Open michaelbenayoun opened 1 year ago
yes, for context: https://github.com/huggingface/optimum/pull/570#discussion_r1045939305
My understanding is that this choice is made because in the _TASKS_TO_AUTOMODELS
mapping, we want to easily map stable-diffusion
to StableDiffusionPipeline
If we were to use text-to-image
as a task in _TASKS_TO_AUTOMODELS
, there would be an issue: several classes can handle text-to-image
, such as StableDiffusionPipeline
from diffusers, but also eBart
from transformers (dall-e).
@mht-sharma similar issue as with timm/transformers, as discussed in the PR
cc @IlyasMoutawwakil
Feature request
As title says.
cc @echarlaix @fxmarty
Motivation
Make things more consistent and less misleading.