Skip to content

Do not save unet.unet keys when training LoRAs #5977

Description

@apolinario

Describe the bug

When training LoRAs, sometimes they end up with a unet.unet key like in this case https://huggingface.co/sayakpaul/new-lora-check-v15/blob/main/pytorch_lora_weights.safetensors5/tree/main

unet.unet.up_blocks.1.attentions.2.transformer_blocks.1.attn2.processor.to_v_lora.up.weight

While loading up this format is supported on diffusers, it is probably undesirable to have such keys and maybe better off deprecated and make sure all keys have a single unet prefix

(Internal Slack thread: https://huggingface.slack.com/archives/C03UQJENJTV/p1701261840645189)

Reproduction

Training a LoRA with our example notebook/script yields the unet.unet output https://huggingface.co/LinoyTsaban/corgy_dog_LoRA

Logs

No response

System Info

diffusers==0.23.1

Who can help?

@sayakpaul

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

bugSomething isn't working

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions