Lower precision in trainnet workflow

Hi,

Is there a way to adjust the precision when training a network using the "trainnet" workflow (2024a)? I'm interested in lowering precision to speed up training, in the same way I would use "torch.set_float32_matmul_precision" in pytorch.

Thanks,

Eric

Answer 1

[Matt J]

Just pass in training data with a lower precision type (e.g., singles versus doubles).