You must be logged in to sponsor compilade
Become a sponsor to compilade
compilade
Canada
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 study in Electrical Engineering, but I like to work on ML in my free time. My hardware is limited (8GB of RAM on a low-power ThinkPad), but it's remarkably sufficient for most of what I've been doing. Still, I hope to eventually get something more powerful.
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, in which case I temporarily rent a cloud instance, but I generally prefer to use local hardware.