3 ms·
llama.cpp runs on the CPU anyways, the whole conversation is on a false premise. The only hard thing about running it on consumer hardware is the quantization s
by zamadatix 4y ago
llama.cpp runs on the CPU anyways, the whole conversation is on a false premise. The only hard thing about running it on consumer hardware is the quantization step (where the largest RAM requirements come in) and that's not even a technical problem since you could just redistribute the quantized version if it weren't for the license of the model.