Release models on HF

[NielsRogge]

Hi @Xuan-World,

Niels here from the open-source team at Hugging Face. I discovered your work through the daily papers: https://huggingface.co/papers/2409.08513 (feel free to claim authorship with your HF account). I work together with AK on improving the visibility of researchers' work on the hub.

I was wondering you'd be up for collaborating on releasing the checkpoints on the 🤗 hub, to improve their discoverability. We can add tags so that people find them when filtering https://huggingface.co/models.

For instance in this case, the "zero-shot-object-detection" tag seems useful: https://huggingface.co/models?pipeline_tag=zero-shot-object-detection&sort=trending.

Uploading models

See here for a guide: https://huggingface.co/docs/hub/models-uploading. In case the models are custom PyTorch model, we could probably leverage the PyTorchModelHubMixin class which adds from_pretrained and push_to_hub to each model. Alternatively, one can leverages the hf_hub_download one-liner to download a checkpoint from the hub.

We encourage researchers to push each model checkpoint to a separate model repository, so that things like download stats also work. Moreover, we can then link the checkpoints to the paper page, improving their visibility.

Let me know if you're interested/need any help regarding this!

Cheers,

Niels ML Engineer @ HF 🤗

[KT313]

@Xuan-World , I'm completely on the same page as Niels, can you please release the weights on a platform that does not require wechat and or a chinese phone number to download the weights, so everyone can test them? You can't really say "We provide the all the Model Weights for community." if it's only accessible to chinese people. I would really like to include your work as part of a project but this just makes it impossible. Preferably huggingface, in safetensors format please. Would really appreciate it :)

我完全同意Niels的看法。请问能否将模型权重发布在不需要微信或中国手机号码的平台上，这样所有人都可以测试这些权重？如果只对中国用户开放，就不能说“我们为社区提供了所有模型权重”。我非常希望能将你们的工作作为项目的一部分，但这种限制使得实现这一点变得不可能。希望能发布到Huggingface上，并且使用safetensors格式。非常感谢！

[Xuan-World]

@Xuan-World , I'm completely on the same page as Niels, can you please release the weights on a platform that does not require wechat and or a chinese phone number to download the weights, so everyone can test them? You can't really say "We provide the all the Model Weights for community." if it's only accessible to chinese people. I would really like to include your work as part of a project but this just makes it impossible. Preferably huggingface, in safetensors format please. Would really appreciate it :)

我完全同意Niels的看法。请问能否将模型权重发布在不需要微信或中国手机号码的平台上，这样所有人都可以测试这些权重？如果只对中国用户开放，就不能说“我们为社区提供了所有模型权重”。我非常希望能将你们的工作作为项目的一部分，但这种限制使得实现这一点变得不可能。希望能发布到Huggingface上，并且使用safetensors格式。非常感谢！

Ok, we will make it possible soon.

[NielsRogge]

Hi @Xuan-World would it be possible to keep the issue open so I can track it?

Happy to send a PR to make the weights available on HF

[Xuan-World]

We have offer you a link to download weights and check the zero-shot results pictures on HuggingFace 🤗 . @NielsRogge @KT313

[NielsRogge]

Great! I’ve opened https://huggingface.co/Xuan-World/Mamba-YOLO-World/discussions/1 for adding a pipeline tag.

Not sure if it’s applicable, but in case your model is an nn.Module which only takes arguments in its init method which are JSON serializable, then we could adopt the PytorchModelHubMixin class linked above, which uses safetensors, which is considered more safe than pickle for storing the weights.

Would you also be interested in creating a Gradio demo, so that people can quickly try out your model?