daquexian

Sorry for the late reply. Could you please try simplifying your onnx model by https://github.com/daquexian/onnx-simplifier? onnx-simplifier will eliminate most Shape operators.

For linux I prefer AppImage

Are you using cmake? :) ONNX_NAMESPACE is defined in cmake via compile definitions

could you please provide your system info like OS version, cmake version and the commands you used for building onnx2dnn?

Thanks for your reply. Sorry for the late response. It works, but not so perfectly. Now I hacked the initialize method: https://github.com/daquexian/tagls/blob/c5a826b8e7945d53bd29c6b1819a3e025955ba4e/tagls/server.py line 139, 204 and 235. IMO it works...

Hi @i4oolish , thanks for your contribution. However, I'm afraid I do not understand what this PR does. It is completely ok that the behavior of opset16 is not the...

BTW, [cibuildwheel](https://cibuildwheel.readthedocs.io/en/stable/) is a great tool to generate Python packages with c++ library because it handles different python versions and architectures and is easy to use. I have successfully used...

> @daquexian do you have bandwidth to add this wheel support before next ONNX release (probably release in Nov. or Dec.)? If not, I can take this work item. Sorry...

这个和 torch.nn.GELU(approximate='tanh') 是重复的吗 https://pytorch.org/docs/stable/generated/torch.nn.GELU.html 相关讨论：https://github.com/huggingface/transformers/issues/15397

从文档来看确实是和 torch.nn.GELU(approximate='tanh') 公式一样的。 oneflow 文档（渲染好像不太正确，以及混用了 input 和 x）：  pytorch 文档（pytorch 文档也没正确渲染 lol）： oneflow 文档里 input 和 x 应该是同一个东西？把 input 换成 x 并放进括号里就和 pytorch 公式一样了。这么来看上面贴的 NewGELUActivation 和 FastGELUActivation 也是一样的。