Albert Zeyer

> or download just that one overlay with `wget https://github.com/raspberrypi/rpi-firmware/raw/master/overlays/pciex1-compat-pi5.dtbo` and then `sudo cp pciex1-compat-pi5.dtbo /boot/firmware/overlays` Thanks! That's what I did now, and this seems to work!

I don't exactly understand your concerns with `DistributeFilesDataset`. We would do it the same as we do it already right now with `DistributeFilesDataset`: The sorting logic is handled by the...

Note, for PyTorch, we have exactly that now, called `default_float_dtype`. It's probably very simple to adapt for any RF usage (also including TF).

> I'm using this PR to run the CI. It's fine that you put intermediate work into a draft PR. But you don't need to use the CI to test...

Hey, what's the state here? Did you test this? Does it work? I.e. it doesn't store the intermediate activations in memory?

> Hey, no new results yet. But what was the state? What were the old results? Did it work? Does it not store the intermediate activations in memory anymore? Or...

Also, I don't exactly understand how this API here could be used for variational noise and/or weight dropout. Can you give an example?

I find the model structure sth which you only really want to see on debugging. I rarely look at it. Or maybe only once when I wrote a new model,...

Acceleration by definition means that you don't have the same amount on each wheel event but that you get a different amount. That's what acceleration means: It increases the velocity...

Can you try with `./main.py -vv --exp 0 --multiplier 2` and also record a video? I'm also interested in the log output you get then.