Quantization: Smaller Weights, Measured Error“We quantized the model to 4-bit and it got dumb.” I’ve heard this many times…