Support my work on llama.cpp related to memory optimizations (also for the convert scripts), fast ternary packing, Mamba, Mamba-2, Jamba, and other recurrent models.

I've also been exploring with rounding algorithms to improve k-quants.

I study in Electrical Engineering, but I like to work on ML in my free time. I like to work on problems where both low-level and high-level concepts need to be considered.

Large external storage is very useful when testing the conversion script(s).

Sometimes, when working on (de)quantization schemes, I need to test things on hardware with specific instruction set extensions and/or GPUs, in which case I temporarily rent a cloud instance, but I generally prefer to use local hardware.