Hi @nota-github 🤗
I'm Niels and work as part of the open-source team at Hugging Face. I discovered your work through Papers with Code as your paper "Quantize the Target, Quantize the Drafter: Efficient Inference with Qwen3.5-4B" was featured.
The Hugging Face paper page (https://huggingface.co/papers/2607.04244) lets people discuss your paper and find related artifacts (such as model checkpoints, code, and demos). You can also claim the paper as yours, which will show up on your public profile on HF, and link your GitHub repository.
I saw in your GitHub README that you plan to release the code and model checkpoints soon. Would you be open to hosting your quantized target model and speculative block-diffusion drafter checkpoints on https://huggingface.co/models?
Hosting them on Hugging Face will make your work much more discoverable and visible to the AI community. We can add metadata tags (such as text-generation, speculative-decoding, quantization, etc.) to the model cards so that users can easily find and load them, and link them directly to your Hugging Face paper page.
If you are interested, you can find a guide on uploading models here. If they are custom PyTorch models, you can leverage the PyTorchModelHubMixin class to easily integrate from_pretrained and push_to_hub methods, or you can directly use hf_hub_download to fetch the checkpoints.
Additionally, you can build a demo on Spaces to showcase your low-latency speculative decoding, and we can provide you with a ZeroGPU grant for free GPU-backed compute.
Let us know if you're interested or if you need any guidance/help setting this up!
Kind regards,
Niels
Hi @nota-github 🤗
I'm Niels and work as part of the open-source team at Hugging Face. I discovered your work through Papers with Code as your paper "Quantize the Target, Quantize the Drafter: Efficient Inference with Qwen3.5-4B" was featured.
The Hugging Face paper page (https://huggingface.co/papers/2607.04244) lets people discuss your paper and find related artifacts (such as model checkpoints, code, and demos). You can also claim the paper as yours, which will show up on your public profile on HF, and link your GitHub repository.
I saw in your GitHub README that you plan to release the code and model checkpoints soon. Would you be open to hosting your quantized target model and speculative block-diffusion drafter checkpoints on https://huggingface.co/models?
Hosting them on Hugging Face will make your work much more discoverable and visible to the AI community. We can add metadata tags (such as
text-generation,speculative-decoding,quantization, etc.) to the model cards so that users can easily find and load them, and link them directly to your Hugging Face paper page.If you are interested, you can find a guide on uploading models here. If they are custom PyTorch models, you can leverage the PyTorchModelHubMixin class to easily integrate
from_pretrainedandpush_to_hubmethods, or you can directly use hf_hub_download to fetch the checkpoints.Additionally, you can build a demo on Spaces to showcase your low-latency speculative decoding, and we can provide you with a ZeroGPU grant for free GPU-backed compute.
Let us know if you're interested or if you need any guidance/help setting this up!
Kind regards,
Niels