[LOW] Improvements to model checkpointing

g-simmons / 289G_NLP_project_FQ2020

0 stars 1 forks source link

Open g-simmons opened 3 years ago

g-simmons commented 3 years ago

wandb data upload sizes are approaching a gig, I assume this is because I added checkpointing after each epoch (full model gets saved to wandb)
Can we reduce the upload size?
Maybe only save the weights, or only save the most recent model?