[Bug] Default process group has not been initialized with autoslim search about mmrazor HOT 8 CLOSED

Thank you for your issue.
At present, distributed mode is needed when searching even if only one gpu is used. It is hacky and we are refactoring the search part. The new version will no longer have this problem.

from mmrazor.

这个问题现在的版本解决了吗？我也遇到一样的问题

from mmrazor.

You can avoid this by trying distributed mode.

Plus, using English is more appreciated for better community discussion around the world.

from mmrazor.

where to do the setup you said

from mmrazor.

where to do the setup you said

You can set the job launcher to one of pytorch, slurm or mpi (ref to here ) to use distributed mode.

from mmrazor.

$ python ./tools/mmcls/search_mmcls.py \
>   configs/pruning/autoslim/autoslim_mbv2_search_8xb1024_ci10.py \
>   output/epoch_50.pth \
>   --work-dir output \
>   --launcher pytorch
/home/tanghuayang/venv_torch/lib/python3.6/site-packages/mmrazor/utils/setup_env.py:33: UserWarning: Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
  f'Setting OMP_NUM_THREADS environment variable for each process '
/home/tanghuayang/venv_torch/lib/python3.6/site-packages/mmrazor/utils/setup_env.py:43: UserWarning: Setting MKL_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
  f'Setting MKL_NUM_THREADS environment variable for each process '
Traceback (most recent call last):
  File "./tools/mmcls/search_mmcls.py", line 181, in <module>
    main()
  File "./tools/mmcls/search_mmcls.py", line 99, in main
    init_dist(args.launcher, **cfg.dist_params)
  File "/home/tanghuayang/venv_torch/lib64/python3.6/site-packages/mmcv/runner/dist_utils.py", line 18, in init_dist
    _init_dist_pytorch(backend, **kwargs)
  File "/home/tanghuayang/venv_torch/lib64/python3.6/site-packages/mmcv/runner/dist_utils.py", line 29, in _init_dist_pytorch
    rank = int(os.environ['RANK'])
  File "/usr/lib64/python3.6/os.py", line 669, in __getitem__
    raise KeyError(key) from None
KeyError: 'RANK'

Is it necessary to configure cfg.dist_params? And, how to configure it?

from mmrazor.

$ python ./tools/mmcls/search_mmcls.py \
>   configs/pruning/autoslim/autoslim_mbv2_search_8xb1024_ci10.py \
>   output/epoch_50.pth \
>   --work-dir output \
>   --launcher pytorch
/home/tanghuayang/venv_torch/lib/python3.6/site-packages/mmrazor/utils/setup_env.py:33: UserWarning: Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
  f'Setting OMP_NUM_THREADS environment variable for each process '
/home/tanghuayang/venv_torch/lib/python3.6/site-packages/mmrazor/utils/setup_env.py:43: UserWarning: Setting MKL_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
  f'Setting MKL_NUM_THREADS environment variable for each process '
Traceback (most recent call last):
  File "./tools/mmcls/search_mmcls.py", line 181, in <module>
    main()
  File "./tools/mmcls/search_mmcls.py", line 99, in main
    init_dist(args.launcher, **cfg.dist_params)
  File "/home/tanghuayang/venv_torch/lib64/python3.6/site-packages/mmcv/runner/dist_utils.py", line 18, in init_dist
    _init_dist_pytorch(backend, **kwargs)
  File "/home/tanghuayang/venv_torch/lib64/python3.6/site-packages/mmcv/runner/dist_utils.py", line 29, in _init_dist_pytorch
    rank = int(os.environ['RANK'])
  File "/usr/lib64/python3.6/os.py", line 669, in __getitem__
    raise KeyError(key) from None
KeyError: 'RANK'

Is it necessary to configure cfg.dist_params? And, how to configure it?

it's runing, use the following command:

$ RANK=0 WORLD_SIZE=1 MASTER_ADDR=127.0.0.1 MASTER_PORT=1692 python ./tools/mmcls/search_mmcls.py \
  configs/pruning/autoslim/autoslim_mbv2_search_8xb1024_ci10.py \
  output/epoch_50.pth \
  --work-dir output \
  --launcher pytorch

from mmrazor.

$ python ./tools/mmcls/search_mmcls.py \
>   configs/pruning/autoslim/autoslim_mbv2_search_8xb1024_ci10.py \
>   output/epoch_50.pth \
>   --work-dir output \
>   --launcher pytorch
/home/tanghuayang/venv_torch/lib/python3.6/site-packages/mmrazor/utils/setup_env.py:33: UserWarning: Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
  f'Setting OMP_NUM_THREADS environment variable for each process '
/home/tanghuayang/venv_torch/lib/python3.6/site-packages/mmrazor/utils/setup_env.py:43: UserWarning: Setting MKL_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
  f'Setting MKL_NUM_THREADS environment variable for each process '
Traceback (most recent call last):
  File "./tools/mmcls/search_mmcls.py", line 181, in <module>
    main()
  File "./tools/mmcls/search_mmcls.py", line 99, in main
    init_dist(args.launcher, **cfg.dist_params)
  File "/home/tanghuayang/venv_torch/lib64/python3.6/site-packages/mmcv/runner/dist_utils.py", line 18, in init_dist
    _init_dist_pytorch(backend, **kwargs)
  File "/home/tanghuayang/venv_torch/lib64/python3.6/site-packages/mmcv/runner/dist_utils.py", line 29, in _init_dist_pytorch
    rank = int(os.environ['RANK'])
  File "/usr/lib64/python3.6/os.py", line 669, in __getitem__
    raise KeyError(key) from None
KeyError: 'RANK'

Is it necessary to configure cfg.dist_params? And, how to configure it?

it's runing, use the following command:

$ RANK=0 WORLD_SIZE=1 MASTER_ADDR=127.0.0.1 MASTER_PORT=1692 python ./tools/mmcls/search_mmcls.py \
  configs/pruning/autoslim/autoslim_mbv2_search_8xb1024_ci10.py \
  output/epoch_50.pth \
  --work-dir output \
  --launcher pytorch

but, how to write these configuration parameters into cfg.dist_params?

from mmrazor.