Run model from hydra setup
#16
Conversation
was this a real question or a statement? |
| offline: False | ||
| project: "molbind" | ||
| entity: "adrianmirza" | ||
| entity: "wandb_username" |
There was a problem hiding this comment.
not sure if this is the ideal default. Perhaps nothing or environment variable is even better.
I mean "better" because this here can lead to silent or hard-to-follow bugs
But this is a minor point.
|
|
||
| rootutils.setup_root(__file__, indicator=".project-root", pythonpath=True) |
There was a problem hiding this comment.
did know this module!
what didn't work without this?
|
|
||
| device_count = torch.cuda.device_count() | ||
| trainer = L.Trainer( | ||
| max_epochs=100, |
There was a problem hiding this comment.
you maybe wand to also make max_epochs and accelerator customizable (for debugging, mps on MacBook might be fine, too)
| data_modalities=train_modality_data, | ||
| batch_size=config.data.batch_size, | ||
| shuffle=True, | ||
| num_workers=1, |
There was a problem hiding this comment.
if you have a more complex loader, you might be happy about having more workers
| dataset=data_modalities[modality], | ||
| modality=modality, | ||
| central_modality=central_modality, | ||
| context_length=256, |
There was a problem hiding this comment.
should this be fixed here?
Squash and merge for clean commits and progress tracking would be nice?