GGUF quantization meta-data format

[mobicham]

Hello!

Are there some resources that explain how the quantized parameters are structured in a GGUF file? We are interested in porting HQQ-quantized models into GGUF format, but in order to do that, we need to know exactly how it is stored. We basically need to know:

• The bitpacking logic
• axis along which quantization is done
• group-sizes associated with different quant types

Thanks!

[phymbert]

Hi, you would better have a look at llama.cpp :

https://github.com/ggerganov/llama.cpp/blob/f184dd920852d6d372b754f871ee06cfe6f977ad/llama.cpp#L13599

[crimson-knight]

@mobicham here is the spec for GGUF for you to use: https://github.com/ggerganov/ggml/blob/master/docs/gguf.md