Plans to make the processor compatible with transformers Trainer?

#58
by RonanMcGovern - opened

As far as I understand, the blocker on using the hf trainer is having a processor that handles the images. Right now, the processor is just a text tokeniser.

Is an integration with hf Trainer on the roadmap?

If there were a simple processor available, that should allow for a data collator to be put together to cover the different use cases (gaze, caption etc.)?

I can try to help out, if given some direction. Thanks.

Sign up or log in to comment