You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
refactor(pt_expt): re-export loss classes from dpmodel instead of wrapping
Loss classes hold only plain Python floats/bools (no tensors, no
NativeOP sub-components), so @torch_module wrapping is unnecessary.
Re-export dpmodel classes directly. Remove .to(device) calls in tests
and relax float32 tolerance in spin loss test.
0 commit comments