Hi,
simple question: Can someone point me to the part of the documentation where I can see how/whether an operation autocasts to fp16 when run with AMP or whether it keeps using fp32. Or can this only be seen from the source code?
Best,
Tim
Hi,
simple question: Can someone point me to the part of the documentation where I can see how/whether an operation autocasts to fp16 when run with AMP or whether it keeps using fp32. Or can this only be seen from the source code?
Best,
Tim
Hi,
See the doc on autocast about autocast operation reference, in particular the part about Op-Specific Behavior where there is such a list. Is it what you are looking for?
Best regards,
Thomas
Thank you very much. That’s exactly it.