LawGPT
LaWGPT 是一系列基于中文法律知识的开源大语言模型。该系列模型在通用中文基座模型的基础上扩充法律领域专有词表，并结合大规模中文法律语料预训练与指令微调，增强了模型在法律领域的语义理解和问答执行能力。
Toolkit
Software
Reusable systems and domain-oriented model releases.
Pianist Transformer
Welcome to Pianist Transformer, an expressive piano performance rendering system that transforms symbolic piano scores into human-like performances. It can generate natural, expressive, and editable MIDI performances from plain scores, making it useful for composers, producers, musicians, and researchers who want to preview, refine, or create realistic piano interpretations efficiently.
Evaluation
Benchmarks
Benchmarks and open testbeds.
FormalML
FormalML is a Lean 4 benchmark for formal subgoal completion in machine learning theory. It studies whether LLMs can fill in short but nontrivial missing proof steps inside human-written proof sketches, combining premise retrieval with research-level formal contexts.
ChinaTravel
A benchmark for evaluating travel-planning abilities of AI agents. Given a query, agents must gather information and generate feasible multi-day itineraries that satisfy hard logical constraints and soft user preferences in realistic Chinese travel scenarios.
TabFSBench
A benchmark for evaluating tabular learning models in open environments. It studies feature-shift scenarios across multiple model families and datasets, and also evaluates the behavior of LLMs and tabular LLMs under distribution changes that are common in real-world deployment.
SSL Benchmark
A benchmark for evaluating semi-supervised learning in open environments. It covers inconsistent data distributions, label spaces, and feature spaces, and reframes robust SSL evaluation with dynamic analysis tools and open-environment-oriented metrics.