denghj3

Results 3 issues of denghj3

## Description the func show below, it describe all nonzero elements become 1, **while it change the the whole mat to 1 in reality**, with **using X.data[:] = 1 or...

enhancement

想请教下,llama-pro训练的显存需求是多少,和lora比要多多少

相同的思路,是否存在套取出预训练数据集的方法?