Jittor[1]的推出,让机器学习多了一个框架可以供人选择。迫不及待地,我在本人的笔记本和服务器上进行了安装尝试。
实验1,台式机的操作系统是windows10,遇到了:
1.pip3 install git+https://github.com/Jittor/jittor.git:错误, 没网络
2.pip3 install git+https://git.net9.org/cjld/jittor.git:错误, 没网络
3.git clone Jittor/jittor && python jittorsetup.py install :遇到了UnicodeDecodeError: 'gbk' codec can't decode byte 0xab in position 177: illegal multibyte sequence
4.修改encoding='utf8',python jittorsetup.py install, :遇到了没有g++,UnicodeDecodeError: 'utf-8' codec can't decode byte 0xb2 in position 8: invalid start byte
5.放弃安装。
实验2,服务器操作系统是centos7+cuda10(nvidia 1080ti),遇到了:
1.需要安装Python-3.7.7,依赖openssl-1.0.2r,修改Python-3.7.7/Modules/Setup安装;
2.需要安装g++5.4.0,安装;
3.git安装好jittor1.0后,执行python3 -m jittor.test.test_example: 遇到错误,
Run cmd: /usr/local/bin/g++ /usr/local/lib/python3.7/site-packages/jittor/src/utils/http://cache_compile.cc /usr/local/lib/python3.7/site-packages/jittor/src/utils/http://log.cc /usr/local/lib/python3.7/site-packages/jittor/src/utils/http://tracer.cc /usr/local/lib/python3.7/site-packages/jittor/src/utils/http://jit_utils.cc -Wall -Werror -Wno-unknown-pragmas -std=c++14 -fPIC -march=native -I/usr/local/include/python3.7m -I/root/.local/include/python3.7m -I/usr/local/lib/python3.7/site-packages/jittor/src -O2 -lstdc++ -ldl -shared -o /root/.cache/jittor/default/g++/jit_utils_core.cpython-37m-x86_64-linux-gnu.so
4.放弃安装。
实验3,在实验2的基础上,服务器环境是centos7-无cuda,遇到了如下问题:
1.需要安装Python-3.7.7,依赖openssl-1.0.2r,修改Python-3.7.7/Modules/Setup安装;
2.需要安装g++5.4.0,依赖大内存,需要配置swap交换文件;
3.git安装好jittor1.0后,执行python3 -m jittor.test.test_example:错误同实验2;
4.放弃安装。
结论:计图Jittor是一个有点丢脸的“贵系”产品,仓促推出有点令人失望。
[1]开源机器学习框架——计图(Jittor) https://cg.cs.tsinghua.edu.cn/jittor/
[2]计图安装步骤 https://cg.cs.tsinghua.edu.cn/jittor/download/