-
Notifications
You must be signed in to change notification settings - Fork 90
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
PoC to test libtorch support for onnxnruntime-extensions #770
base: main
Are you sure you want to change the base?
PoC to test libtorch support for onnxnruntime-extensions #770
Conversation
38a0978
to
bb4f87a
Compare
@thiagocrepaldi please read the following Contributor License Agreement(CLA). If you agree with the CLA, please reply with the following information.
Contributor License AgreementContribution License AgreementThis Contribution License Agreement (“Agreement”) is agreed to by the party signing below (“You”),
|
bb4f87a
to
01129d3
Compare
01129d3
to
67d96f8
Compare
This PR prototypes an extension that adds the ability to
onnxruntime-extensions
to execute kernels implemented using PyTorch'slibtorch
.For this experiment, only libtorch for CPU only from https://download.pytorch.org/libtorch/nightly/cpu/libtorch-shared-with-deps-latest.zip was tested, but it should also work for GPU-enable libtorch
It is important to notice that the libtorch version (cpu/cuda) used to build the extension must match the PyTorch installed for runtime, otherwise undefined reference issues will happen
As a proof of concept, a dummy
MyRelu
op was introduced, in which it calls intotorch::ReLu
operator. This serves as a minimal example. There are implementations for both CUDA and CPU Relu, but only the CPU version has been tested.To replicate this PR on a clean environment, users must install the following as requirements:
Next, users must
export OCOS_LIBTORCH_PATH=/path/to/libtorch
before callingpip install .
from the onnxruntime-extensions repo.As an alernative, the full working environment for cpu-only can be pulled and tested from Docker hub using the following command: