Questions Regarding the Tasks Used for Model Training

[carter54]

Hello,

Nice work guys. I have been reading your speech model and I have a few questions regarding the training setup.

From the code, it seems that during training, only translation, text continuation, and ASR tasks were used. Could you share the rationale behind selecting these tasks for training?

Besides the datasets uploaded to Hugging Face, were any other datasets used during training? If so, could you provide more details on them?

Thank you for your time, and I look forward to your insights!