Songyang Zhang
Researcher at 腾讯
Xuhui District, Shanghai, China
Summary
🤩
Rockstar🎓
Top SchoolSongyang Zhang is a researcher with a decade of experience building and evaluating large multimodal and language models, currently leading a team that develops open-source evaluation and pre-training tooling such as OpenCompass and contributions to InternLM and OpenMMLab. Based in Shanghai and affiliated with Shanghai AI Lab and Tencent, he blends deep research experience (PhD-level training) with hands-on engineering—particularly backend evaluation pipelines and dataset/configuration refinement for benchmarks like MMLU and LCSTS. His work improves model assessment and post-training alignment at scale, and he has a track record of strengthening LLM judge capabilities for more robust automated evaluation. An uncommon strength is his simultaneous focus on open-source platform maintenance and core pre-training research, bridging reproducible tooling with frontier model development.
10 years of coding experience
3 years of employment as a software developer
Doctor of Philosophy - PhD Computer Science and Technology, Doctor of Philosophy - PhD Computer Science and Technology at ShanghaiTech UniversityDoctor of Philosophy - PhD, Doctor of Philosophy - PhD at University of Chinese Academy of SciencesBachelor of Engineering - BE Electronical Information and Engineering, Bachelor of Engineering - BE Electronical Information and Engineering at Beihang University
Chinese, English
Github Skills (9)
configuration-management10
lm10
benchmarking10
benchmark10
large-language-models10
python10
evaluation10
llm10
data-processing9
Programming languages (10)
Github contributions (5)
open-compass/opencompass
Jul 2023 - Mar 2025
OpenCompass is an LLM evaluation platform, supporting a wide range of models (Llama3, Mistral, InternLM2,GPT-4,LLaMa2, Qwen,GLM, Claude, etc) over 100+ datasets.
Role in this project:
Back-end Developer
Back-end DeveloperContributions:1 release, 359 reviews, 229 PRs in 1 year 9 months
Contributions summary:Songyang's commits focus on modifying the configuration files related to summarization and datasets, particularly for the MMLU benchmark and the LCSTS dataset. These changes involve adjusting summarization groups, and dataset configurations, suggesting a focus on refining the evaluation platform for Large Language Models (LLMs). Furthermore, the modifications to include various configurations for the LLM Judge point to the user's contributions in improving the system's evaluation capabilities. These tasks suggest the user plays a role in data pre-processing or evaluation pipeline development.
benchmarkchatgptevaluationlarge-language-modelllama2
tonysy/STANet-PyTorch
Nov 2018 - Sep 2020
Contributions:6 commits, 5 pushes, 2 comments in 1 year 10 months
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.