NNPOps 0.2
This release brings a full support of TorchANI via NNPOps.OptimizeTorchANI
. Additionally, the PyTorch wrappers for the continuous-filter convolution (CFConv) operations have been implemented, which will be used to accelerate TorchMD-Net.
What's Changed
- Optimization to SchNet by @peastman in #42
- Fix the execution of ANISymmetryFunctions on a non-default CUDA stream by @raimis in #37
- Move the TorchANI imports from the top level by @raimis in #45
- Optimize EnergyShifter of TorchANI by @raimis in #38
- Optimize SpeciesConverter of TorchANI by @raimis in #39
- Optimize TorchANI by @raimis in #41
- Enable CPU-only builds by @raimis in #48
- Improve the wrapper of
ANISymmetryFunctions
by @raimis in #47 - PyTorch wrapper for SchNet operations by @raimis in #40
- Refactor the CUDA error check by @raimis in #49
Full Changelog: v0.1...v0.2