3 ms·
:) Oh ye from the paper it looks like if one uses alpha = 2*rank, sometimes LoRA does even better than full finetuning
by danielhanchen 2y ago
:) Oh ye from the paper it looks like if one uses alpha = 2*rank, sometimes LoRA does even better than full finetuning