Deliberately not posting much, since i dont want to clickbait and head to the next thing, currently looking into improving performance by a mile in multiple areas at a time if u want to keep up with what i do checkout Remora-Lab im gonna update the repo with a llama.cpp fork too
@0xSero Been working on local ai daily for months, trying to squeeze more out of my 9060 xt. mostly kernels, quants, long context/kv in ram and expert offloading. Im in sweden too, a 3090 would give me a 24gb cuda baseline to test it all against and become another research card.
@ItsmeAjayKV personally i like DS4.1 flash better in this prompt u made but i mean its still really good, love seeing more competitors in the space ngl