Release K-LoRA models on Hugging Face

[NielsRogge]

Hi @ouyangziheng 🤗

I'm Niels and work as part of the open-source team at Hugging Face. I discovered your work through Hugging Face's daily papers as yours got featured: https://huggingface.co/papers/2502.18461. The paper page lets people discuss about your paper and lets them find artifacts about it (your models for instance), you can also claim the paper as yours which will show up on your public profile at HF.

Your paper on K-LoRA is very interesting and your results are impressive. It would be great to have your model checkpoints available on the Hugging Face Hub to improve their discoverability and visibility. We can add tags to the model cards so that people can easily find them when searching on the Hugging Face Model Hub. Your paper, accepted by CVPR 2025, is a significant contribution to the field.

Would you be interested in hosting your K-LoRA models (for both Stable Diffusion and FLUX) on Hugging Face? Hosting on Hugging Face will also allow users to easily download and use your models via the Hugging Face Hub's API:

from diffusers import StableDiffusionPipeline

pipe = StableDiffusionPipeline.from_pretrained("your-hf-org-or-username/k-lora-stable-diffusion")

If you are interested, here's a guide on uploading models to the Hugging Face Hub: https://huggingface.co/docs/hub/models-uploading. If it's a custom PyTorch model, you can use the PyTorchModelHubMixin class which adds from_pretrained and push_to_hub to the model, allowing users to upload and download models easily. We can help with this process. We encourage researchers to push each model checkpoint to a separate model repository, so that things like download stats also work. We can then also link the checkpoints to the paper page.

After uploading, we can link the models to the paper page (https://huggingface.co/docs/hub/en/model-cards#linking-a-paper) for better discoverability. You could also consider creating a demo on Hugging Face Spaces (https://huggingface.co/spaces) which we can even support with a ZeroGPU grant (https://huggingface.co/docs/hub/en/spaces-gpus#community-gpu-grants) to provide free A100 GPUs.

Let me know if you're interested or need any guidance.

Kind regards,

Niels
ML Engineer @ HF 🤗

[ouyangziheng]

Hi, @NielsRogge

I'm very happy to receive your recognition and support! I'm highly interested in deploying to Hugging Face.

However, I have a question. In the images shown in the README, we actually used two LoRAs from Glif Loradex Trainer and fused them using our K-LoRA method to generate the results.

This means there isn't actually a merged model like "your-hf-org-or-username/k-lora-stable-diffusion". How should we handle this situation?

I appreciate your reply.

[NielsRogge]

I'll ping @multimodalart here!