-
Notifications
You must be signed in to change notification settings - Fork 2.4k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Make PaperCut and VoxelArt models load again #1730
Conversation
This corrects a regression in loading of these models due to a change of the embedding_manager parameter `num_vectors_per_token` Fixes #1718
@damian0815 I do not recall why we changed |
also fixes #1657 after setting this back to 1, please test if https://huggingface.co/datasets/Nerfgun3/bad_prompt/blob/main/bad_prompt_version2.pt loads (it has a vector length of 8) i do not know how to verify if it's actually working but @blessedcoolant has a prompt created in a similar way. |
Both |
IIUC, it's a "stock" negative prompt which one would use for all generations by including 'bad_prompt' or some such in every prompt (as opposed to copy-pasting a long list every time). Impossible to tell what it actually does without knowing what terms are included in the negation list. |
a bunch of the ones we make fun of for being pointless. nevertheless, commonly used. |
right, it's just that not knowing what it does makes it hard to know if it's being loaded correctly (all 8 vectors) or if it's being truncated (just the 1st vector is loaded/used and it "seems" ok but actually isnt). |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
let's merge it
I've just updated main using Git and running the latest, and I see Papercut and voxel_art are not currently loading. Do we need to run any update scripts to get these to work? |
This corrects a regression in loading of these models due to a change of the embedding_manager parameter
num_vectors_per_token
Reverting to the original version (1) does not seem to affect concept embeddings.
Fixes #1718