Ask a Question Question Hi, i want to try convert the opset version of my onnx model. I want to convert from 11 to 10. But its failed because there is no gather RuntimeError: /github/workspace/onnx/version_converter/BaseConverter.h:66: adapter_lookup: Assertion `false` failed: No Adapter From Versio...

System information main branch What is the problem that this feature solves? I'm working on Issue #4160 through PR #5263 . Current design works fine for Op version > 5. For Op version <= 5. I found the following error: I looks like the Reshape creation arguments do not fit for Op version <...

Ask a Question Question Why does the ONNX specification state that the scale and bias in GroupNormalization should be equivalent to the number of groups, while the GroupNorm paper suggests that they should have the shape of the number of channels? Further information I see the _group_normalization f...

Bug Report https://github.com/onnx/onnx/blob/main/docs/Operators.md#Einsum Einsum An einsum of the form term1, term2 -> output-term produces an output tensor using the following equation output[output-term] = reduce-sum( input1[term1] * input2[term] ) <<<<need...

Bug Report Is the issue related to model conversion? No Describe the bug ONNX spec should explicitly say if each integer attribute value can be negative. For example, the conv operator has 6 attributes: auto_pad, dilations , group, kernel_shape, pads , strides. They are stored as signed values. It...

Hi! I am strugling with build of protobuf+onnx inside my project. I want to build for Windows either: static protobuf, static onnx or static protobuf, dynamic onnx and then use some model loading inside my code of course. For example I just define onnx::ModelProto in main.cpp. Mode is Release, con...

Ask a Question Question I have a model that outputs masks, then there's a postprocessing part to get the final output from those masks. There is two types of postprocessing depending on a boolean value split. I first exported the model from pytorch as onnx, I then exported the two postprocessings. A...

Ask a Question Does the output generated by the last slide of window divided by remaining pixels or kernal size when ceil_mode is True. Question I found in current ORT, when ceil mode is on, the last output from the last slide of window is divided by kernal size regardless the remaining pixels (incl...

Ask a Question Question I have two onnx models i exported from pytorch. I named them split.onnx and nosplit.onnx. They make some sort of postprocessing. i want to have a single ONNX model with 2 computation graphs by using an if node on top of the models.the Then branch is the graph of split, and th...

Ask a Question Question import onnx from onnx import helper import numpy as np def make_col2im(): input = helper.make_tensor_value_info("input", onnx.TensorProto.FLOAT, [1, 5, 5]) image_shape = helper.make_tensor_value_info("image_shape", onnx.TensorProto.INT32, [5, 5]) block_shape = helper.make_te...