Install pytorch commit fd25a2a

I was running evaluation on cityscape and did not find args.test_scale anywhere. Its not mentioned in options.py aswell.

Error:

if len(args.test_scale) == 1:
AttributeError: 'Namespace' object has no attribute 'test_scale'

Running this from terminal: CUDA_VISIBLE_DEVICES=0 python3.6 test.py --dataset cityscapes --model danet --resume-dir cityscapes/model --crop-size 768 --workers 1 --backbone resnet101 --multi-grid --multi-dilation 4 8 16 --eval

[huanghoujing]

There are some environment variables in setup.py to set for customizing your CUDA, cudnn, NCCL paths. Thanks to God, it is in Python logic. The example in my case:

1. Install CUDA 8.0 and cudnn 7
2. Install Anaconda 3
3. Install Pytorch from source

# Install certain version of pytorch from source

INSTALL_PYTORCH_SCRIPT_DIR=${HOME}/Software/install_pytorch

pip uninstall -y torch

# Install basic dependencies
conda install --yes numpy pyyaml mkl mkl-include setuptools cmake cffi typing
conda install --yes -c mingfeima mkldnn
# Add LAPACK support for the GPU
conda install --yes -c pytorch magma-cuda80 # or magma-cuda90 if CUDA 9

cd ${HOME}/Software
git clone --recursive https://github.com/pytorch/pytorch
cd pytorch
git checkout fd25a2a
git submodule init # There may be submodules no longer existing in new commits
git submodule update --recursive

rm -rf build
rm -rf torch.egg-info
export CUDA_HOME=/mnt/data-1/data/houjing.huang/Software/cuda-8.0
export USE_SYSTEM_NCCL=1
export NCCL_LIB_DIR=${CUDA_HOME}/lib64
export NCCL_INCLUDE_DIR=${CUDA_HOME}/include
export CMAKE_PREFIX_PATH="$(dirname $(which conda))/../" # [anaconda root directory]

python setup.py install 2>&1 | tee ${INSTALL_PYTORCH_SCRIPT_DIR}/install_pytorch.log
cd ${INSTALL_PYTORCH_SCRIPT_DIR}
python test_data_parallel.py 2>&1 | tee test_pytorch_data_parallel.log

# test_data_parallel.py
import torch
import torch.nn as nn
from torch.nn.parallel import DataParallel

model = nn.Linear(10, 20).cuda()
x = torch.ones(100, 10).float().cuda()
model_w = DataParallel(model, device_ids=[0,1,2,3])
x = model_w(x)
# x = model(x)
print(x.size())

[emma-sjwang]

@wmj1238 Actually, after I successfully installed pytorch@fd25a2a, I did not find that I need this ‘nervanagpu’ third_party. Meybe you forget to change to the correct pytorch version by:

git checkout fd25a2a