Ensure .backwards works with new thermo code #199

Activity

1. mattwthompson

   added this to the MVP milestone on Oct 15, 2026
2. mattwthompson

   self-assigned this

   on Oct 15, 2026
3. mattwthompson

   added a parent issue on Oct 15, 2026

   • Phase 1 MVP #171
4. mattwthompson commented on Oct 20, 2026

   

   mattwthompson

   on Oct 20, 2026

   MemberAuthor

   More actions

   Something is not quite working here, but I haven't dug deeply into it

   ipdb> y_ref, y_ref_std, y_pred, y_pred_std = predict(
       dataset=thermo_data,
       force_field=current_ff,
       epoch_dir=root / f"EPOCH_{epoch}",
   )
   ipdb> loss = (y_pred - y_ref).square().mean()
   ipdb> loss.backward()
   *** RuntimeError: Function RemoteEnsembleAverageBackward returned an invalid gradient at index 2 - got [208] but expected shape compatible with [287]

   This is dropping in around here
5. mattwthompson

   closed this as completedin #207on Oct 27, 2026