2 questions, i havent tried installing it yet.

[sprappcom]

1. do i need 24gb vram to run it? it doesnt seem to have -ngl number of layers to offload to cpu parameter like llama.cpp
2. can i do pure cpu inference without using gpu? how is the speed up like compared with llama.cpp pure cpu inference?

[sprappcom]

finished reading this
https://github.com/kvcache-ai/ktransformers/blob/main/doc/en/deepseek-v2-injection.md

realised need vram.

what's the ratio of vram to dram that is needed and how to effectively calculate how much ram to use for each?