GNER, a Generative Named Entity Recognition framework, which demonstrates enhanced zero-shot capabilities across unseen entity domains. Basically this is a GGUF version of alternatively trained text encoder that you can use as a part of CLIP for Flux, Krea, Chroma or HiDream and many other models that can use FLAN-T5-XXL.
The original model lives here: https://huggingface.co/dyyyyyyyy/GNER-T5-xxl - this Q_8 is a direct quantization of their repo.
Examples of GNER vs FLAN:
PROMPT: Professional raw photo of femme fatale wearing alluring red dress and pointing a pistol at viewer, closeup shot, dark alley, midnight time.
Made with GNER:
Made with FLAN:
...As you can see, it's a matter of artistic taste which one is better!
Description
this is not a TE-only version, hence it is almost 2x times bigger, for TE-only model (only in safetensors file format) go to:
1) https://huggingface.co/silveroxides/GNER-T5-xxl-encoder-only/tree/main
2) https://huggingface.co/wikeeyang/GNER-T5-xxl-encoder-only/tree/main
FAQ
Comments (5)
I don't quite understand. Is this just a text encoder?
or
generation diffusion model + text encoder
is it a compact model?
I'll just load this with load model gguf in comfyui, but don't I need to install another text encoder in addition? I don't quite understand. Do you have an example workflow?
(Newbie question)
this is a TEXT Encoder = in case of Chroma it is a full CLIP, but for Flux, Krea and HiDream - it is a part of CLIP
example workflows are in recommended resources of this model, you need to select this model in CLIP loader instead of FLAN
@OliviaRossi Thank you
How to use this in Forge?
put the gguf file where it belongs (not sure if forge can use clip made as gguf - if not, use safetensors versions from links in release notes of this model)