can we possibly train (ideally continual pretraining) Gemma 3 series of models on t5x ?

[StephennFernandes]

is there a possibility to train (ideally continual pretraining) Gemma 3 series of models on t5x ?

why choose t5x over maxtext ?
because i plan to continually pretrain gemma 3 on Mixture of Denoisers and over bi-directional attention similar to UL2 and PaLM 2 models

i am also open to suggestions whether it would be ideal and less clumbersome to explore this option on maxtext