Proposal is to augment the docs with benchmark info for embeddings generation, simultaneously documenting performance on systems such as apple Silicon (mps). Could be similar to https://omnicloudmask.readthedocs.io/en/latest/performance.html
Noting MPS use requires some additional config since rslearn passes trainer options directly to Lightning and sets a default DDP strategy, but Lightning DDP is not supported on MPS.
trainer:
accelerator: mps
devices: 1
strategy: auto
Proposal is to augment the docs with benchmark info for embeddings generation, simultaneously documenting performance on systems such as apple Silicon (mps). Could be similar to https://omnicloudmask.readthedocs.io/en/latest/performance.html
Noting MPS use requires some additional config since rslearn passes trainer options directly to Lightning and sets a default DDP strategy, but Lightning DDP is not supported on MPS.