Can I install PyTorch/XLA nightly build using conda instead of pip

Installing PyTorch Nightly with Conda

To install the PyTorch nightly build using conda, you can use the following command:

bash
conda install pytorch torchvision cudatoolkit=10.0 -c pytorch-nightly

This command installs PyTorch and torchvision from the nightly channel, which includes the latest development versions[1].

Installing PyTorch/XLA Nightly with Pip

For PyTorch/XLA, you typically need to use pip to install the nightly builds. You can download the wheels directly from Google Cloud Storage and install them using pip. Here's an example command for installing a PyTorch/XLA nightly build:

bash
pip install https://storage.googleapis.com/pytorch-xla-releases/wheels/tpuvm/torch-nightly+20240601-cp310-cp310-linux_x86_64.whl
pip install https://storage.googleapis.com/pytorch-xla-releases/wheels/tpuvm/torch_xla-nightly+20240601-cp310-cp310-linux_x86_64.whl

Replace `+20240601` with the desired date for the nightly build you want to install[3].

Conclusion

While you can install PyTorch nightly builds using conda, PyTorch/XLA nightly builds are generally installed via pip due to their specific distribution method. If you need to use PyTorch/XLA nightly builds, using pip is the recommended approach.