Conversation
thiagocrepaldi
left a comment
There was a problem hiding this comment.
I would create a folder that indicates the type of sample, such as samples/training/ortmodule/nodejs, samples/training/training/orttrainer/nodejs, samples/inference
This way we can properly set code owners without conflicts
Training samples should go to https://github.com/microsoft/onnxruntime-training-examples. Does this folder have training samples? In the new repo I will follow your advice and setup code owners. For example, iOS to the mobile team and c_cxx to the people under Pranav. |
|
I found one: samples/python/pytorch_transformer. It is for training. |
|
Another one: samples/python/mnist |
|
Related PR: #8335 . Sherlock is moving the training samples to training dir. |
There are 2 for ORTTrainer pytorch forntend: python/mnist and python/pytorch_transformer. They are used for some end-to-end test |
|
Thank you @thiagocrepaldi , I excluded them from this PR and kept them untouched. |
thiagocrepaldi
left a comment
There was a problem hiding this comment.
Check with quantization team about this removal. They often check in stuff to the samples folder
Description:
Move the samples to a new repo https://github.com/microsoft/onnxruntime-inference-examples
Motivation and Context
To organize them better and setup build pipelines for them.
Also, it would be easier for us to manage open source usage registrations. Samples typically use 3rd party libraries much more. I don't want to mix them with the main library. Like, I don't want to put tensorflow into https://github.com/microsoft/onnxruntime/blob/master/ThirdPartyNotices.txt just because an ONNX Runtime sample uses it. It would confuse people who just want to import the main onnxruntime library.
Related PR: microsoft/onnxruntime-inference-examples#12