2026
TBQ: Trajectory-Aligned Branch-Aware Quantization for Few-Step MoE-Like Video Diffusion Models
Jinyang Du, Shenghao Jin, Ziqian Xu, Ruihao Gong, yang yong, Shiqiao Gu, Jinyang Guo, Xianglong Liu
The fourth International Workshop on Rich Media with Generative AI (RichMediaGAI '26) 2026 First Author Workshop
TBQ studies the deployment gap of MoE-like video diffusion models and combines few-step distillation with branch-aware low-bit calibration to make compression more stable under tight memory and latency budgets.
TBQ: Trajectory-Aligned Branch-Aware Quantization for Few-Step MoE-Like Video Diffusion Models
Jinyang Du, Shenghao Jin, Ziqian Xu, Ruihao Gong, yang yong, Shiqiao Gu, Jinyang Guo, Xianglong Liu
The fourth International Workshop on Rich Media with Generative AI (RichMediaGAI '26) 2026 First Author Workshop
TBQ studies the deployment gap of MoE-like video diffusion models and combines few-step distillation with branch-aware low-bit calibration to make compression more stable under tight memory and latency budgets.
Half-S: Halving the Scale for Near-Lossless 4-Bit LLM Training
Jinyang Du, Ruihao Gong#, Linghan Ai, Zining Wang, Yunke Peng, Yao Wang, Lei Yan, Xuefei Wang, Yaoyuan Wang, Jinyang Guo, Dahua Lin, Xianglong Liu (# corresponding author)
Findings of the Association for Computational Linguistics (ACL Findings) 2026 First Author ACL Findings
Half-S revisits FP4 scaling for heavy-tailed LLM tensors and proposes a minimal scale correction that improves quantization grid utilization for practical near-lossless 4-bit training.
Half-S: Halving the Scale for Near-Lossless 4-Bit LLM Training
Jinyang Du, Ruihao Gong#, Linghan Ai, Zining Wang, Yunke Peng, Yao Wang, Lei Yan, Xuefei Wang, Yaoyuan Wang, Jinyang Guo, Dahua Lin, Xianglong Liu (# corresponding author)
Findings of the Association for Computational Linguistics (ACL Findings) 2026 First Author ACL Findings
Half-S revisits FP4 scaling for heavy-tailed LLM tensors and proposes a minimal scale correction that improves quantization grid utilization for practical near-lossless 4-bit training.
2025
Low-bit FlashAttention Accelerated Operator Design Based on Triton
Jinyang Du, Jinyang Guo, Yifu Ding, Xianglong Liu
IEEE/CVF International Conference on Computer Vision Workshops (ICCVW) 2025 First Author
This work designs low-bit FlashAttention operators with Triton, using operator fusion and mixed-precision execution to improve long-context quantized inference efficiency.
Low-bit FlashAttention Accelerated Operator Design Based on Triton
Jinyang Du, Jinyang Guo, Yifu Ding, Xianglong Liu
IEEE/CVF International Conference on Computer Vision Workshops (ICCVW) 2025 First Author
This work designs low-bit FlashAttention operators with Triton, using operator fusion and mixed-precision execution to improve long-context quantized inference efficiency.
A Survey of Low-bit Large Language Models: Basics, Systems, and Algorithms
Ruihao Gong, Yifu Ding, Zining Wang, Chengtao Lv, Xingyu Zheng, Jinyang Du, Yong Yang, Shiqiao Gu, Haotong Qin, Jinyang Guo, Dahua Lin, Michele Magno, Xianglong Liu
Neural Networks 2025 Journal
This survey reviews low-bit quantization for large language models from basic formats, system support, and algorithmic strategies, connecting practical toolchains with future efficient LLM deployment.
A Survey of Low-bit Large Language Models: Basics, Systems, and Algorithms
Ruihao Gong, Yifu Ding, Zining Wang, Chengtao Lv, Xingyu Zheng, Jinyang Du, Yong Yang, Shiqiao Gu, Haotong Qin, Jinyang Guo, Dahua Lin, Michele Magno, Xianglong Liu
Neural Networks 2025 Journal
This survey reviews low-bit quantization for large language models from basic formats, system support, and algorithmic strategies, connecting practical toolchains with future efficient LLM deployment.