Thank you for your excellent work and open-source contribution!
Regarding the experimental setup in Section 4.1, I would like to inquire whether the number 126K might be a typo.
As far as I know, 126K is the total number of multilingual instructions from the original RxR dataset. However, the total number of instructions in RxR-CE should not be 126K, right? Instead, it should equal the total number of unique episode_id entries in the RxR-CE dataset, since each episode corresponds to one instruction.
Could you please clarify this point? Thank you very much for your time and help!