[docs] Adds a doc on LoRA support for diffusers#2086
Conversation
|
The documentation is not available anymore as the PR was closed or merged. |
| import torch | ||
|
|
||
| model_path = "sayakpaul/sd-model-finetuned-lora-t4" | ||
| pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", torch_dtype=torch.float16) |
There was a problem hiding this comment.
(nit) maybe we can show how to retrieve the base_model from the model card by loading the yaml code via huggingface_hub
There was a problem hiding this comment.
There was a problem hiding this comment.
from huggingface_hub.repocard import RepoCard
card = RepoCard.load("sayakpaul/sd-model-finetuned-lora-t4")
card.data.to_dict()["base_model"]
# 'CompVis/stable-diffusion-v1-4'I guess we would want to show it in a separate code snippet from the doc?
There was a problem hiding this comment.
Nice! Maybe include it as a tip below the current snippet?
There was a problem hiding this comment.
For me it's fine in the same code snippet
There was a problem hiding this comment.
See if the current changes make sense.
docs/source/en/training/lora.mdx
Outdated
|
|
||
| - Previous pretrained weights are kept frozen so that model is not prone to [catastrophic forgetting](https://www.pnas.org/doi/10.1073/pnas.1611835114). | ||
| - Rank-decomposition matrices have significantly fewer parameters than original model, which means that trained LoRA weights are easily portable. | ||
| - LoRA attention layers allow to control to which extent the model is adapted toward new training images via a `scale` parameter. |
There was a problem hiding this comment.
nit: so far we've only mentioned "update matrices", but not how they work or whether they contain attention layers. Maybe we should very briefly introduce the concept? Something simple like "LoRA matrices are added to the model attention layers and they control ..." could work.
There was a problem hiding this comment.
See if the current edits make sense.
| import torch | ||
|
|
||
| model_path = "sayakpaul/sd-model-finetuned-lora-t4" | ||
| pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", torch_dtype=torch.float16) |
There was a problem hiding this comment.
Nice! Maybe include it as a tip below the current snippet?
|
|
||
| prompt = "A pokemon with green eyes and red legs." | ||
| image = pipe(prompt, num_inference_steps=30, guidance_scale=7.5).images[0] | ||
| image.save("pokemon.png") |
There was a problem hiding this comment.
Just wondering, maybe display the image here? We never do it in the docs, what's your opinion about starting doing it to make things more visual?
There was a problem hiding this comment.
Diffusion for computer vision is definitely about visuals. I like the idea and I think we should definitely add it :)
patil-suraj
left a comment
patil-suraj
left a comment
There was a problem hiding this comment.
Thanks a lot for working on this. Looks great!
Co-authored-by: Pedro Cuenca <pedro@huggingface.co>
Co-authored-by: Pedro Cuenca <pedro@huggingface.co>
* add: a doc on LoRA support in diffusers. * Apply suggestions from code review Co-authored-by: Pedro Cuenca <pedro@huggingface.co> * apply PR suggestions. * Apply suggestions from code review Co-authored-by: Pedro Cuenca <pedro@huggingface.co> * remove visually incoherent elements. Co-authored-by: Pedro Cuenca <pedro@huggingface.co>
5 participants
Closes #2028.