LinesHogan

I am Zeng Yuanhao (曾缘浩), a 1st-year PhD student at ShanghaiTech University and BIGAI. I received my bachelor’s degree from the CS Experimental Class at BUPT. I am currently interning at ByteDance.

My research focuses on algorithm-system codesign for large language model agents — I design efficient algorithms that solve real-world challenges and build systems that make them practical. I enjoy (very) smart ideas in theory, algorithms, and systems, and will share some cool insights in my blogs.

I develop tLLM, a “mod loader” for vLLM that enables efficient decoding with test-time algorithms that intervene on the model’s internal states.

My recent work introduces Exploratory Sampling (ESamp), a decoding-time method that encourages LLMs to explore semantically diverse reasoning paths via online latent distillation. See you in Seoul!

news

Jul 02, 2026	Our work on Exploratory Sampling (ESamp), Large Language Models Explore by Latent Distilling, has been accepted to ICML 2026.

selected publications

ICML

Large Language Models Explore by Latent Distilling

Yuanhao Zeng, Ao Lu, Lufei Li, and 3 more authors

In International Conference on Machine Learning, 2026

Abs HTML Code

Exploratory Sampling (ESamp) is a decoding-time method for large language models that encourages semantic diversity during generation. It trains a lightweight Latent Distiller online to predict deep-layer hidden representations from shallow-layer representations; the prediction error acts as a novelty signal for reweighting candidate continuations toward less-explored reasoning paths.