orbital-materials/orb-models#158 -- they squeeze single atom system energy breaking required tensor shapes. -- fixed here by #548, pending upstream.
~~ orbital-materials/orb-models#159 -- direct models with 20 nbrs show batch discrepancies in the predicted energy. ~~ This issue was closed upstream. This level of stochastic effect is just part and parcel of their 20 nbr models as they break ties differently in limiting to 20 in a batch dependent manner. Leave as Xskip for the interface test.
#533 chose to mark these as expected failures and this issue tracks their resolution upstream. Once resolved the minimum pin should be bumped. Using the conservative orb model for batched simulations is fine and will not surface these issues. The model interface is 99% correct and the single atom is a minor edge case.
The energy issues with the batched energy are a function of the model and can be reproduced outside the torchsim ecosystem. This issue not connected to torchsim.
orbital-materials/orb-models#158 -- they squeeze single atom system energy breaking required tensor shapes. -- fixed here by #548, pending upstream.
~~ orbital-materials/orb-models#159 -- direct models with 20 nbrs show batch discrepancies in the predicted energy. ~~ This issue was closed upstream. This level of stochastic effect is just part and parcel of their 20 nbr models as they break ties differently in limiting to 20 in a batch dependent manner. Leave as Xskip for the interface test.
#533 chose to mark these as expected failures and this issue tracks their resolution upstream. Once resolved the minimum pin should be bumped. Using the conservative orb model for batched simulations is fine and will not surface these issues. The model interface is 99% correct and the single atom is a minor edge case.
The energy issues with the batched energy are a function of the model and can be reproduced outside the torchsim ecosystem. This issue not connected to torchsim.