-
Notifications
You must be signed in to change notification settings - Fork 233
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
What is the full batch size if mesh_dim is set to 1,1,-1, on TPUv3-8? #94
Comments
Different mesh dims correspond to different sharding strategies. While they do not define a batch size, they do incur certain constraints on the possible batch size.
|
Did you get this to run with a v3? I seem to always get HLO out of memory errors. |
A single v3-8 only has 128GB of memory in total, which might not be sufficient for training a 7B model. |
Hi, thanks for this amazing repo.
I was wondering how should I set batch size to make a desirable full batch size.
For example, if I set train_dataset.huggingface_dataset.batch_size to 1 on TPUv3-8,
what is the full batch size given mesh_dim 1,1,-1 / 1,-1,1 / -1,1,1 ?
Are all of them 8? or 1?
Thanks!
The text was updated successfully, but these errors were encountered: