Question about training time on NuScenes

[dabi4256]

    您好，我自己用完整的nuScenes数据集训练，显卡是两张3090， BATCH_SIZE_PER_GPU=8，用的是‘cbgs_voxel0075_voxelnext.yaml’配置文件。训练时间稳定的话大概是200个小时，这是正常的吗。

[yukang2017]

你好，感觉不是很正常。我用4卡V100，BATCH_SIZE_PER_GPU=4. 训练大概2天多点。我猜测可能训练速度瓶颈在dataloader的数据读取上。

[dabi4256]

你好，感谢你的回复。
请问这个时间是实际跑完需要2天，还是程序给出的预估时间。 我代码还没有训练完，只是看到程序给出的时间是200个小时。
而且，我用双卡3090跟单卡3090的预估时间差别不大。

[yukang2017]

你好，我预估时间和实际时间基本差不多。

[learnuser1]

    您好，我自己用完整的nuScenes数据集训练，显卡是两张3090， BATCH_SIZE_PER_GPU=8，用的是‘cbgs_voxel0075_voxelnext.yaml’配置文件。训练时间稳定的话大概是200个小时，这是正常的吗。

你好，这个自己训练的话效果跟原论文比的话怎么样。然后这个模型跟centerpoint比的话要好很多嘛

[dabi4256]

    您好，我自己用完整的nuScenes数据集训练，显卡是两张3090， BATCH_SIZE_PER_GPU=8，用的是‘cbgs_voxel0075_voxelnext.yaml’配置文件。训练时间稳定的话大概是200个小时，这是正常的吗。

你好，这个自己训练的话效果跟原论文比的话怎么样。然后这个模型跟centerpoint比的话要好很多嘛

我的机器跑的很慢，20个epoch还没跑完。但是我测试了第10个epoch，mAP已经到了59.32了。CenterPoint我没跑过，但是论文里面写的是CenterPoint的mAP为 58.6.

[learnuser1]

    您好，我自己用完整的nuScenes数据集训练，显卡是两张3090， BATCH_SIZE_PER_GPU=8，用的是‘cbgs_voxel0075_voxelnext.yaml’配置文件。训练时间稳定的话大概是200个小时，这是正常的吗。

你好，这个自己训练的话效果跟原论文比的话怎么样。然后这个模型跟centerpoint比的话要好很多嘛

我的机器跑的很慢，20个epoch还没跑完。但是我测试了第10个epoch，mAP已经到了59.32了。CenterPoint我没跑过，但是论文里面写的是CenterPoint的mAP为 58.6.

好的，感谢

[rockywind]

@learnuser1 @yukang2017 @dabi4256 @yanwei-li 大家好
我跑多卡训练，会报这个错误。

Traceback (most recent call last):
  File "train.py", line 246, in <module>
Traceback (most recent call last):
  File "train.py", line 246, in <module>
    main()
  File "train.py", line 179, in main
    main()
  File "train.py", line 179, in main
    model = nn.parallel.DistributedDataParallel(model, device_ids=[cfg.LOCAL_RANK % torch.cuda.device_count()])
  File "/opt/conda/envs/VoxelNet/lib/python3.8/site-packages/torch/nn/parallel/distributed.py", line 674, in __init__
    model = nn.parallel.DistributedDataParallel(model, device_ids=[cfg.LOCAL_RANK % torch.cuda.device_count()])
  File "/opt/conda/envs/VoxelNet/lib/python3.8/site-packages/torch/nn/parallel/distributed.py", line 674, in __init__
    _verify_param_shape_across_processes(self.process_group, parameters)
  File "/opt/conda/envs/VoxelNet/lib/python3.8/site-packages/torch/distributed/utils.py", line 118, in _verify_param_shape_across_processes
    _verify_param_shape_across_processes(self.process_group, parameters)
  File "/opt/conda/envs/VoxelNet/lib/python3.8/site-packages/torch/distributed/utils.py", line 118, in _verify_param_shape_across_processes
    return dist._verify_params_across_processes(process_group, tensors, logger)
torch.distributed.DistBackendError:     return dist._verify_params_across_processes(process_group, tensors, logger)NCCL error in: ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:1275, internal error, NCCL version 2.14.3
ncclInternalError: Internal check failed.
Last error:
Duplicate GPU detected : rank 1 and rank 0 both on CUDA device 8a000

torch.distributed.DistBackendError: NCCL error in: ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:1275, internal error, NCCL version 2.14.3
ncclInternalError: Internal check failed.
Last error:
Duplicate GPU detected : rank 0 and rank 1 both on CUDA device 8a000
ERROR:torch.distributed.elastic.multiprocessing.api:failed (exitcode: 1) local_rank: 0 (pid: 1132223) of binary: /opt/conda/envs/VoxelNet/bin/python
Traceback (most recent call last):
  File "/opt/conda/envs/VoxelNet/lib/python3.8/runpy.py", line 194, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/opt/conda/envs/VoxelNet/lib/python3.8/runpy.py", line 87, in _run_code
    exec(code, run_globals)
  File "/opt/conda/envs/VoxelNet/lib/python3.8/site-packages/torch/distributed/launch.py", line 196, in <module>
    main()
  File "/opt/conda/envs/VoxelNet/lib/python3.8/site-packages/torch/distributed/launch.py", line 192, in main
    launch(args)
  File "/opt/conda/envs/VoxelNet/lib/python3.8/site-packages/torch/distributed/launch.py", line 177, in launch
    run(args)
  File "/opt/conda/envs/VoxelNet/lib/python3.8/site-packages/torch/distributed/run.py", line 785, in run
    elastic_launch(
  File "/opt/conda/envs/VoxelNet/lib/python3.8/site-packages/torch/distributed/launcher/api.py", line 134, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/opt/conda/envs/VoxelNet/lib/python3.8/site-packages/torch/distributed/launcher/api.py", line 250, in launch_agent
    raise ChildFailedError(
torch.distributed.elastic.multiprocessing.errors.ChildFailedError: