Hugging Face's logo

ZeroCool94
/
mm-v1-0

Model card Files Community
4
Edit model card
YAML Metadata Warning: empty or missing yaml metadata in repo card (https://huggingface.co/docs/hub/model-cards#model-card-metadata)
license: creativeml-openrail-m

This repository contains several merged models using the provided merge.py script on the repository.

Why?

The idea behind this is to test how different models specialized in different tasks or trained on different data and merged together with the exact same weight can improve the general model or not, so far multiple tests and comparisons seem to show that it does improve the general quality of the model while still still able to generate the same content the original model was good at without losing any data or losing so little that its not noticeable. If this research is successful we might be able to just train a single main model with high resolution image such as Stable Diffusion and then train smaller models and then merge it to the main model and achieve the same quality as if a big fine tune was done on the main model from the start. This would allow us to train models that fit consumer cards on specific datasets for smaller tasks similar to Textual Inversion but being actually trained from scratch, meaning they would return a .ckpt file that can be used by itself and teaches a model to do something specific instead of a .pt embed file which depends on another model to run and just reuses what the model already now to be better at something.

The term we will be using around here to refer to the merged models is "Mega Model" as that is what we are trying to achieve, a single mega model.

A few things to keep in mind when browsing this model card and repository:

  • When possible every model used when doing the merge will be described on this model card.

  • The filenames for the models will be using short versions of the models in it for brevity.

  • Always assume that all models used in the Mega Model have the exact same weight even when this is not mentioned, unless specified all models will try to follow this rule of thumb as this is what ensured that little to no data is lost during the merge.

  • We try to not use models that are leaks and are only available openly, if you find any mode that do not follow this please let us know.

  • Community discussion on new models to add to the Merged Model are welcome.

Note: We do realize that the list of models used was mostly of NSFW models, this is because there is a lack of SFW models out there, if you have some SFW models please let us know so we can use them.

Mega Models List:

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference API
Unable to determine this model's library. Check the docs .