Zkkkai (SII)

icon company Shanghai Innovation Institute icon location SII is an institution dedicated to innovation in education and research in the field of AI.

Results 2 comments of Zkkkai (SII)

The issue with HASAC

Thanks for your reply!

SFT中 loss mask与最终loss没有对齐，会mask掉eos_token的loss

似乎应该将下面这行代码去除，这样最后的loss mask是正确的，对于response的第一个token和最后一个token（EOS）的预测都会计算相应的loss https://github.com/volcengine/verl/blob/504696245b39dd22162579b36303706ee61a731a/verl/utils/dataset/sft_dataset.py#L197