Hi @mingkunlei 馃
I'm Niels and work as part of the open-source team at Hugging Face. I discovered your work on BudCache through Arxiv and was very interested in the acceleration results you achieved for FLUX and Wan2.1.
I noticed your GitHub repository and saw the note in the README that "The additional release materials will be released later." I was wondering whether you would like to host the pre-searched cache policies (Stage 1 checkpoints) and the refined sampling schedules (Stage 2 checkpoints) on https://huggingface.co/models?
Hosting these artifacts on Hugging Face would significantly improve their discoverability and make it much easier for the community to accelerate their own generations using BudCache. We can add metadata tags (linking them to the respective base models) so that people find them easily, and they can be linked directly to your paper page.
If you're interested, you can find a guide for uploading models here. For custom checkpoint files like these, users can simply use hf_hub_download to pull them directly into their inference scripts.
Additionally, you might consider building a demo for BudCache on Spaces. We can provide you with a ZeroGPU grant to support the compute for the demo.
Let me know if you're interested or need any guidance!
Kind regards,
Niels
ML Engineer @ HF 馃
Hi @mingkunlei 馃
I'm Niels and work as part of the open-source team at Hugging Face. I discovered your work on BudCache through Arxiv and was very interested in the acceleration results you achieved for FLUX and Wan2.1.
I noticed your GitHub repository and saw the note in the README that "The additional release materials will be released later." I was wondering whether you would like to host the pre-searched cache policies (Stage 1 checkpoints) and the refined sampling schedules (Stage 2 checkpoints) on https://huggingface.co/models?
Hosting these artifacts on Hugging Face would significantly improve their discoverability and make it much easier for the community to accelerate their own generations using BudCache. We can add metadata tags (linking them to the respective base models) so that people find them easily, and they can be linked directly to your paper page.
If you're interested, you can find a guide for uploading models here. For custom checkpoint files like these, users can simply use hf_hub_download to pull them directly into their inference scripts.
Additionally, you might consider building a demo for BudCache on Spaces. We can provide you with a ZeroGPU grant to support the compute for the demo.
Let me know if you're interested or need any guidance!
Kind regards,
Niels
ML Engineer @ HF 馃