Question on Training Datasets for One-Stage Training in TA-TiTok

[yingShen-ys]

Thank you for the amazing work on TA-TiTok!

I’m curious about the training process—specifically, what datasets were used during the one-stage training for TA-TiTok? Is it a single dataset or a combination of several public datasets?

If it’s the latter, could you share details about the specific datasets involved?

Looking forward to your response!

[TACJu]

Hi,

Thank you for your interest in our work. Regarding the training datasets used for TA-TiTok, as detailed in Table 7 of our paper, we exclusively used DataComp-1B with a resolution filter.

[yingShen-ys]

Thank you for the prompt response!