zqiu

2 ms·

1. 1.▲
   by zqiu 4mo ago

   Thank you so much, glad you liked it :)
2. 2.▲
   by zqiu 4mo ago

   Thank you so much :)
3. 3.▲
   by zqiu 4mo ago

   Hi, thank you so much :) Yes, you are very sharp, to make truely it scalable, we have to finetune quantized LLMs directly, similar ideas as QLoRA. The algorithmic innovation is based on our research about orthogonal finetuning OFT and QOFT,