Took me a while to understand what their "hot" and "cold" neurons meant, since in most ML I do, there is no such notion. And their paper doesn't directly define it (or I missed it)

After some thoughts, in ReLU it does make sense, because half of the function is constant, so you can say that you're "cold" if that neuron's ReLU-ed output is often 0 . So I checked whether ReLU was common in LLMs, original llama doesn't use ReLU. But after (re-)reading the github, it actually only works on ReLU models. Turns out that there is a group of people "fine-tuning" (I would rather call that re-training, since you start by breaking the model?) models to use ReLU to allow for that sparsity: https://huggingface.co/SparseLLM

So this is sadly not applicable to any model you can find on the internet, but that sounds like a great progress anyway. Possibly this might shift the compromises back to bigger models but with "less ideal" activations. Also I'm curious what would be the legal impacts on it (since USA and EU refers to a model's FLOPs/number of parameters... How do you compute it with sparsity? Do you average?)

I think that a possible avenue for future research in that area is keeping original activation (like llama keeping SwiGLU), but using quantification to define "hot" and "cold" neurons to be saturation areas. (For example, saying that this activation function, below -1. at 8 bit, is equivalent to -infinity, and thus this is a cold neuron)