[BUG] Can't pickle local object 'instrument_w_nvtx.<locals>.wrapped_fn'

[annopackage]

Describe the bug A clear and concise description of what the bug is. bug occurs while calling dataloader with multi num workers. Here, ’trainer‘ is initialized from transformers. If I only debug with dataloader as follows, the code works.
‘’‘ for _, data in tqdm(enumerate(trainer.get_train_dataloader())): print('dataloader: ', _, data.keys()) ’‘’

However, if running trainer.train(), then the code throws the error as the title.

To Reproduce Steps to reproduce the behavior:

1. Go to '...'
2. Click on '....'
3. Scroll down to '....'
4. See error

Expected behavior A clear and concise description of what you expected to happen.

ds_report output Please run ds_report to give us details about your setup.

Screenshots

System info (please complete the following information):

• OS: [e.g. Ubuntu 18.04]
• GPU count and types [e.g. two machines with x8 A100s each]
• Interconnects (if applicable) [e.g., two machines connected with 100 Gbps IB]
• Python version
• Any other relevant info about your setup

Launcher context launching your experiment with the deepspeed launcher, MPI, or something else?

Docker context Do not use docker.

Additional context Add any other context about the problem here.

[loadams]

Hi @annopackage - can you share a full minimal repro script with us please?

[annopackage]

hi, thanks for your quick reply. My dataloader contains function for video decoding with gpu, which maybe the reason of the issue. If I comment this part, the dataset with transformer and deepspeed works well. The minimal script for reproduction would be provided later.

[annopackage]

@loadams Hi, the reason to the issue is multiprocessing_context. I think the error could be reproduced while setting multiprocessing_context in dataloader as 'spawn'.

[annopackage]

With zero2 and spawn, the code runs well. With zero3 and fork, the code runs well. With zero3 and spawn, the bug occurs.