opendiffusionai/t5-v1_1-base-encoder-only

This model exists only as a convenience, to be able to use google t5 base conveniently for comparison and testing sake. That being said... it is significantly inferior to the other models.

If you raally need a model with 768dim output.. you are most likely better off using a larger one and a projection layer, rather than using this one at native 768 size, unless you just need a tiny model.

(projection layers are surprisingly easy to use. only about 10 more lines of code. poke around at our other t5 based sd models to find examples of this.)