Evaluation of GGUF?

[MoonRide303]

GGUF (llama.cpp) is a very popular format used for handling quantized models - would be nice to see evaluation of that, and also quants like Q6_K (0.16% PPL difference vs fp16).

Evaluations for Llama 2 70B are available on llama.cpp project: https://github.com/ggerganov/llama.cpp/blob/master/examples/perplexity/README.md