To Mask or Not to Mask: The Effect of Prompt Tokens on Instruction Tuning | Towards Data Science

Implementing prompt-loss-weight, and why we should replace prompt-masking with prompt-weighting