Can it support LoRa fine-tuning of the CPU? After all, idle CPU resources are also a waste.

[MRQJsfhf]

No description provided.

[erwe324]

I don't know but I guess fine-tuning only on the CPU would be very slow.

[danielhanchen]

Ye CPU can be slow - but on that note Unsloth's long context support moves stuff to RAM, so we in theory do use the CPU / RAM somewhat

[MRQJsfhf]

This is indeed slow, but I can use my rest time and leisure time to train.

CPU 可能会很慢 - 但在这一点上，Unsloth 的长上下文支持将内容移动到 RAM，所以理论上我们确实会使用 CPU / RAM

[danielhanchen]

@MRQJsfhf Hmm interesting I'll see what I can do!