PyTorch 1.8.1 comes with PyTorch Profiler – SD Times

The newest model of the machine studying library PyTorch is now out there. PyTorch 1.8.1 introduces the PyTorch Profiler, which is a software for efficiency evaluation and troubleshooting for large-scale deep studying fashions. 

According to PyTorch’s staff, there was a scarcity of obtainable instruments for such a process, and the instruments that did exist missed out on vital PyTorch-specific data. Machine studying scientists ended up needing to make use of a mixture of instruments or manually including correlation data as a way to meet their wants. 

PyTorch Profiler collects each GPU {hardware} and PyTorch data, correlates them, performs detection of mannequin bottlenecks, and generates suggestions on easy methods to ease these bottlenecks. The data is visualized in TensorBoard so the consumer can simply see it.

In addition, the brand new Profiler API is natively supported in PyTorch, which suggests customers don’t want to put in extra packages as a way to make use of the software. 

Other new enhancements in PyTorch 1.8.1 embody enabling autocast for PyTorch xla, making the torch. submodule import extra autocomplete-friendly, and extra. 

The full launch notes for PyTorch 1.8.1 can be found here


Please enter your comment!
Please enter your name here