Thanks for the PR @neuralsorcerer !

TODO: My one qualm is that I don't think it's worth having both trim_weights AND trim_and_normalize_weights.
I think it makes more sense to add target_sum_weights to trim_weights (with None as default, and use other values for the various cases). Could you please make that change and update the PR?

NOTE: for visibility, I need two people to accept your PR for it to land - so it might be that it would land only between 0-2 days after I accept - since my friends/colleagues may only give a green light next week - just for you to have visibility on why things might take time (I'll do my best to review ASAP, but the actual land depends on +1 other person to be available).

TODO: My one qualm is that I don't think it's worth having both trim_weights AND trim_and_normalize_weights. I think it makes more sense to add target_sum_weights to trim_weights (with None as default, and use other values for the various cases). Could you please make that change and update the PR?

Updated.

NOTE: for visibility, I need two people to accept your PR for it to land - so it might be that it would land only between 0-2 days after I accept - since my friends/colleagues may only give a green light next week - just for you to have visibility on why things might take time (I'll do my best to review ASAP, but the actual land depends on +1 other person to be available).

No worries. Thanks for the update.

@talgalili has imported this pull request. If you are a Meta employee, you can view this in D87063176.

@talgalili merged this pull request in ca963b7.