🚀 Feature

Add support for torch.max with:

• CUDA bfloat16
• CPU float16 and bfloat16

Motivation

Currently, torch.max has support for CUDA float16:

>>> torch.rand(10, dtype=torch.float16, device='cuda').max()
tensor(0.8530, device='cuda:0', dtype=torch.float16)

But all three other combinations of CPU/CUDA and float16/bfloat16 are not supported:

>>> torch.ra

Description

Implement a YOLO model and add it to the DJL model zoo

References

• https://arxiv.org/abs/1506.02640

Issue to track tutorial requests:

• Deep Learning with PyTorch: A 60 Minute Blitz - #69

It would be helpful to have visualisation tools to plot/debug information in the SNNs. I also think we should be slightly careful to use correct/adequate abstractions since it is 1) not the primary purpose of Norse, and 2) we don't want to maintain something that will change a lot in the future.

Here are a few suggestions for visualisations:

• Layer parameters
  • Weights / tuning curves

• asarray : converts all array-likes to a numpy array; no copy for arrays
• astensor converts all array-likes to a mygrad tensor; returns tensors unchanged
• assert_almost_equal with various options to include checks of gradients and to include checks ac