Move all .cpu().numpy() to after backward call.

InfluenceFunctional / MXtalTools

BSD 3-Clause "New" or "Revised" License

8 stars 1 forks source link

Closed InfluenceFunctional closed 9 months ago

InfluenceFunctional commented 9 months ago

See autoencoder step. Appears to be some speedup here, possibly from cuda's async

InfluenceFunctional commented 9 months ago

Coming in autoencoder merge