Questions about the Flan Held-Out Tasks (pubmed_qa) and experiment details in the flan collection paper

[nanyyyyyy]

Are the Flan Held-Out Tasks mentioned in the flan collection paper (PubmedQA, WANLI, CondaQA, MedNLI, CxC) included for training in the checkpoints released on Hugging Face? Because I do see multiple PubmedQA tasks in the FLAN2022 mixture, but I cannot find the other four tasks in the final mixture. So I am a little bit confused here. Are the checkpoints released on Hugging Face trained on the complete flan 2022 dataset? Why pubmed qa is a held out task. Thank you so much for your help!!