运行jittor.test.test_example报错

[Mountchicken]

你好，请问一下我在运行python -m jittor.test.test_example的时候出现了这样的问题，似乎和nvcc有关

(jittor_gpu) wjp@wjp-System-Product-Name:~$ python -m jittor.test.test_example [i 0805 17:20:53.835541 52 compiler.py:869] Jittor(1.2.3.87) src: /home/wjp/miniconda2/envs/jittor_gpu/lib/python3.7/site-packages/jittor [i 0805 17:20:53.837007 52 compiler.py:870] g++ at /usr/local/bin/g++(5.4.0) [i 0805 17:20:53.837062 52 compiler.py:871] cache_path: /home/wjp/.cache/jittor/default/g++ /bin/sh: 1: /usr/loca/cuda/bin/nvcc: not found Traceback (most recent call last): File "/home/wjp/miniconda2/envs/jittor_gpu/lib/python3.7/runpy.py", line 183, in _run_module_as_main mod_name, mod_spec, code = _get_module_details(mod_name, _Error) File "/home/wjp/miniconda2/envs/jittor_gpu/lib/python3.7/runpy.py", line 109, in _get_module_details import(pkg_name) File "/home/wjp/miniconda2/envs/jittor_gpu/lib/python3.7/site-packages/jittor/init.py", line 18, in from . import compiler File "/home/wjp/miniconda2/envs/jittor_gpu/lib/python3.7/site-packages/jittor/compiler.py", line 892, in nvcc_path = env_or_try_find('nvcc_path', 'nvcc') or
File "/home/wjp/miniconda2/envs/jittor_gpu/lib/python3.7/site-packages/jittor/compiler.py", line 816, in env_or_try_find version = jit_utils.get_version(path) File "/home/wjp/miniconda2/envs/jittor_gpu/lib/python3.7/site-packages/jittor_utils/init.py", line 260, in get_version version = run_cmd(output+" --version") File "/home/wjp/miniconda2/envs/jittor_gpu/lib/python3.7/site-packages/jittor_utils/init.py", line 143, in run_cmd raise Exception(err_msg) Exception: Run cmd failed: /usr/loca/cuda/bin/nvcc --version

(jittor_gpu) wjp@wjp-System-Product-Name:~$ nvcc -V nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2019 NVIDIA Corporation Built on Fri_Feb__8_19:08:17_PST_2019 Cuda compilation tools, release 10.1, V10.1.105

[Jittor]

您的nvcc在什么位置呢，which nvcc输出怎样

[Jittor]

建议您让jittor帮您找到合适版本的nvcc和cuda，您可以运行python3 -m jittor_utils.install_cuda

[Mountchicken]

nvcc在 /usr/loacl/cuda-10.1/bin中，我也执行了 export nvcc_path="/usr/loacl/cuda-10.1/bin"的指令

[Mountchicken]

pytorch是能正常运行的

[Mountchicken]

然后我也测试jittor的其它代码，是可以正常运行的，包括在cpu上训练网络。但是只要指定jt.flags.use_cuda=1就会报错

[Jittor]

错误信息里面显示Exception: Run cmd failed: /usr/loca/cuda/bin/nvcc --version， 您的环境变量是不是local误写成loca了？