tch-ext
| Crates.io | tch-ext |
| lib.rs | tch-ext |
| version | 0.2.0 |
| source | src |
| created_at | 2023-05-18 07:23:55.307118 |
| updated_at | 2024-10-17 17:01:29.208575 |
| description | Sample Python extension using tch to interact with PyTorch. |
| homepage | |
| repository | https://github.com/LaurentMazare/tch-rs |
| max_upload_size | |
| id | 867538 |
| size | 4,486 |
Laurent Mazare (LaurentMazare)
documentation
README
Python extensions using tch
This sample crate shows how to use tch to write a Python extension
that manipulates PyTorch tensors via PyO3.
This is currently experimental hence requires some unsafe code until this has been stabilized.
In order to build the extension and test the plugin, run the following in a Python environment that has torch installed from the root of the github repo.
LIBTORCH_USE_PYTORCH=1 cargo build -p tch-ext && cp -f target/debug/libtch_ext.so tch_ext.so
python examples/python-extension/main.py
It is recommended to run the build with LIBTORCH_USE_PYTORCH set, this will
result in using the libtorch C++ library from the Python install in tch and
will ensure that this is at the proper version (having tch using a different
libtorch version from the one used by the Python runtime may result in segfaults).
Colab Notebook
tch based plugins can easily be used from colab (though it might be a bit slow
to download all the crates and compile), see this example
notebook.