James Reed
James Reed
@dzhulgakov wrote that tutorial
I don't really see how benchmarking channels-last as an explicit API precludes us from doing automatic optimizations in our compiler. On the contrary, it can help expose gaps and create...
So I think my thinking comes from the fact that the API-facing layout (NCHW) was actually an arbitrary choice iiuc in that that's what CuDNN did at the time. However,...
> @jamesr66a can you add any specifics around which models you'd like to quantize and what the minimal number of quantized models is that would be useful to enable? As...
@wconstab since it's changing the semantics of the model it usually requires Human intervention (though there are some things that attempt to do auto-tuning) I think it would make sense...
To clarify, I think there's actually three execution modes being discussed here: a) Training with QAT quant-dequant nodes inserted at various program points b) The calibration step of post-training quantization,...
@vkuzo I agree with that, iiuc you're referring to (c) in my list?
This should be fixed now by https://github.com/pytorch/pytorch/pull/33002
@mdcramer the fix was landed after the 1.4.0 branch was cut. It will be present in either the nightly build or the 1.5.0 release, when that happens
Hey @cbalioglu any update? :)