mdm-prime-v2-slimpajama

MDM-Prime-v2-Slimpajama

Paper | Project Page | GitHub

MDM-Prime-v2 is an enhanced version of the MDM-Prime framework. MDM-Prime is a discrete diffusion model enhanced with the Partial masking scheme (Prime). It enables fine-grained denoising and improves generation quality across both image and text domains. Refer to our papers for more details:

• MDM-Prime-v2: Binary Encoding and Index Shuffling Enable Scaling of Diffusion Language Models.
• Beyond Masked and Unmasked: Discrete Diffusion Models via Partial Masking.

---

Model Details

• Dataset: Slimpajama
• Model Size: 1.1B
• Context Length: 2,048

---

How to Use

To download the weights, one can download the huggingface_hub library via pip install -U huggingface_hub and perform the following python code:

from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="chen-hao-chao/mdm-prime-v2-slimpajama",
    filename="${checkpoint_name}"
)

Replace ${checkpoint_name} with mdm-prime-v2-3300flops.pth or mdm-prime-v2-6600flops.pth. Checkpoints with -weight-only indicates that only the weights of the model are included. This enables faster download and inference. This repository is organized as follows:

mdm-prime-v2-slimpajama/
├── README.md
├── mdm-prime-v2-3300flops.pth
├── mdm-prime-v2-3300flops-weight-only.pth
├── mdm-prime-v2-6600flops.pth
└── mdm-prime-v2-6600flops-weight-only.pth

For more details regarding the training and inference processes, please refer to our github repository: chen-hao-chao/mdm-prime-v2.

---

Citing MDM-Prime and MDM-Prime-v2

If you find this repository useful, please consider citing our papers.

@article{chao2026mdmprimev2,
      title = {{MDM-Prime-v2: Binary Encoding and Index Shuffling Enable Scaling of Diffusion Language Models}}, 
      author = {Chen-Hao Chao, Wei-Fang Sun, Junwei Quan, Chun-Yi Lee, Rahul G. Krishnan},
      year = {2026},
}
@inproceedings{chao2025mdmprime,
      title = {{Beyond Masked and Unmasked: Discrete Diffusion Models via Partial Masking}}, 
      author = {Chen-Hao Chao, Wei-Fang Sun, Hanwen Liang, Chun-Yi Lee, Rahul G. Krishnan},
      booktitle = {Proceedings of the Conference on Neural Information Processing Systems (NeurIPS)},
      year = {2025},
}