Yapei Chang | University of Maryland

About Me

Hi! I’m a PhD student at the University of Maryland, College Park, advised by Prof. Mohit Iyyer with a research focus on NLP. Prior to transferring to UMD, I was affiliated with the UMass Amherst NLP Lab, where I was also advised by Prof. Iyyer. My current main research interests include:

Reliable / cost-effective metrics for LLM training [BLEUBERI (NeurIPS’25)] and evaluation [BooookScore (ICLR’24)]
Scalable pipelines for model evaluation and training based on synthetic data [CLIPPER (COLM’25)] or rewriting the web [How2everything (preprint)].

I like gaming and tinkering with mechanical keyboards in the free time I aggregate from my fluctuating work-life balance (it’s a work in progress ┐(‘～` )┌).

Updates

[Feb 2026] Released How2Everything, a web-scale framework mining step-by-step procedures across diverse domains + a reliable scoring protocol for scalable evaluation and RL-based improvement of step-by-step generation.
[Dec 2025] Olmo3 is out! All code, training data, and intermediate model checkpoints are publicly available.
[Jul 2025] Invited talk at the Google Multilinguality Reading Group.
[Jun 2025] Started my internship at Ai2 (OLMo team), working with Kyle Lo & Luca Soldaini!
[May 2025] Released BLEUBERI, which shows that simple string-matching metrics like BLEU can be used as effective rewards to improve general instruction following.
[Jan. 2025] Moved to UMD, College Park! Also got my MS degree from UMass Amherst.
[Oct. 2024] ✈ Attended COLM to present FABLES, a paper on faithfulness evaluation in book-length summarization.
[May 2024] ✈ Attended ICLR and gave an oral presentation for BooookScore, a paper on book-length summarization and finegrained coherence evaluation.
[Sep. 2022] Started my PhD!

Selected Publications

For a more up-to-date list of publications, please visit my Google Scholar page.

Preprint

How2Everything: Mining the Web for How-To Procedures to Evaluate and Improve LLMs

Yapei Chang, Kyle Lo, Mohit Iyyer, Luca Soldaini

Preprint, 2026.

PDF Code

Technical Report

Olmo 3

Olmo Team, including Yapei Chang

Technical Report, 2025.

PDF

Preprint

Semantically-Aware Rewards for Open-Ended R1 Training in Free-Form Generation

Zongxia Li, Yapei Chang, Yuhang Zhou, Xiyang Wu, Zichao Liang, Yoo Yeon Sung, Jordan Lee Boyd-Graber

Preprint, 2025.

PDF Code

NeurIPS

BLEUBERI: BLEU is a surprisingly effective reward for instruction following

Yapei Chang, Yekyung Kim, Michael Krumdick, Amir Zadeh, Chuan Li, Chris Tanner, Mohit Iyyer

Conference on Neural Information Processing Systems (NeurIPS), 2025.

PDF Code

COLM

CLIPPER: Compression enables long-context synthetic data generation

Chau Minh Pham, Yapei Chang, Mohit Iyyer

Conference on Language Modeling (COLM), 2025.

PDF Code

EMNLP

PostMark: A Robust Blackbox Watermark for Large Language Models

Yapei Chang, Kalpesh Krishna, Amir Houmansadr, John Wieting, Mohit Iyyer

Conference on Empirical Methods in Natural Language Processing (EMNLP), 2024.

PDF Code

COLM

FABLES: Evaluating faithfulness and content selection in book-length summarization

Yekyung Kim, Yapei Chang, Marzena Karpinska, Aparna Garimella, Varun Manjunatha, Kyle Lo, Tanya Goyal, Mohit Iyyer

Conference on Language Modeling (COLM), 2024.

PDF Code Project Page

ICLR

BooookScore: A systematic exploration of book-length summarization in the era of LLMs

Yapei Chang, Kyle Lo, Tanya Goyal, Mohit Iyyer

International Conference on Learning Representations (ICLR), 2024.

PDF Code Project Page Oral Presentation (Top 1.2%)

Projects

📚 Book summarization with LLMs