BRbrucethemoose2·Sep 24, 2023
> brute aggressive quantization
Quantization like ExLlama's EX2 or even llama.cpp's k-quants is already far from brute force: https://github.com/turboderp/exllamav2#exl2-quantization
> The format allows for mixing quantization levels within a model to achieve any average bitrate between 2 and 8 bits per weight. Moreover, it's possible to apply multiple quantization levels to each linear layer, producing something akin to sparse quantization wherein more important weights (columns) are quantized with more bits. The same remapping trick that lets ExLlama work efficiently with act-order models allows this mixing of formats to happen with little to no impact on performance. Parameter selection is done automatically by quantizing each matrix multiple times, measuring the quantization error (with respect to the chosen calibration data) for eac
Read the rest
h of a number of possible settings, per layer. Finally, a combination is chosen that minimizes the maximum quantization error over the entire model while meeting a target average bitrate.
Llama.cpp is also working on a feature that let's a small model "guess" the output of a big model which then "checks" it for correctness. This is more of a performance feature, but you could also arrange it to accelerate a big model on a small GPU.