k2-fsa / k2

FSA/FST algorithms, differentiable, with PyTorch compatibility.

https://k2-fsa.github.io/k2

Apache License 2.0

1.11k stars 213 forks source link

Fix building conda packages for torch 1.13 #1116

Closed csukuangfj closed 1 year ago

csukuangfj commented 1 year ago

It turns out the way to install PyTorch using conda has been changed since PyTorch >= 1.13.0

Before PyTorch 1.13.0

conda install -c pytorch -c conda-forge pytorch=1.12.1 cudatoolkit=11.6

Since PyTorch 1.13.0

conda install -c pytorch -c nvidia pytorch=1.13 pytorch-cuda=11.6

The main differences are:

(1) It uses pytorch-cuda to replace cudatoolkit
(2) the channel nvidia replaces conda-forge

Note: The way to install a CPU version of PyTorch is still the same.

csukuangfj commented 1 year ago

Usage

conda install -c k2-fsa -c pytorch -c nvidia k2 pytorch=1.13.0 pytorch-cuda=11.6 python=3.8

conda install -c k2-fsa -c pytorch -c nvidia k2 pytorch=1.13.0 pytorch-cuda=11.7 python=3.8

See https://anaconda.org/k2-fsa/k2/files