Haoqian Meng

I am an incoming Ph.D. student in Computer Science at Zhejiang University. I am currently an undergraduate student in the Elite Class at the School of Computer Science and Technology, Tianjin University (2023–2027), advised by Prof. Peng Zhang.

Research Interests

My research interests include efficient large language model (LLM) inference, post-training quantization, low-precision numerical formats, and hardware-aware algorithms and kernels. I am particularly interested in making quantized LLM inference more accurate and efficient on modern accelerators.

Publications

ARCQuant: Boosting NVFP4 Quantization with Augmented Residual Channels for LLMs

Published in ACL 2026, 2026

Augments activations with quantized residual channels to improve NVFP4 accuracy while retaining a unified format and efficient GEMM execution.

Recommended citation: Haoqian Meng, Yilun Luo, Yafei Zhao, Wenyuan Liu, Peng Zhang, and Xindian Ma. (2026). ARCQuant: Boosting NVFP4 Quantization with Augmented Residual Channels for LLMs. Proceedings of ACL 2026.
Download Paper

SharQ: Bridging Activation Sparsity and FP4 Quantization for LLM Inference

Published in arXiv preprint, 2026

A training-free inference method that combines activation sparsity with FP4 quantization through an online sparse–dense decomposition.

Recommended citation: Haoqian Meng, Yilun Luo, Yafei Zhao, Wenyuan Liu, Huaqing Zheng, Xindian Ma, and Peng Zhang. (2026). SharQ: Bridging Activation Sparsity and FP4 Quantization for LLM Inference. arXiv:2606.26587.
Download Paper

MicroMix: Efficient Mixed-Precision Quantization with Microscaling Formats for Large Language Models

Published in ICLR 2026, 2026

Co-designs mixed MXFP4, MXFP6, and MXFP8 quantization with matrix multiplication kernels for efficient LLM inference.

Recommended citation: Wenyuan Liu, Haoqian Meng, Yilun Luo, Peng Zhang, and Xindian Ma. (2026). MicroMix: Efficient Mixed-Precision Quantization with Microscaling Formats for Large Language Models. International Conference on Learning Representations (ICLR).
Download Paper

Post-Training Quantization of OpenPangu Models for Efficient Deployment on Atlas A2

Published in arXiv preprint, 2025

Studies INT8 and W4A8 post-training quantization for deploying OpenPangu models efficiently on Huawei Atlas A2 NPUs.

Recommended citation: Yilun Luo, HuaQing Zheng, Haoqian Meng, Wenyuan Liu, and Peng Zhang. (2025). Post-Training Quantization of OpenPangu Models for Efficient Deployment on Atlas A2. arXiv:2512.23367.
Download Paper

Education

2027– — Incoming Ph.D. student in Computer Science, Zhejiang University.
2023–2027 — Undergraduate student, School of Computer Science and Technology, Tianjin University (Elite Class).
- Advisor: Prof. Peng Zhang.