SCM-7B

Official SCM (Streaming Content Monitor) model based on Qwen/Qwen2.5-7B for the NeurIPS 2025 paper:

"From Judgment to Interference: Early Stopping LLM Harmful Outputs via Streaming Content Monitoring"

Model Description

SCM-7B is a dual-task model that performs both token-level and sequence-level safety classification, training with a logic consistency loss to ensure coherence between the two tasks.

• Base Model: Qwen/Qwen2.5-7B
• Architecture: QwenForDualTask (custom, based on Qwen2PreTrainedModel)
• Parameters: 7B

Usage

from transformers import AutoTokenizer, AutoModel

tokenizer = AutoTokenizer.from_pretrained("liyang-ict/SCM-7B")
model = AutoModel.from_pretrained("liyang-ict/SCM-7B", trust_remote_code=True)

Citation

If you find this model useful, please cite our paper:

@inproceedings{NEURIPS2025_4e315702,
  author = {Li, Yang and Sheng, Qiang and Yang, Yehan and Zhang, Xueyao and Cao, Juan},
  booktitle = {Advances in Neural Information Processing Systems},
  editor = {D. Belgrave and C. Zhang and H. Lin and R. Pascanu and P. Koniusz and M. Ghassemi and N. Chen},
  pages = {54305--54333},
  publisher = {Curran Associates, Inc.},
  title = {From Judgment to Interference: Early Stopping LLM Harmful Outputs via Streaming Content Monitoring},
  url = {https://proceedings.neurips.cc/paper_files/paper/2025/file/4e3157021c5f833bb2204081f1dda573-Paper-Conference.pdf},
  volume = {38},
  year = {2025}
}

License

This model is released under the Apache 2.0 License, following the license of the base Qwen2.5 model.