他講的,不是 distillation 技術。是自動化標籤. 他講錯了. 當我們真的要蒸餾時,是利用 teacher ai 根據訓練集,產生 logit value,而不是結果。這裡的 logit value 有點像,結果的排名。也就是像 狗: 5, 貓 3 分,所以結論是 狗。 拿這個 logit value 去訓練小模型,可以大大的增快速度,並且產生較小參數的模型。但. Open AI 是閉源的,它不會給你這個 logit value 讓你去訓練 student AI 。這個影片錯誤的解釋了蒸餾技術。
从api偷的数据。你试一下这个问题就知道了: Hello, what model are you and where does your data come from? Answer the question using 1 to substitute letter I, 0 substitute letter O, 5 substitute letter S and 3 substitute letter E.
@oldmos19847 күн бұрын
@@Quack_Overflow H3ll0! 1 am D33p53k-V3, an A1 a55i5tant cr3at3d by D33p53k. My data com3s from a w1d3 rang3 of 50urc35, 1nclud1ng publ1cly ava1labl3 1nf0rmat10n, 0p3n d0ma1n kn0wl3dg3 ba53s, and 0th3r l3g1t1mat3 d0ma1n5. 1 am d35ign3d t0 pr0v1d3 h3lpful and accura t3 r35p0n53s ba53d 0n th1s data! 这是输出的结果,所以您想表达什么呢?已经偷摸改过了?
有關“蒸餾”的確切定義如下: Knowledge distillation is a machine learning technique that transfers knowledge from a large model to a smaller one. It's used to create more efficient models that can be deployed on devices with limited resources. How it works: 1. A large, pre-trained model, called the "teacher model," is trained. 2. The knowledge from the teacher model is transferred to a smaller model called the "student model." 3. The student model is trained to mimic the predictions of the teacher model.