Pretraining data
Improving corpus quality, mixture design, long-tail knowledge coverage, and multimodal alignment at scale.
Large Language Model Data
Bingli Wang.
I work on data for large language models, spanning pretraining corpora, evaluation benchmarks, and the loops that turn model behavior into better training signals.
01 / About
Data is where my
research begins.
I am a PhD student in Computer Science at Southeast University. My research focuses on data for large language models.
I work across the full data lifecycle: constructing pretraining corpora, designing evaluation benchmarks, and turning evaluation signals into targeted data for the next training cycle.
Evaluation data
Designing benchmarks that measure knowledge, multimodal understanding, and agent capabilities with high signal.
Evaluation-to-training loops
Turning model failures, expert feedback, and evaluation traces into targeted data for the next training cycle.
02 / Education
Academic background.
2026 - Present
Southeast University
PhD in Computer Science and Technology
2022 - 2026
Sichuan Agricultural University
Undergraduate degree in Computer Science and Technology
03 / Selected work
Research, selected.
All publications
SuperGPQA: Scaling LLM Evaluation across 285 Graduate Disciplines
A 26K-question benchmark built through expert annotation and automated quality control.
Paper
Intern-S1-Pro: Scientific Multimodal Foundation Model at Trillion Scale
Contributed to long-tail pretraining data and the multimodal caption data pipeline.
Paper
COHERENCE: Benchmarking Fine-Grained Image-Text Alignment
A benchmark for image-text alignment in naturally interleaved multimodal contexts.
PaperMore papers, citation details, and the complete publication record.
View Google Scholar04 / Experience
Research in practice.
May 2026 - Present
Ubiquant ยท IQust Foundation Model
Research Intern
Building evaluation-to-training loops for long-horizon agent capabilities.
Jul 2025 - May 2026
Shanghai AI Laboratory
Research Intern
Delivered multimodal caption data and built evaluation-to-training loops for multimodal understanding.
Aug 2024 - May 2025
M-A-P
Research Intern
Led large-scale benchmark data construction and quality control for graduate-level LLM evaluation.
Apr 2024 - May 2025
Tsinghua University (Shenzhen)
Research Intern
Conducted research on machine learning and model evaluation.