pytorch · hamzaqureshi5 · May 16, 2025 · May 16, 2025 · May 16, 2025 · May 16, 2025
diff --git a/third_party/gloo b/third_party/gloo
diff --git a/third_party/googletest b/third_party/googletest
@@ -169,7 +169,8 @@ Node* createONNXUnsqueeze(
   Node* unsqueeze_node = graph->create(onnx::Unsqueeze, 1);
   unsqueeze_node->addInput(input);
   unsqueeze_node->insertBefore(n_to_insert_before);
-  if (opset_version >= OPSET_VERSION_13) {
+  if (opset_version >= OPSET_VERSION_13) 
+  {
     // ONNX spec sets `axes` as input for opset >= 13.
     Node* unsqueeze_axes = graph->create(onnx::Constant, 1);
     unsqueeze_axes->insertBefore(unsqueeze_node);

@@ -24,6 +24,7 @@
     "symbolic_opset18",
     "symbolic_opset19",
     "symbolic_opset20",
+    "symbolic_opset21",
     # Enums
     "OperatorExportTypes",
     "TrainingMode",
@@ -90,6 +91,9 @@
     symbolic_opset18,
     symbolic_opset19,
     symbolic_opset20,
+    symbolic_opset21,
+    symbolic_opset22,
+    symbolic_opset23,
     utils,
 )
 

@@ -4,8 +4,8 @@
 
 ONNX_BASE_OPSET = 9
 ONNX_MIN_OPSET = 7
-ONNX_MAX_OPSET = 20
-ONNX_TORCHSCRIPT_EXPORTER_MAX_OPSET = 20
+ONNX_MAX_OPSET = 23
+ONNX_TORCHSCRIPT_EXPORTER_MAX_OPSET = 23
 # ONNX_DEFAULT_OPSET generated by tools/onnx/update_default_opset_version.py
 ONNX_DEFAULT_OPSET = 17
 ONNX_CONSTANT_FOLDING_MIN_OPSET = 9

@@ -0,0 +1,37 @@
+# mypy: allow-untyped-defs
+"""This file exports ONNX ops for opset 21.
+
+Note [ONNX Operators that are added/updated in opset 21]
+
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+https://github.com/onnx/onnx/blob/main/docs/Changelog.md#version-21-of-the-default-onnx-operator-set
+New operators:
+    - Gelu
+"""
+
+import functools
+
+import torch.nn.functional as F
+from torch import _C
+from torch.onnx import symbolic_helper
+from torch.onnx._internal import jit_utils, registration
+
+# EDITING THIS FILE? READ THIS FIRST!
+# see Note [Edit Symbolic Files] in symbolic_helper.py
+
+__all__ = [
+    "gelu",
+]
+
+def convert_grid_sample_mode(mode_s):
+    return (
+        "linear" if mode_s == "bilinear" else "cubic" if mode_s == "bicubic" else mode_s
+    )
+
+_onnx_symbolic = functools.partial(registration.onnx_symbolic, opset=21)
+
+
+@_onnx_symbolic("aten::gelu")
+@symbolic_helper.parse_args("v", "s")
+def gelu(g: jit_utils.GraphContext, self: _C.Value, approximate: str = "none"):
+    return g.op("Gelu", self, approximate_s=approximate)
@@ -0,0 +1,31 @@
+# mypy: allow-untyped-defs
+"""This file exports ONNX ops for opset 22.
+
+Note [ONNX Operators that are added/updated in opset 22]
+
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+https://github.com/onnx/onnx/blob/main/docs/Changelog.md#version-22-of-the-default-onnx-operator-set
+New operators:
+    Selu
+"""
+
+import functools
+import torch
+from torch import _C
+from torch.onnx import symbolic_helper
+from torch.onnx._internal import jit_utils, registration
+
+__all__ = [
+    "selu",
+]
+
+_onnx_symbolic = functools.partial(registration.onnx_symbolic, opset=22)
+
+@_onnx_symbolic("aten::selu")
+@symbolic_helper.parse_args("v")
+def selu(g: jit_utils.GraphContext, self: _C.Value):
+    # Use default alpha and gamma values as per ONNX Selu-22 spec
+    alpha = 1.6732632
+    gamma = 1.0507
+    return g.op("Selu", self, alpha_f=alpha, gamma_f=gamma)
+
@@ -0,0 +1,93 @@
+# mypy: allow-untyped-defs
+"""This file exports ONNX ops for opset 23.
+
+Note [ONNX Operators that are added/updated in opset 23]
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+https://github.com/onnx/onnx/blob/main/docs/Changelog.md#version-23-of-the-default-onnx-operator-set
+New operators:
+    Attention
+    RMSNormalization
+    Reshape
+    RotaryEmbedding
+"""
+
+import functools
+from torch import _C
+from torch.onnx import symbolic_helper
+from torch.onnx._internal import jit_utils, registration
+
+__all__ = ["attention", "rms_normalization", "reshape", "rotary_embedding"]
+
+_onnx_symbolic = functools.partial(registration.onnx_symbolic, opset=23)
+
+
+@_onnx_symbolic("aten::attention")
+@symbolic_helper.parse_args("v", "v", "v", "v", "v", "v", "i", "i", "i", "f", "f", "i")
+def attention(
+    g: jit_utils.GraphContext,
+    q: _C.Value,
+    k: _C.Value,
+    v: _C.Value,
+    attn_mask: _C.Value,
+    past_key: _C.Value,
+    past_value: _C.Value,
+    q_num_heads: int,
+    kv_num_heads: int,
+    qk_matmul_output_mode: int,
+    scale: float,
+    softcap: float,
+    softmax_precision: int,
+):
+    inputs = [q, k, v]
+    if attn_mask.node().kind() != "prim::Constant" or attn_mask.type().kind() != "NoneType":
+        inputs.append(attn_mask)
+    if past_key.node().kind() != "prim::Constant" or past_key.type().kind() != "NoneType":
+        inputs.append(past_key)
+    if past_value.node().kind() != "prim::Constant" or past_value.type().kind() != "NoneType":
+        inputs.append(past_value)
+
+    return g.op(
+        "Attention",
+        *inputs,
+        q_num_heads_i=q_num_heads,
+        kv_num_heads_i=kv_num_heads,
+        qk_matmul_output_mode_i=qk_matmul_output_mode,
+        scale_f=scale,
+        softcap_f=softcap,
+        softmax_precision_i=softmax_precision,
+        outputs=4,
+    )
+
+
+@_onnx_symbolic("aten::rms_norm")
+@symbolic_helper.parse_args("v", "v", "i", "f")
+def rms_normalization(g, input, scale, axis, epsilon):
+    squared = g.op("Mul", input, input)
+    rank = symbolic_helper._get_tensor_rank(input)
+    axes = list(range(axis if axis >= 0 else rank + axis, rank)) # type: ignore
+    mean = g.op("ReduceMean", squared, axes_i=axes, keepdims_i=1)
+    mean_eps = g.op("Add", mean, g.op("Constant", value_t=symbolic_helper._scalar(epsilon)))
+    rms = g.op("Sqrt", mean_eps)
+    normalized = g.op("Div", input, rms)
+    return g.op("Mul", normalized, scale)
+
+
+@_onnx_symbolic("aten::reshape")
+@symbolic_helper.parse_args("v", "v")
+def reshape(g, input, shape):
+    return g.op("Reshape", input, shape)
+
+
+@_onnx_symbolic("aten::rotary_embedding")
+@symbolic_helper.parse_args("v", "v", "v", "v", "i", "i", "i")
+def rotary_embedding(g, input, position_ids, sin_cache, cos_cache, interleaved, rotary_embedding_dim, num_heads):
+    return g.op(
+        "RotaryEmbedding",
+        input,
+        position_ids,
+        sin_cache,
+        cos_cache,
+        interleaved_i=interleaved,
+        rotary_embedding_dim_i=rotary_embedding_dim,
+        num_heads_i=num_heads
+    )
@@ -353,7 +353,7 @@ def export(
 
                     Models exported this way are probably runnable only by Caffe2.
 
-        opset_version (int, default 17): The version of the
+        opset_version (int, default 21): The version of the
             `default (ai.onnx) opset <https://github.com/onnx/onnx/blob/master/docs/Operators.md>`_
             to target. Must be >= 7 and <= 17.
         do_constant_folding: Apply the constant-folding optimization.
@@ -1393,7 +1393,7 @@ def _export(
     if opset_version is None:
         opset_version = _constants.ONNX_DEFAULT_OPSET
 
-    # torch.onnx.export does not support opset versions >=18
+    # torch.onnx.export does not support opset versions >=21
     if opset_version > _constants.ONNX_TORCHSCRIPT_EXPORTER_MAX_OPSET:
         # We do not want to fail because we should still allow users to create
         # custom symbolic functions for opset>17
+0 −1		CMakeLists.txt
+0 −10		gloo/CMakeLists.txt
+0 −2		gloo/config.h.in
+0 −9		gloo/cuda.cu
+0 −4		gloo/cuda_allreduce_bcube.cc
+0 −4		gloo/cuda_allreduce_halving_doubling.cc
+0 −4		gloo/cuda_allreduce_local.cc
+0 −4		gloo/cuda_allreduce_ring.cc
+0 −4		gloo/cuda_allreduce_ring_chunked.cc
+0 −4		gloo/cuda_broadcast_one_to_all.cc
+0 −4		gloo/cuda_private.h
+13 −27		BUILD.bazel
+2 −2		CMakeLists.txt
+21 −28		MODULE.bazel
+12 −3		README.md
+10 −39		WORKSPACE
+29 −47		ci/linux-presubmit.sh
+4 −7		ci/macos-presubmit.sh
+13 −25		ci/windows-presubmit.bat
+50 −70		docs/advanced.md
+13 −0		docs/faq.md
+20 −53		docs/gmock_cook_book.md
+2 −2		docs/primer.md
+30 −23		docs/quickstart-bazel.md
+1 −2		docs/reference/actions.md
+2 −9		docs/reference/assertions.md
+4 −7		docs/reference/matchers.md
+1 −98		docs/reference/testing.md
+4 −32		fake_fuchsia_sdk.bzl
+18 −101		googlemock/include/gmock/gmock-actions.h
+76 −343		googlemock/include/gmock/gmock-matchers.h
+8 −8		googlemock/include/gmock/gmock-more-actions.h
+6 −4		googlemock/include/gmock/gmock-spec-builders.h
+5 −0		googlemock/include/gmock/internal/gmock-internal-utils.h
+0 −1		googlemock/include/gmock/internal/gmock-port.h
+4 −4		googlemock/src/gmock-cardinalities.cc
+7 −71		googlemock/test/gmock-actions_test.cc
+2 −2		googlemock/test/gmock-function-mocker_test.cc
+27 −210		googlemock/test/gmock-matchers-arithmetic_test.cc
+5 −131		googlemock/test/gmock-matchers-comparisons_test.cc
+19 −317		googlemock/test/gmock-matchers-containers_test.cc
+23 −82		googlemock/test/gmock-matchers-misc_test.cc
+10 −39		googlemock/test/gmock-more-actions_test.cc
+1 −1		googlemock/test/gmock-pp_test.cc
+4 −3		googlemock/test/gmock-spec-builders_test.cc
+1 −1		googlemock/test/gmock_link_test.h
+3 −3		googletest/README.md
+1 −1		googletest/cmake/internal_utils.cmake
+0 −7		googletest/include/gtest/gtest-assertion-result.h
+3 −3		googletest/include/gtest/gtest-matchers.h
+39 −95		googletest/include/gtest/gtest-param-test.h
+0 −39		googletest/include/gtest/gtest-printers.h
+65 −61		googletest/include/gtest/gtest-typed-test.h
+2 −2		googletest/include/gtest/gtest.h
+44 −40		googletest/include/gtest/internal/gtest-internal.h
+14 −48		googletest/include/gtest/internal/gtest-param-util.h
+66 −16		googletest/include/gtest/internal/gtest-port.h
+0 −4		googletest/src/gtest-internal-inl.h
+34 −157		googletest/src/gtest.cc
+17 −51		googletest/test/BUILD.bazel
+0 −38		googletest/test/googletest-fail-if-no-test-linked-test-with-disabled-test_.cc
+0 −38		googletest/test/googletest-fail-if-no-test-linked-test-with-enabled-test_.cc
+0 −169		googletest/test/googletest-fail-if-no-test-linked-test.py
+0 −19		googletest/test/googletest-filter-unittest.py
+16 −95		googletest/test/googletest-json-output-unittest.py
+0 −70		googletest/test/googletest-param-test-test.cc
+1 −31		googletest/test/googletest-printers-test.cc
+2 −2		googletest/test/googletest-setuptestsuite-test_.cc
+4 −52		googletest/test/gtest_unittest.cc
+26 −67		googletest/test/gtest_xml_output_unittest.py
+1 −21		googletest/test/gtest_xml_output_unittest_.cc
+7 −7		googletest_deps.bzl