fix build for new RTX GPUs

kengz / SLM-Lab

Modular Deep Reinforcement Learning framework in PyTorch. Companion library of the book "Foundations of Deep Reinforcement Learning".

https://slm-lab.gitbook.io/slm-lab/

MIT License

1.23k stars 263 forks source link

Closed kengz closed 2 years ago

kengz commented 2 years ago

fix breakage with newer GPUs that require newer NVIDIA CUDA version and newer pytorch version
fix device assignment error in total_loss by assigning tensor to the same device as self.net.device. This fixes:
- fix #490
- fix #494

Thanks to @karl-grantham for help with debugging!