In this episode I talk about reduced precision floating point formats float16 (aka half precision) and bfloat16. I'll discuss what floating point numbers are, how these two formats vary, and some of the practical considerations that arise when you are working with numeric code in PyTorch that also needs to work in reduced precision. Did you know that we do all CUDA computations in float32, even if the source tensors are stored as float16? Now you know!Further reading.<ul><li>The Wikipedia article on IEEE floating point is pretty great <a href="https://en.wikipedia.org/wiki/IEEE_754">https://en.wikipedia.org/wiki/IEEE_754</a></li><li>How bfloat16 works out when doing training <a href="https://arxiv.org/abs/1905.12322">https://arxiv.org/abs/1905.12322</a></li><li>Definition of acc_type in PyTorch <a href="https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/AccumulateType.h">https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/AccumulateType.h</a></li></ul>

Tensor subclasses and Liskov substitution principle

The PyTorch Developer Podcast is a place for the PyTorch dev team to do bite sized (10-20 min) topics about all sorts of internal development topics in PyTorch.