In Chapter 1, we defined the starting point as the minimum information of the universe. In Chapter 2, we examined the process by which differences are generated from that minimum information and ...
Replay (rehearsal) dataset generation for mitigating catastrophic forgetting during SFT. Instead of mixing public SFT datasets (which are distributionally mismatched), this pipeline reconstructs the ...
This page has been translated to Chinese by monklof. I started writing down this list as my personal notes of topics I discussed with colleagues and friends, and that I wanted to deepen... I'm not a ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results