Sagemaker Studio notebook instances restricted to 64 megabytes not allow to train Pytorch multiprocess

Question

Sagemaker Studio notebook instances restricted to 64 megabytes not allow to train Pytorch multiprocess with the default dataloaders. **How can I add more capacity to /dev/shm or what kernel can I use to train with Pytorch multiprocess?**
```
uname -a
Linux tensorflow-2-3-gpu--ml-g4dn-xlarge-33edf42bcb5531c041d8b56553ba 4.14.231-173.361.amzn2.x86_64 #1 SMP Mon Apr 26 20:57:08 UTC 2021 x86_64 x86_64 x86_64 GNU/Linux
df -h | grep -E 'shm|File'
Filesystem Size Used Avail Use% Mounted on
shm 64M 0 64M 0% /dev/shm
```

Answer

This is being tracked in the GitHub issue linked below.
- https://github.com/aws/sagemaker-python-sdk/issues/937

A possible workaround is to use a regular Notebook Instance instead of a Studio Notebook Instance. On a regular Notebook Instance of the same size (ml.g4dn.xlarge), /dev/shm is 7.7G
```
df -h | grep -E 'shm|File'
Filesystem      Size  Used Avail Use% Mounted on
tmpfs           7.7G     0  7.7G   0% /dev/shm
```

Sagemaker Studio notebook instances restricted to 64 megabytes not allow to train Pytorch multiprocess

Relevanter Inhalt