google-deepmind / google-deepmind/weathernext
How many L40S (48 GB VRAM) need to train GenCast or GraphCast at 0.25 deg?
Open
- Dominant language
- Python
- Stars
- 7.7k
- Forks
- 986
- PR merge metrics
- No merged PRs in 30d
Description
As a newbie, I'm really struggled to make JAX GenCast (or GraphCast) tensor parallelism 😢
Can I train PyTorch GenCast (or GraphCast) at 0.25 deg with 4 ~ 8 L40S?
BTW, each L40S have 48GB VRAM
Contributor guide
Assessment
This issue has not been assessed yet.