Wenhua Cheng 程文华

Wenhua Cheng 程文华

Wenhua Cheng is a Software Engineer at Intel, with a strong background in large language model (LLM) quantization, model compression, and computer vision. Previously, he worked at Alibaba Cloud as a software engineer and at Intel Labs as a researcher. Wenhua holds a Master’s degree from Zhejiang University and a Bachelor’s degree from Nanjing University of Science and Technology.

Wenhua’s expertise spans two main domains:

  1. LLM Compression: As the first author, he has contributed to methods such as SignRound and SignRoundV2, signed gradient descent-based rounding optimizations for LLM quantization, and TEQ (Trainable Equivalent Transformation).

  2. Computer Vision: As a co-first author, he won two championships in the DawnBench competition, outperforming teams from Huawei, Google, and others. As the first author, he also ranked 4th and 6th in two tracks of the 2017 ICDAR Scene Text Detection Competition.

Wenhua has filed 21 patents, 11 of which have been granted. Over the past four years, he has contributed to 300+ merged PRs.

Wenhua is open to fully remote positions that align with his background. Reaching out via email or any other ways would be greatly appreciated.

📝 Publications

sym

[2025] SignRoundV2: Closing the Performance Gap in Extremely Low-Bit Post-Training Quantization for LLMs arxiv GitHub Stars

Wenhua Cheng, Weiwei Zhang, Heng Guo, Haihao Shen

sym

[2023] Optimize Weight Rounding via Signed Gradient Descent for the Quantization of LLMs EMNLP GitHub Stars

Wenhua Cheng, Weiwei Zhang, Haihao Shen, Yiyang Cai, Xin He, Lv Kaokao, Yi Liu

sym

[2023] TEQ: Trainable Equivalent Transformation for Quantization of LLMs arxiv GitHub Stars

Wenhua Cheng, Yiyang Cai, Kaokao Lv, Haihao Shen

🎖 Honors and Awards

  • 2020 Ranked 1st in DawnBench competition (Training Time track)
  • 2020 Ranked 1st in DawnBench competition (Training Cost track)
  • 2017 Ranked 4th in DawnBench competition (Scene text detection track)
  • 2017 Ranked 6th in DawnBench competition (COCO text detection track)

📖 Educations

  • 2011.09 – 2014.06, Zhejiang University – Control Science and Engineering

  • 2005.09 – 2009.06, Nanjing University of Science and Technology – Automation