Configurable vit parameters
#68
by
gmastrapas
- opened
Should this be like the proposed change?
I'm trying to do model = AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-large', trust_remote_code=True, vision_config={'some_vision_option': some_value})
gmastrapas
changed pull request title from
Update configuration_florence2.py
to Configurable vit parameters
And getting AssertionError: only DaViT is supported for now cause apparently model_type == '' in PretrainedConfig
@gmastrapas this is a more proper fix but since the authors abandoned this project it wont get updated: https://huggingface.co/microsoft/Florence-2-large/discussions/62