Slater Pooling is very slow

NLESC-JCER / QMC

Deep Learning for Quantum Monte Carlo Simulations

Apache License 2.0

7 stars 2 forks source link

Slater Pooling is very slow #7

Closed NicoRenaud closed 4 years ago

NicoRenaud commented 5 years ago

SP is by far the most time consuming step of the forward pass. Can we speed that up. pytorch 1.2.0 support batched determinant calculations. We should try those instead of my own batch det

NicoRenaud commented 4 years ago

the batch det of pytorch is used by defaults (require pytroch>=1.4)