Coverage decreased (-1.3%) to 92.28% when pulling 9ff88319e25b2acd1668b08449455ae0c986726e on autodiff into 87762dce11d52b1d91a300d1ac289c46602e2514 on develop.
Coverage decreased (-1.3%) to 92.28% when pulling 9ff88319e25b2acd1668b08449455ae0c986726e on autodiff into 87762dce11d52b1d91a300d1ac289c46602e2514 on develop.
Coverage increased (+0.3%) to 93.932% when pulling 96bc4f50fd0a942eb1e9b55579eefc37bace4978 on autodiff into 87762dce11d52b1d91a300d1ac289c46602e2514 on develop.
Seems like not the best API if comparing multiple optimizers. Maybe do w = t3f.prepare_var(w); define tensor yourself; t3f.grad(tensor, w)?