} \ No newline at end of file diff --git a/build/torch215-cu132-x86_64-linux/symbols.json b/build/torch215-cu132-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch215-cu132-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +} diff --git a/build/torch215-cu134-x86_64-linux/__init__.py b/build/torch215-cu134-x86_64-linux/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..946160f16b9dc91fefeea037fb7ac84fd6afd802 --- /dev/null +++ b/build/torch215-cu134-x86_64-linux/__init__.py @@ -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", +] \ No newline at end of file diff --git a/build/torch215-cu134-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so b/build/torch215-cu134-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so new file mode 100644 index 0000000000000000000000000000000000000000..21ec3e109c3159fa55f53990a268889353a37372 --- /dev/null +++ b/build/torch215-cu134-x86_64-linux/_layer_norm_cuda_fe2bb66.abi3.so @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bf6fbdca4a2928bbef41bf361b7d009810d53ce88da563de188c89d7627e0251 +size 1228462312 diff --git a/build/torch215-cu134-x86_64-linux/_ops.py b/build/torch215-cu134-x86_64-linux/_ops.py new file mode 100644 index 0000000000000000000000000000000000000000..425d1f9d89080e874561a0316e5486dab6ee8bfa --- /dev/null +++ b/build/torch215-cu134-x86_64-linux/_ops.py @@ -0,0 +1,9 @@ +import torch +from . import _layer_norm_cuda_fe2bb66 +ops = torch.ops._layer_norm_cuda_fe2bb66 + +def add_op_namespace_prefix(op_name: str): + """ + Prefix op by namespace. + """ + return f"_layer_norm_cuda_fe2bb66::{op_name}" diff --git a/build/torch215-cu134-x86_64-linux/layers.py b/build/torch215-cu134-x86_64-linux/layers.py new file mode 100644 index 0000000000000000000000000000000000000000..7ed883f42ead452f8b60f498ec11302c53d3cf74 --- /dev/null +++ b/build/torch215-cu134-x86_64-linux/layers.py @@ -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) \ No newline at end of file diff --git a/build/torch215-cu134-x86_64-linux/metadata.json b/build/torch215-cu134-x86_64-linux/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..ebaec9b23dac71cda3e56c47ebe37ce45f560f17 --- /dev/null +++ b/build/torch215-cu134-x86_64-linux/metadata.json @@ -0,0 +1,40 @@ +{ + "name": "layer-norm", + "id": "_layer_norm_cuda_fe2bb66", + "version": 1, + "kernels-minver": "0.14.0", + "license": "BSD-3-Clause", + "python-depends": [], + "kernel-depends": [], + "backend": { + "type": "cuda", + "archs": [ + "10.0", + "12.0", + "8.0", + "8.9", + "9.0" + ] + }, + "digest": { + "algorithm": "sha256", + "files": { + "__init__.py": "jDU4y5fj5PX2zHRl093nsBRJeljX4k0cH3xhCkY8TII=", + "_layer_norm_cuda_fe2bb66.abi3.so": "v2+9ykopKLvvQb82G30AmBDVPOiNpWPeGIyJ12J+AlE=", + "_ops.py": "Bz98v8Ry/7lpJVStlnEKKEvzErnRTgXwlW+sroJ2+tM=", + "layers.py": "6MG0mvDTvgmjgeOE8+zQHno3oudtLhqdXVh5x8a+2fo=", + "symbols.json": "uzzPh37RmlWx6HBnBUyTrknEd17yJD0FN+qjhwr3unI=" + } + }, + "provenance": { + "kernel-builder": { + "version": "0.18.0-dev0", + "commit": "d054906a13710b811b96697b4635f03020842733", + "dirty": false + }, + "kernel": { + "commit": "fe2bb66fee6c2a77eccf08a0508dc57428d488ed", + "dirty": false + } + } +} \ No newline at end of file diff --git a/build/torch215-cu134-x86_64-linux/metadata.json.sigstore b/build/torch215-cu134-x86_64-linux/metadata.json.sigstore new file mode 100644 index 0000000000000000000000000000000000000000..db887d4b0015121dcf1c3b7d58af153594f676b2 --- /dev/null +++ b/build/torch215-cu134-x86_64-linux/metadata.json.sigstore @@ -0,0 +1 @@ +{"mediaType":"application/vnd.dev.sigstore.bundle.v0.3+json", "verificationMaterial":{"certificate":{"rawBytes":"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"}, "tlogEntries":[{"logIndex":"3323917737", "logId":{"keyId":"wNI9atQGlz+VWfO6LRygH4QUfY/8W4RFwiT5i5WRgB0="}, "kindVersion":{"kind":"hashedrekord", "version":"0.0.1"}, "integratedTime":"1792429891", "inclusionPromise":{"signedEntryTimestamp":"MEUCIFGRUNnubn0bdAWCZ3QTem/dnlsiu/DlMLj9qgOJwtviAiEAvRZ4FlJUw0XFlRi/D9sKLHNK7+p0mNLfOnG7zkw76w0="}, "inclusionProof":{"logIndex":"3202013475", "rootHash":"S84GqfSffRNrFMX4GgDWq3Z9vKYvNin2wG6yraSylA0=", "treeSize":"3202013822", "hashes":["TJ9dQ7F1Cv3YGYP20VWsIhQevjcra95gXy131R31w5I=", "8G+SfDo0C21bW6kwE+2T3YHRmicGFI2gnpsNkwj0B7A=", "YHta20bj2MZx3F2vsW3uOiAE0GDnMS+QPPqCM1kOWuU=", "p1AVmeePoJSgXdjDBERsZkTzY0Cwe0PttkpzoJStIUY=", "vTXxwppvV+bl1LnNkhM11UnzZGQsdPfd1iObfLo2Jus=", "fnmTlD8zJ6SizAhaK5c6uHiKU8BIY3WAEhYxinZgF5s=", "YvC3hLkf0oVg6LjcNuf6sdHJqqPUeYPfbL4WIfky52g=", "rEUMPHuKzsSsMkiNCuCISTM7oHlV6Ux3LCIDmamZ1Oc=", "EirorqpYZTq1EhK4u28OkkzWv1z9gsH0PJEEsEAc1ew=", "U34PxYgyuMAYvPA6HPi8QDiSMv80yt3NDx45E64jeMQ=", "PHGbo2WyIyVQqY8QuKi7/3rJG+1gQJmkeq/V5JmmTpg=", "X1OwG1pW/NAf1zSALYc7dJD4MRPAS3+RElCmZVkRTNQ=", "ZoEKTEi+vNH82Rv/iMhp5QJLlu/Ocy2iHNm45P+eFP8=", "Fmu5dEaq836T8rKgvh+xxNtPcazEy+AMz4HOuyONfho=", "EA/gesnuWcmqzCu/bYVmY/gXaHfv+DBoAgaVMuR/WMg=", "MD4OgR/R/G+2HawrFMhrsxg9GvdgOeQsIipPmrKXl38=", "j0WYhlZCgHWFR/OJIjCQT0Q8ta+HnlghbtUUUTyZptI=", "7WFpoaiQCzLwM3o2YI5Ngo8J7JD0OfyCCQtmXhWxVes=", "nyQV/+cWQUG5Hb2PKjQQNkQIJmz2qzAZKBQlkHGIknM=", "G+nk7mgE5h6dUmv2UY7NDJrY3WvrZY31raaO1AETt+o=", "qlTWNDzERnTBQVcHHXXVPjiUP0SsikGxfcxVlgqVFDo=", "gg9GNApNke3+zNhl4tsWtmIycCFrOMCB3lys6XuSQNQ=", "NlOclldmKPA1tMKIskqHAwuQDYHp31YwpK+OOMrMccc=", "qxzHanAzz57SDdmJe0B7bJK72NTIBbwEMGBBKvDOROw=", "xH/DCseLHr9eKoYT8qsORZK7zVdEGYWHuVtsVrD95wY="], "checkpoint":{"envelope":"rekor.sigstore.dev - 1193050959916656506\n3202013822\nS84GqfSffRNrFMX4GgDWq3Z9vKYvNin2wG6yraSylA0=\n\n— rekor.sigstore.dev wNI9ajBFAiAXeoiUfCGXHY1uVZ7K8IBXdNgZKn6PLFY43eB/EazawgIhAJnar1NqP4qPWrAu5qIhgmIYtM9r7uxhrQqrOSKAFGsF\n"}}, "canonicalizedBody":"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"}], "timestampVerificationData":{"rfc3161Timestamps":[{"signedTimestamp":"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"}]}}, "messageSignature":{"messageDigest":{"algorithm":"SHA2_256", "digest":"ugmtqofD2iGKUu1x9XnU6vgz3B1NrdGs9Uir96Da0g4="}, "signature":"MEUCIQCIjO2b1Q89Pry22QLhQatFmZIcPuRragRrKsi5SxF9VwIgT8m139/DjsvrJsP8+wkhQ/a5CZotYUevhS8XFkYU3YM="}} \ No newline at end of file diff --git a/build/torch215-cu134-x86_64-linux/symbols.json b/build/torch215-cu134-x86_64-linux/symbols.json new file mode 100644 index 0000000000000000000000000000000000000000..b5710aa9803e3d5004b684884a8c1b84d575ff1f --- /dev/null +++ b/build/torch215-cu134-x86_64-linux/symbols.json @@ -0,0 +1,361 @@ +{ + "schema_version": 1, + "module": "_layer_norm_cuda_fe2bb66", + "functions": [ + { + "name": "dropout_add_ln_fwd", + "qualname": "dropout_add_ln_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_bwd", + "qualname": "dropout_add_ln_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "colscale", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "z_subset", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rowscale_const", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x0_numrows", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_fwd", + "qualname": "dropout_add_ln_parallel_residual_fwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "input", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "beta1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "epsilon", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gen", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "residual_in_fp32", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + }, + { + "name": "dropout_add_ln_parallel_residual_bwd", + "qualname": "dropout_add_ln_parallel_residual_bwd", + "module": "_layer_norm_cuda_fe2bb66", + "kind": "function", + "doc": null, + "signature": { + "parameters": [ + { + "name": "dz0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dz1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dx", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "x", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "mu", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "rsigma", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma0", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "gamma1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "dropout_p", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_x1", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "has_residual", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + }, + { + "name": "is_rms_norm", + "kind": "POSITIONAL_OR_KEYWORD", + "default": null, + "annotation": null + } + ], + "return_annotation": null + } + } + ], + "layers": [] +}