News for hackers. Stuff that matters.
Re:The Q, K, V Matrices by lostmsu on Thursday January 08, 2026 @11:01AM (#46542537)
> and in all three cases I observed quality degradation (when training from scratch).

At the same model size and training FLOPS?

Re:The Q, K, V Matrices by p1esk on Thursday January 08, 2026 @11:46AM (#46543194)
No. Each projection is ~5% of total FLOPs/params. Not enough model capacity change to care. From what I remember, removing one of them was worse than other two, I think it was Q. But in all three cases, degradation (in both loss and perplexity) was significant.