Guide to Install on Windows

[siddagra]

You have to use WSL2!!

In WSL2:

First you have to make sure you install torch version that matches your nvcc version:

python3 -c 'import torch; print(torch.version.cuda)'
nvcc --version

You can match the versions by installing an older version of torch which uses a different cuda version:

pip3 install -U torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu<major_version><minor_version>

then clone causal conv 1d and pip install:

git clone https://github.com/Dao-AILab/causal-conv1d.git
cd causal_conv1d
git checkout v1.0.2  # this is the highest compatible version allowed by Mamba
CAUSAL_CONV1D_FORCE_BUILD=TRUE pip3 install .

then clone mamba and pip install:

https://github.com/state-spaces/mamba.git
cd mamba
pip3 install .

[siddagra]

May be nice to add to README

[kjerk]

"How to install on windows: Use linux" is not the solution you think it is.

[siddagra]

I said use WSL2, which is on Windows. And it's better than not being able to use it on windows at all. This way you can use WSL2 on windows and make it work (which is better than the typical response "we are not sure / we do not support windows").

[kjerk]

WSL2 is a full fledged Linux VM, anything you run in it is not 'on windows', which just makes the title and explanation misleading or incorrect. It's a bandaid non-native solution that everyone already knows by default. "How to run nodejs on windows: Well first install virtualbox and download the linux binaries". No. Not the goal.

People dropping by the repo looking to "run on windows" are looking for native dependency-less support for a litany of reasons, for which there are other issues actually describing real efforts at tackling the issue #12 (comment)