Ubuntu+GPU上Debug Torch项目fairseq

1,下载并启动zbstudio

$ git clone https://github.com/soumith/zbs-torch.git  
$ cd zbs-torch  
$ ./zbstudio.sh  

2,安装Torch
http://torch.ch/docs/getting-started.html

3,安装nccl

git clone https://github.com/NVIDIA/nccl
cd nccl
make install
sudo ldconfig
luarocks install nccl

找到libnccl.so*的目录在/etc/bashrc里加上:

export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda-8.0/lib64:/usr/local/lib"

3,安装fairseq
https://github.com/facebookresearch/fairseq#requirements-and-installation

4,把fairseq导入zbstudio
在zbstudio里选择project->project directory->choose

5,preprocess数据
https://github.com/facebookresearch/fairseq#data-pre-processing

6,开始Debug
在train.lua里修改datadir参数指向fairseq/data-bin/iwslt14.tokenized.de-en
运行train.lua,也选中train.lua就是点那个绿色三角

你可能感兴趣的:(Torch)