Biography
Hello! I am a final-year PhD candidate in Computer Science at Hong Kong Baptist University (HKBU), where I am advised by Dr. Jing Ma. In 2024, I also visited the National University of Singapore (NUS), working under the guidance of Professor Mohan Kankanhalli. My research interests lie in the development of novel foundation models and their applications. Specifically, this includes:
I will graduate in the fall of 2025 and am seeking a full-time research scientist position. If you are interested, please feel free to contact me. (email: cszyluo AT comp.hkbu.edu.hk, wechat ID: chiyeunglaw)
For those who are interested in working with me, please feel free to email me. Remote collaboration is also welcome!
Representative Works
- Evaluation for LMMs: VideoAutoArena: An Automated Arena for Evaluating Large Multimodal Models in Video Analysis through User Simulation. We introduce the first arena-style benchmark for Large Multimodal Models (LMMs) in video analysis, designed to evaluate LMMs through user-centric assessments—similar to MT-Bench and AlpacaEval. (Page, Arxiv 2024)
- Code Reasoing of LLMs: WizardCoder: Empowering Code Large Language Models with Evol-Instruct This is the first work that truly closes the gap between open-source and closed-source Code LLMs, widely used in the following works. [2024/01/04] We released WizardCoder-33B-V1.1, the SOTA OSS Code LLM on EvalPlus Leaderboard. [2023/08/26] We released WizardCoder-Python-34B-V1.0, which achieves the 73.2 pass@1 and surpasses GPT4 (2023/03/15), ChatGPT-3.5, and Claude2 on the HumanEval. [2023/06/16] Our WizardCoder-15B-V1.0 achieves 57.3 pass@1 score on HumanEval, more than 20 points higher than the SOTA open-source LLMs. (ICLR 2024).
- Multimodal Retrieval: LexLIP: Lexicon-Bottlenecked Language-Image Pre-Training for Large-Scale Image-Text Sparse Retrieval. Our LexLIP introduces a new paradigm for image-text retrieval, which outperforms CLIP with 5.8x faster retrieval speed and 19.1x less index storage memory. (Codes, ICCV 2023)
- Outlier Neurons in LLMs: Positional Artefacts Propagate Through Masked Language Model Embeddings We are the first to discover the outlier neurons phenomenon in the Transformer-based LMs. This finding is widely used to quantize transformers in the recent works. (ACL 2021)
Working Experiences
- July. 2024 - Dec. 2024, Research Intern, Rhymes AI (aka. 01.AI).
- Feb. 2024 - June. 2024, Research Intern, Language Technology Lab, Alibaba DAMO Academy - Singapore.
- May. 2023 - Jan. 2024, WizardCoder Core Contributor, WizardLM Team, Microsoft.
- Nov. 2022 - Jan. 2024, Research Intern, Data, Knowledge, and Intelligence Group, Microsoft Research Lab - Asia (MSRA).
- Jan. 2022 - July. 2022, Research Intern, NLP Research Group, Fuxi AI Lab, NetEase, Inc.
- Jan. 2021 - July. 2021, Research Intern, NLP Research Group, Fuxi AI Lab, NetEase, Inc.
- May. 2020 - Aug. 2020, Engineer Intern, Jovi Chatbot Group, VIVO Communication Technology Co. Ltd.
Academic Services
- Conference Program Committee Member: ICLR 2025, EMNLP 2022-23, ACL 2023, ACL ARR 2023-24, COLING 2025, AAAI 2023-25, ACCV 2022, ECCV 2024, CVPR 2025
- Workshop Reviewer: Instruction Workshop @NeurIPS 2023
- Student Volunteer: EMNLP 2022, COLING 2025
Invited Talks
- "WizardCoder: Empowering Code Large Language Models with Evol-Instruct" at Neurosymbolic Reading Group, MIT. (Slides)