From pre-compiled wheels (recommended)

We provide pre-compiled wheels for the following platforms:

• Pre-compiled CPU wheels (Linux)
• Pre-compiled CUDA wheels (Linux)
• Pre-compiled CPU wheels (macOS)
• Pre-compiled CPU wheels (Windows)

We recommend that you use this approach to install k2.

Hint

Please always install the latest version of k2.

Installation examples

We provide several examples below to show you how to install k2

Linux (CPU) example

Suppose that we want to install the following wheel

https://huggingface.co/csukuangfj/k2/resolve/main/cpu/k2-1.24.3.dev20230719+cpu.torch2.0.1-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl

we can use one of the following methods:

# method 1
pip install torch==2.0.1+cpu -f https://download.pytorch.org/whl/torch_stable.html
pip install k2==1.24.3.dev20230719+cpu.torch2.0.1 -f https://k2-fsa.github.io/k2/cpu.html

# method 2
pip install torch==2.0.1+cpu -f https://download.pytorch.org/whl/torch_stable.html
wget https://huggingface.co/csukuangfj/k2/resolve/main/cpu/k2-1.24.3.dev20230719+cpu.torch2.0.1-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
pip install ./k2-1.24.3.dev20230719+cpu.torch2.0.1-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl

Hint

You can visit https://k2-fsa.github.io/k2/cpu.html to see avaiable versions of k2.

macOS (CPU) example

Suppose we want to use the following wheel:

https://huggingface.co/csukuangfj/k2/resolve/main/macos/k2-1.24.3.dev20230720+cpu.torch2.0.1-cp38-cp38-macosx_10_9_x86_64.whl

we can use the following methods:

# method 1
pip install torch==2.0.1
pip install k2==1.24.3.dev20230720+cpu.torch2.0.1 -f https://k2-fsa.github.io/k2/cpu.html

# method 2
pip install torch==2.0.1
wget https://huggingface.co/csukuangfj/k2/resolve/main/macos/k2-1.24.3.dev20230720+cpu.torch2.0.1-cp38-cp38-macosx_10_9_x86_64.whl
pip install ./k2-1.24.3.dev20230720+cpu.torch2.0.1-cp38-cp38-macosx_10_9_x86_64.whl

Hint

You can visit https://k2-fsa.github.io/k2/cpu.html to see avaiable versions of k2.

Windows (CPU) example

Suppose we want to install the following wheel

https://huggingface.co/csukuangfj/k2/resolve/main/windows-cpu/k2-1.24.3.dev20230726+cpu.torch2.0.1-cp38-cp38-win_amd64.whl

we can use the one of the following methods:

# method 1
pip install torch==2.0.1+cpu -f https://download.pytorch.org/whl/torch_stable.html
pip install k2==1.24.3.dev20230726+cpu.torch2.0.1 -f https://k2-fsa.github.io/k2/cpu.html

# method 2
pip install torch==2.0.1+cpu -f https://download.pytorch.org/whl/torch_stable.html
wget https://huggingface.co/csukuangfj/k2/resolve/main/windows-cpu/k2-1.24.3.dev20230726+cpu.torch2.0.1-cp38-cp38-win_amd64.whl
pip install k2-1.24.3.dev20230726+cpu.torch2.0.1-cp38-cp38-win_amd64.whl

Hint

If you want to build k2 with CUDA support on Windows, please consider compiling k2 from source.

Hint

You can visit https://k2-fsa.github.io/k2/cpu.html to see avaiable versions of k2.

Linux (CUDA) example

Suppose we want to install

https://huggingface.co/csukuangfj/k2/resolve/main/cuda/k2-1.24.3.dev20230718+cuda11.7.torch2.0.1-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl

we can use the following methods:

# method 1
pip install torch==2.0.1+cu117 -f https://download.pytorch.org/whl/torch_stable.html
pip install k2==1.24.3.dev20230718+cuda11.7.torch2.0.1 -f https://k2-fsa.github.io/k2/cuda.html

# method 2
pip install torch==2.0.1+cu117 -f https://download.pytorch.org/whl/torch_stable.html
wget https://huggingface.co/csukuangfj/k2/resolve/main/cuda/k2-1.24.3.dev20230718+cuda11.7.torch2.0.1-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
pip install ./k2-1.24.3.dev20230718+cuda11.7.torch2.0.1-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl

Hint

You can visit https://k2-fsa.github.io/k2/cuda.html to see avaiable versions of k2.