Skip to content

Host MeshFlow checkpoints on the Hugging Face Model Hub #4

Description

@NielsRogge

Hi @qiisun 🤗

I'm Niels, part of the community science team at Hugging Face.

First, congratulations on MeshFlow getting accepted to SIGGRAPH 2026! It looks like a really impressive and fast approach to direct 3D mesh generation.

I noticed that you have already uploaded your dataset and model checkpoints to Hugging Face under the dataset repository: https://huggingface.co/datasets/qsun2001/meshflow. This is awesome!

However, since you have released 4 pre-trained checkpoints (for bench, chair, lamp, and table categories), would you like to host them on the Hugging Face Model Hub (https://huggingface.co/models) instead of keeping the .pt files inside the dataset repository?

Hosting your checkpoints as proper model repositories on Hugging Face provides several key benefits:

  1. Better Discoverability: They will be categorized and searchable as machine learning models, making them much easier for the 3D generation community to find.
  2. Individual Download Statistics: You will get dedicated download tracking and analytics for each of your model checkpoints.
  3. Structured Metadata & Model Cards: You can add specific model cards, metadata tags (like PyTorch, 3D, or custom categories), and cleanly separate your training data from your model checkpoints.
  4. Paper Page Integration: Once uploaded to the Model Hub, we can link each model directly to your official Hugging Face Paper Page (https://huggingface.co/papers/2606.23489), so readers can find your models instantly.

Additionally, since you already have a gradio_demo.py script in your repository, you can easily build an interactive web demo for MeshFlow on Spaces. We can provide you with a free ZeroGPU grant, which gives you free GPU-backed compute for eligible demo Spaces.

If you are open to it, you can easily create model repositories via the UI or CLI and upload your weights. Here is a quick guide on uploading models to the Hub.

Let me know if you are interested or if you need any assistance with this!

Best regards,

Niels
ML Engineer @ HF 🤗

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions