[covoeus]

llama.cpp is general purpose in the sense that it supports many different model architectures. ds4 is laser focused on deepseek v4 flash, thus having a leaner codebase