Hi, In my opinion, the DCO leads to problems every now and then, and even experienced Onnx contributors forget about it from time to time. As I currently see it, it is at least that online when changing via the github online interface. I would suggest to turn on "require-sign-off-on-web-based-commit...

At least StableHLO shares part of the vision of ONNX, which were designed as "multiple-layers" operators though. :-)

System information No response What is the problem that this feature solves? UBSan finds undefined behavior in pytorch/pytorch#93174. We can use it in our CI (maybe)? https://clang.llvm.org/docs/UndefinedBehaviorSanitizer.html Alternatives considered No response Describe the feature No response Will...

Bug Report Is the issue related to model conversion? Describe the bug ONNX reference evaluator fails on the SplitToSequence node when running the gpt2 model. System information ONNX version: 1.14 Reproduction instructions Describe the code to reproduce the behavior. import onnx import numpy a...

Bug Report Is the issue related to model conversion? No Describe the bug Error in run "pip install -e" Before pip install, all are okay. cd onnx git submodule update --init --recursive # prefer lite proto set CMAKE_ARGS=-DONNX_USE_LITE_PROTO=ON In file included from /home/v-junyan1/adv_training/on...

Bug Report Is the issue related to model conversion? No Describe the bug import onnxscript import onnx from onnxscript import FLOAT from onnxscript.function_libs.torch_aten.ops.nn import aten_gelu @onnxscript.script() def expected_model(x: FLOAT[2, 3, 4]): return aten_gelu(x, approximate="tan...

Bug Report Is the issue related to model conversion? No Describe the bug Documented Gather operator's behavior is not matched w/ onnxruntime's result. Gather-1 / Gather-11 example descriptibed in Changelog.md is as follow: data = [ [1.0, 1.2, 1.9], [2.3, 3.4, 3.9], [4.5, 5.7, 5.9...

Hi, I am trying to export the torch-native PTQ model using the timm repo. I am working on exporting the deit_tiny_patch_16 model. The model conversion is however failing with runtime error: RuntimeError: ONNX symbolic expected node type prim::TupleConstruct, got `%456 : Tensor = onnx::Concat[axis=1]...

Bug Report Is the issue related to model conversion? Nope. Describe the bug Changelog.md is broken below the section "Version 15 of the default ONNX operator set", showing raw markdown (e.g. <dt><tt>epsilon</tt> : float (default is 1e-05)</dt>) after #4747. System information...

I am analyzing DFT op in ONNX (https://github.com/onnx/onnx/blob/main/docs/Operators.md#DFT) and try to run provided tests (https://github.com/onnx/onnx/blob/main/onnx/backend/test/case/node/dft.py) on ONNXRuntime (CPU) as a backend. Unfortunately, all (three of them) fail with mismatch (about 5%)....