Is there a way to cache a specific dataset and reuse its encoded image latents and text embeddings

[primecai]

I am working on training a model, but since I have not finalized my dataset, I am progressively adding small chunks of images and captions once every few days. However, so far, I have always deleted all the cached image latents and text embeddings and recreated new ones when I started to train a new model. My dataset could be as follows:

--dataset
----data_0
------image_0.png
------image_0.txt
------image_1.png
------image_1.txt
------...
----data_1
------image_0.png
------image_0.txt
------image_1.png
------image_1.txt
------...
----data_2
------...
----data_x

The dataset is getting huge and it takes >2 days to do the caching. So I am trying to find a way to reuse the encoded latent and embeddings from data_0 to data_2 when I add new data, data_x, to the dataset. Is this somehow doable by the available features in SimpleTuner? Thanks in advance!