First, we use the optimizer we all know and love. Next, we use category cross-entry and categorical accuracy for our los

Pop! OS uses systemd-boot instead of grub so it took me a little while to figure out how to make the switch but after a few rounds of DuckDuckGo, I was able to lock in a pretty easy way to swap out the kernel.,At home, I turned on the news in America. I listened to an interview with a crying nurse. She was crying because the number of cases in America had risen exponentially, and, on her drive home after spending her day watching people die from the virus, she passed bar after bar packed with maskless people. She was talking about how helpless she felt, desperate for people to understand the severity and horror she was experiencing every day.,Around 3 years ago, triggered by getting a Lenovo X1C6 to replace my Macbook Air, I started tinkering with performance tweaks a bit more than before and one of the things I noticed is that if I use a low latency kernel, at least for my specific type of usage, the overall experience is a bit snappier. Since then, I have used the low latency kernel with Ubuntu on pretty much most of my daily drivers and the experience has been consistently better at least for what I do.,One of the main perks of buying a Linux laptop is that it comes optimized and there’s very little to tweak beyond your own personal visual preferences and such. So I have been just running on the stock kernel and had completely forgotten about the low latency option.,Note: If training BERT layers too, try Adam optimizer with weight decay — which can help reduce overfitting and improve generalization [1]. I would recommend this article for understanding why.,Alternatively, (although I found this to be detrimental) we can even use BERTs pre-pooled output tensors by swapping out last_hidden_state with pooler_output — but that is for another time.,If you are following me here already, you will probably have noticed that I recently switched to a System 76 Lemur Pro as my daily driver and since running an OS made by the hardware vendor has obvious merits, I also went with Pop! OS when I made the switch. So-far, my experience has been great with very minimal gripes which I will save for another post.,A few days ago, as I started reading up a bit on low latency kernels for audio production (what most people need the low latency kernel for) during my nightly wind down, I got a bit curious about the performance difference if I were to run the low latency kernel with Pop on my Lemur Pro. Then, today, I got a message from a fellow Lemur Pro owner asking about it so it kicked me over the edge and I decided to give it a go.,And… voila! The system booted up with the low latency kernel. However, when it first booted up, after I logged in, input (mouse and keyboard) was frozen up for a little bit. After a couple of minutes, it was back to normal and then things not only went back to normal but the system as a whole felt a bit snappier. Though I am immediately noticing that I am loosing a nominal amount of battery life (less than 5% of battery in a 4 hour time frame) which is to be expected.,These are pretty great results for such a simple output network. Further fine-tuning, the addition of CNNs, LSTMs, or other more expressive networks may improve our results even further.,Alternatively, (although I found this to be detrimental) we can even use BERTs pre-pooled output tensors by swapping out last_hidden_state with pooler_output — but that is for another time.,So for the next few days, I will be running on this kernel a bit more and will report back if there’s anything noteworthy. My only concerns now are one, I am not familiar with systemd-boot so I am unsure if the above method has any repercussions and two, I am not sure how safe it is to run the low latency kernel with the rest of System76’s customized pieces against this hardware. I guess off to more reading and perhaps maybe even ask System76 support.,If you are stuck on CPU, try out Google Colab — it’s a free, cloud-based notebook service provided by Google. Colab includes a GPU as standard — albeit not a particularly powerful one (but it is free).,Disclaimer: I’m not responsible for any damages or injury, including but not limited to special or consequential damages, that result from your use of the below instructions.,So for the next few days, I will be running on this kernel a bit more and will report back if there’s anything noteworthy. My only concerns now are one, I am not familiar with systemd-boot so I am unsure if the above method has any repercussions and two, I am not sure how safe it is to run the low latency kernel with the rest of System76’s customized pieces against this hardware. I guess off to more reading and perhaps maybe even ask System76 support.