This release brings support for FP32, TF32, FP16 and BF16 precision modes, allowing faster trainings and inference on hardware that supports it.
If you had previously added a remote server, refresh it to make those modes visible. They will appear next to the inference or training device selector.
Full Changelog
- Add support for FP32, TF32, FP16 and BF16 precision modes
- Improve SSH connectivity and sync
- Fix python clients staying alive on local and remote servers when tasks were complete
- fix module parameters not properly restored when loading a project
- fix pin on top state when saving and loading a project
- minor improvements to RandomGenerator dataset
Subscribe to TorchStudio channels for live updates:
- Twitter: https://twitter.com/TorchStudioAI
- LinkedIn: https://linkedin.com/company/TorchStudioAI
- Facebook: https://www.facebook.com/TorchStudioAI
- Youtube: https://www.youtube.com/@TorchStudioAI