NVIDIA's apex only for inferences

[damiox]

Hi. I’m looking for some feedback about the question I posted here:

Is training a requirement to use apex O1 mode?

Hi. I am trying to use apex only for inferences with an existing model that was trained with FP32 (with no mixed precision, and no apex support). But I’m wondering if what I’m doing is wrong. I can’t find in the documentation if apex can be used only for inferences without re-training the model. I cannot re-train my model. I posted this question here: https://github.com/NVIDIA/apex/issues/750 but I’m confused about the feedback I got. Also It’d be really good if I can find something that is su…

I know NVIDIA’s apex library brings a lot of features to help training, but it’s also specified the inferences will run faster with this library.
My question is: Can I use apex for inferences only without re-training the FP32 model with Mixed Precision? Is there a real improvement by doing so in comparison with just casting my model to half precision?

Thanks in advance,
Damian