anyway, the agent finished what I asked it to do when I left for work It took about an hour, which is honestly pretty good considering I asked it to put together a good set of parameters, referencing online and documentation and stuff, for a qwen3.5-4b model to be used as the dedicated compression model
Samu ／人◕ ‿‿ ◕人＼
>>1184559 other way around GPU have more raw computing power but only fit GPU shaped problems CPUs can do anything but only do one or two steps at a time
S C
ah
Samu ／人◕ ‿‿ ◕人＼
also for llms its all about having enough memory bandwidth to feed in weights constantly cpu much more bottlenecked
