Slim-Llama employs binary/ternary quantization which reduces the precision of model weights to just 1 or 2 bits, ...