I Quantized a Custom Evolutionary AI to 4-Bits From Scratch. It Got Smarter. (opens in new tab)
Most quantization papers focus on the same old routine: train a huge transformer using backpropagation, compress it to INT4, and pray the…
Read the original article