Is it possible to use Torchserve for Huggingface transformers` VisionEncoderDecoderModel?

[kangsan0420]

Hi, I'm wondering whether I can use Torchserve for the below model.
(I'd like to use it for TrOCR)

from transformers import VisionEncoderDecoderModel
encoder_decoder = ('google/vit-base-patch16-224-in21k', 'kykim/bert-kor-base')
model = VisionEncoderDecoderModel.from_encoder_decoder_pretrained(*encoder_decoder)

[msaroufim]

@agunapal has recently looked at video models, it should be entirely possible - just need to expand on the huggingface custom handler to support this

This recent PR should give you a good idea on how to do this #1904