페이스북의 파이토치(PyTorch) 버전이 빠르게 올라가고 있습니다. 지난번 포스팅 이후 두번의 릴리즈가 더 있었습니다. 0.1.7 버전에는 텐서플로우에는 없는 LBFGS 알고리즘이 추가 되었고 0.1.8 버전에는 분산 처리를 위한 THD 라이브러리가 추가 되었습니다. 콘다를 사용하면 파이토치 바이너리를 편하게 설치할 수 있습니다.

### conda (linux, osx)
$ conda install pytorch torchvision -c soumith

### pip
# linux: python2.7 + cuda7.5
$ pip install https://s3.amazonaws.com/pytorch/whl/cu75/torch-0.1.8.post1-cp27-none-linux_x86_64.whl

# linux: python2.7 + cuda8.0
$ pip install https://s3.amazonaws.com/pytorch/whl/cu80/torch-0.1.8.post1-cp27-none-linux_x86_64.whl

# linux: python3.5 + cuda7.5
$ pip install https://s3.amazonaws.com/pytorch/whl/cu75/torch-0.1.8.post1-cp35-cp35m-linux_x86_64.whl

# linux: python3.5 + cuda8.0
$ pip install https://s3.amazonaws.com/pytorch/whl/cu80/torch-0.1.8.post1-cp35-cp35m-linux_x86_64.whl

# linux: python3.6 + cuda7.5
$ pip install https://s3.amazonaws.com/pytorch/whl/cu75/torch-0.1.8.post1-cp36-cp36m-linux_x86_64.whl

# linux: python3.6 + cuda8.0
$ pip install https://s3.amazonaws.com/pytorch/whl/cu80/torch-0.1.8.post1-cp36-cp36m-linux_x86_64.whl

# osx: python2.7
$ pip install https://s3.amazonaws.com/pytorch/whl/torch-0.1.8.post1-cp27-none-macosx_10_7_x86_64.whl

# osx: python3.5
$ pip install https://s3.amazonaws.com/pytorch/whl/torch-0.1.8.post1-cp35-cp35m-macosx_10_6_x86_64.whl

# osx: python3.6
$ pip install https://s3.amazonaws.com/pytorch/whl/torch-0.1.8.post1-cp36-cp36m-macosx_10_6_x86_64.whl

### torchvision
$ pip install torchvision