Is there a model architecture beyond Transformer to generate good text with small a dataset, a few GPUs and "few" parameters? It is enough generating coherent English text as short answers.
Is it true that deepseek for training has distilled from OpenAl models or is it just a move by Microsoft not to lose too much in the stock market?
What is the difference between open-source and free software?
Is there a benchmark to understand how good is a small LLM in generating text between 100M and 400M? Because most of benchmarks out there are for trillions of parameters.
I have finally solved my problem creating from scratch in rust a driver to write files on a disk, to save data permanently. Just wanted to say: never give up. I thought it was impossible, but I've done it. Same thing might occur with your project.
Can i still have a good chatbot from an LLM i've trained from scratch with only one GPU but with 168M parameters? My goal is to have a chatbot able to speak generic English conversations, nothing super-stargate (just hobby)
In the rust os Moros, how is managed the persistancr storage on a device disk.img? I tried to see the code but i couldn't understand how it allow to save strings on a disk.img.
Ĉu estas aŭtoroj kiuj verkis poemojn aŭ librojn en Esperanto aŭ eĉ filmojn?
At the moment i am developing a Rust OS and it is only one kernel cargo no_std. I'd like now to add userspace implementation in my kernel (i followed the phil opp tutorial) to create other cargo and maybe using the std. Where can i learn how to implement such a thing in my OS?
Are there some resources that I can consult to continue to develop my OS if until now I have followed the tutorial https://os.phil-opp.com/ ?
Are there some resources that I can consult to continue to develop my OS if until now I have followed the tutorial https://os.phil-opp.com/?
How do I create a file system for my monolithic kernel in Rust?
I saw on Gitlab the source code of RedoxFS and i saw that it depends of the standard library, but how is it possible if by definition an OS must be deployed in no_std mode? Is maybe possible to create a file system using the Rust std and implementing it with a no_std kernel or what?
In your opinion, to train an LLM of 900M parameters based on GPT architecture how many A100 40GB GPUs do I need (if one is enough or i need more on Colab)? And how long would training on a 20GB dataset on Google Colab with "pay as you go" plan (I have a tokenizer that has 35000 tokens) last?
I am creating an operating system (a real project not just basic) and I'd like to receive feedbacks for new features you'd like to have in an OS (or problems you have with Windows, MacOS or Linux)
I tried QubesOS but it is simply not usable in real life: due to virtualization and it is really too heavy. Does anyone use QubesOS in real life as main OS or people use it "just for fun"?
I read that less than 3% of people use Linux for PCs, but Windows users are continuing to decline year after year. So all these people are also looking for an alternative to Linux for some reason?
