Install using conda (Linux/macOS/Windows)
All you need is the following line
$ conda install -c k2-fsa -c pytorch -c conda-forge k2 python=3.8 cudatoolkit=11.1 pytorch=1.8.1
to install the latest k2 with Python 3.8, CUDA 11.1, and PyTorch 1.8.1.
You can change the versions of Python, CUDA, and PyTorch as you need.
To Install a CPU version, use:
conda install -c k2-fsa -c pytorch cpuonly k2 python=3.8 pytorch=1.8.1
Hint
If you are using macOS and encounter the following error while
running python3 -m k2.version after installation:
ImportError: dlopen(/Users/fangjun/software/miniconda3/envs/foo/lib/python3.8/site-packages/_k2.cpython-38-darwin.so, 2): Library not loaded: @rpath/libk2context.dylib
Referenced from: /Users/fangjun/software/miniconda3/envs/foo/lib/python3.8/site-packages/_k2.cpython-38-darwin.so
Reason: image not found
You can use:
export DYLD_LIBRARY_PATH=$CONDA_PREFIX/lib/python3.8/site-packages:$DYLD_LIBRARY_PATH python3 -m k2.version # now it should work
Read the following if you want to learn more.
Supported versions
The conda package k2 is available from the channel k2-fsa, so
you have to use -c k2-fsa or --channel k2-fsa for conda install.
It is recommended to specify the versions of Python, CUDA, and PyTorch while installing k2. Supported versions are given in the above badges.
Search for available versions
conda search -c k2-fsa k2
lists the available versions. An example output is given below:
$ conda search -c k2-fsa k2
Loading channels: done
# Name Version Build Channel
k2 0.3.3.dev20210509_cuda10.1 cuda10.1_py3.6_torch1.6.0 k2-fsa
k2 0.3.3.dev20210509_cuda10.1 cuda10.1_py3.6_torch1.7.0 k2-fsa
k2 0.3.3.dev20210509_cuda10.1 cuda10.1_py3.6_torch1.7.1 k2-fsa
(lots of lines are omitted)
k2 0.3.3.dev20210510_cuda11.0 cuda11.0_py3.6_torch1.7.1 k2-fsa
k2 0.3.3.dev20210510_cuda11.1 cuda11.1_py3.6_torch1.8.0 k2-fsa
k2 0.3.3.dev20210510_cuda11.1 cuda11.1_py3.6_torch1.8.1 k2-fsa
(lots of lines are omitted)
Detailed installation instructions
To install k2, let’s first create a virtual environment:
$ conda create -n foo python=3.7
Collecting package metadata (current_repodata.json): done
Solving environment: done
## Package Plan ##
environment location: /xxx/anaconda3/2020.11/envs/foo
added / updated specs:
- python=3.7
The following NEW packages will be INSTALLED:
_libgcc_mutex pkgs/main/linux-64::_libgcc_mutex-0.1-main
ca-certificates pkgs/main/linux-64::ca-certificates-2021.4.13-h06a4308_1
certifi pkgs/main/linux-64::certifi-2020.12.5-py37h06a4308_0
ld_impl_linux-64 pkgs/main/linux-64::ld_impl_linux-64-2.33.1-h53a641e_7
libffi pkgs/main/linux-64::libffi-3.3-he6710b0_2
libgcc-ng pkgs/main/linux-64::libgcc-ng-9.1.0-hdf63c60_0
libstdcxx-ng pkgs/main/linux-64::libstdcxx-ng-9.1.0-hdf63c60_0
ncurses pkgs/main/linux-64::ncurses-6.2-he6710b0_1
openssl pkgs/main/linux-64::openssl-1.1.1k-h27cfd23_0
pip pkgs/main/linux-64::pip-21.0.1-py37h06a4308_0
python pkgs/main/linux-64::python-3.7.10-hdb3f193_0
readline pkgs/main/linux-64::readline-8.1-h27cfd23_0
setuptools pkgs/main/linux-64::setuptools-52.0.0-py37h06a4308_0
sqlite pkgs/main/linux-64::sqlite-3.35.4-hdfb4753_0
tk pkgs/main/linux-64::tk-8.6.10-hbc83047_0
wheel pkgs/main/noarch::wheel-0.36.2-pyhd3eb1b0_0
xz pkgs/main/linux-64::xz-5.2.5-h7b6447c_0
zlib pkgs/main/linux-64::zlib-1.2.11-h7b6447c_3
Proceed ([y]/n)? y
Preparing transaction: done
Verifying transaction: done
Executing transaction: done
#
# To activate this environment, use
#
# $ conda activate foo
#
# To deactivate an active environment, use
#
# $ conda deactivate
Second, activate the environment foo that we just created:
$ conda activate foo
(foo) $
Third, determine the versions of Python, CUDA, and PyTorch you want to use.
Assuming you want to use Python 3.8, CUDA 11.1, and PyTorch 1.8.1. Then you can use the following command to install k2:
(foo) $ conda install -c k2-fsa -c pytorch -c conda-forge k2 python=3.8 cudatoolkit=11.1 pytorch=1.8.1
Caution
You have to specify -c k2-fsa -c pytorch -c conda-forge. Otherwise, the installation will fail.
-c k2-fsatells conda where to findk2.
-c pytorchtells conda where to findpytorch=1.8.1
-c conda-forgetells conda where to findcudatoolkit=11.1
The output of the above command is:
(foo) $ conda install -c k2-fsa -c pytorch -c conda-forge k2 python=3.8 cudatoolkit=11.1 pytorch=1.8.1
Collecting package metadata (current_repodata.json): done
Solving environment: done
## Package Plan ##
environment location: /xxx/anaconda3/2020.11/envs/foo
added / updated specs:
- cudatoolkit=11.1
- k2
- python=3.8
- pytorch=1.8.1
The following packages will be downloaded:
package | build
---------------------------|-----------------
cudatoolkit-11.1.1 | h6406543_8 1.20 GB conda-forge
k2-0.3.3.dev20210509_cuda11.1|cuda11.1_py3.8_torch1.8.1 61.4 MB k2-fsa
pytorch-1.8.1 |py3.8_cuda11.1_cudnn8.0.5_0 1.27 GB pytorch
------------------------------------------------------------
Total: 2.54 GB
The following NEW packages will be INSTALLED:
_openmp_mutex conda-forge/linux-64::_openmp_mutex-4.5-1_llvm
blas pkgs/main/linux-64::blas-1.0-mkl
cudatoolkit conda-forge/linux-64::cudatoolkit-11.1.1-h6406543_8
k2 k2-fsa/linux-64::k2-0.3.3.dev20210509_cuda11.1-cuda11.1_py3.8_torch1.8.1
libuv conda-forge/linux-64::libuv-1.41.0-h7f98852_0
llvm-openmp conda-forge/linux-64::llvm-openmp-11.1.0-h4bd325d_1
mkl conda-forge/linux-64::mkl-2020.4-h726a3e6_304
mkl-service conda-forge/linux-64::mkl-service-2.3.0-py38h1e0a361_2
mkl_fft conda-forge/linux-64::mkl_fft-1.3.0-py38h5c078b8_1
mkl_random conda-forge/linux-64::mkl_random-1.2.0-py38hc5bc63f_1
ninja conda-forge/linux-64::ninja-1.10.2-h4bd325d_0
numpy pkgs/main/linux-64::numpy-1.19.2-py38h54aff64_0
numpy-base pkgs/main/linux-64::numpy-base-1.19.2-py38hfa32c7d_0
python_abi conda-forge/linux-64::python_abi-3.8-1_cp38
pytorch pytorch/linux-64::pytorch-1.8.1-py3.8_cuda11.1_cudnn8.0.5_0
six conda-forge/noarch::six-1.16.0-pyh6c4a22f_0
typing_extensions conda-forge/noarch::typing_extensions-3.7.4.3-py_0
The following packages will be UPDATED:
certifi pkgs/main::certifi-2020.12.5-py37h06a~ --> conda-forge::certifi-2020.12.5-py38h578d9bd_1
libgcc-ng pkgs/main::libgcc-ng-9.1.0-hdf63c60_0 --> conda-forge::libgcc-ng-9.3.0-h2828fa1_19
libstdcxx-ng pkgs/main::libstdcxx-ng-9.1.0-hdf63c6~ --> conda-forge::libstdcxx-ng-9.3.0-h6de172a_19
pip pkgs/main/linux-64::pip-21.0.1-py37h0~ --> conda-forge/noarch::pip-21.1.1-pyhd8ed1ab_0
python 3.7.10-hdb3f193_0 --> 3.8.8-hdb3f193_5
The following packages will be SUPERSEDED by a higher-priority channel:
_libgcc_mutex pkgs/main::_libgcc_mutex-0.1-main --> conda-forge::_libgcc_mutex-0.1-conda_forge
ca-certificates pkgs/main::ca-certificates-2021.4.13-~ --> conda-forge::ca-certificates-2020.12.5-ha878542_0
openssl pkgs/main::openssl-1.1.1k-h27cfd23_0 --> conda-forge::openssl-1.1.1k-h7f98852_0
setuptools pkgs/main::setuptools-52.0.0-py37h06a~ --> conda-forge::setuptools-49.6.0-py38h578d9bd_3
Proceed ([y]/n)? y
Downloading and Extracting Packages
k2-0.3.3.dev20210509 | 61.4 MB | ############################################################################################ | 100%
pytorch-1.8.1 | 1.27 GB | ############################################################################################ | 100%
cudatoolkit-11.1.1 | 1.20 GB | ############################################################################################ | 100%
Preparing transaction: done
Verifying transaction: done
Executing transaction: - By downloading and using the CUDA Toolkit conda packages, you accept the terms and conditions of the CUDA End
User License Agreement (EULA): https://docs.nvidia.com/cuda/eula/index.html
done
To verify that k2 is installed successfully, use:
(foo) $ python3 -m k2.version
It should print something like the following:
(foo) $ python3 -m k2.version
/xxx/anaconda3/2020.11/envs/foo/lib/python3.8/runpy.py:127: RuntimeWarning: 'k2.version' found in sys.modules after import of package 'k2', but prior to execution of 'k2.version'; this may result in unpredictable behaviour
warn(RuntimeWarning(msg))
Collecting environment information...
k2 version: 0.3.3
Build type: Release
Git SHA1: 397b5b154d93b82eb58bc2eee11ca835b5aa138c
Git date: Sun May 9 06:38:52 2021
Cuda used to build k2: 11.1
cuDNN used to build k2: 8.0.5
Python version used to build k2: 3.8
OS used to build k2: Ubuntu 16.04.7 LTS
CMake version: 3.18.4
GCC version: 5.5.0
CMAKE_CUDA_FLAGS: --expt-extended-lambda -gencode arch=compute_35,code=sm_35 --expt-extended-lambda -gencode arch=compute_50,code=sm_50 --expt-extended-lambda -gencode arch=compute_60,code=sm_60 --expt-extended-lambda -gencode arch=compute_61,code=sm_61 --expt-extended-lambda -gencode arch=compute_70,code=sm_70 --expt-extended-lambda -gencode arch=compute_75,code=sm_75 -D_GLIBCXX_USE_CXX11_ABI=0 --compiler-options -Wall --compiler-options -Wno-unknown-pragmas
CMAKE_CXX_FLAGS: -D_GLIBCXX_USE_CXX11_ABI=0
PyTorch version used to build k2: 1.8.1
PyTorch is using Cuda: 11.1
NVTX enabled: True
Disable debug: True
Sync kernels : False
Disable checks: False
The following code verifies that you can create an example FSA with k2:
(foo) $ python3
Python 3.8.8 (default, Apr 13 2021, 19:58:26)
[GCC 7.3.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import k2
>>> s = '''
... 0 1 -1 0.1
... 1
... '''
>>> fsa = k2.Fsa.from_str(s)
>>> fsa.draw('hello.svg', title='hello')
<graphviz.dot.Digraph object at 0x7ff88b1ca610>
Congratulations! You have installed k2 successfully.