search

TorchStudio / torchstudio

IDE for PyTorch and its ecosystem
https://torchstudio.ai
MIT License
378 stars 27 forks source link

Remote environment not possible #59

Closed DarkSky7 closed 6 months ago

DarkSky7 commented 8 months ago

https://www.torchstudio.ai/remoteservers/ suggests the possibility of running torch on a network server, but the dialog "TorchStudio Python Environment" doesn't allow for that possibility. It assumes that the local machine is providing python and it wants 'PyTorch' or 'PyTorch with CUDA' selected. I'm running on a Windows-10 machine (where tensorflow always lags its availability on Linux); TensorFlow is built from source to run on the CPU with optimizations: --copt=-march=native --copt=-mavx --copt=-mavx2 --copt=-mfma --copt=-msse4.2 on a nearby Fedora-38 machine; the last four build options are redundant with the first, but were included in the bazel build just in case. I don't mind storing python sources locally, but the execution environment has to be that Fedora-38 machine. When network execution supports my run-time configuration, I can begin using TorchStudio; until then it's not an option.

divideconcept commented 6 months ago

@DarkSky7 this should now be fixed in 0.9.17: https://github.com/TorchStudio/torchstudio/releases/tag/0.9.17