Tag
This paper from Tsinghua University introduces Single-rollout Asynchronous Optimization (SAO) for reinforcement learning post-training of LLMs. SAO replaces batch-based GRPO with single-rollout asynchronous training to reduce idle GPU time and improve stability, and it was used to train the GLM-5.2 model (750B-A40B), achieving state-of-the-art results on agentic coding and reasoning benchmarks.
Tsinghua University NLP lab's graduate destination statistics show that PhDs in large model direction can earn an annual salary of over 6 million yuan, and master's degree holders over 1 million yuan, sparking criticism of self-media's distorted reporting.
Tsinghua University Language Processing Lab is recruiting postdocs, researchers, and interns to work on cutting-edge large model research and development. It offers ample computing power, data, funding, and competitive salaries, with a focus on research and open source.
SEAGym is a new evaluation environment for self-evolving LLM agents that measures agent harness updates across training, validation, test, replay, and cost records, providing complementary signals about the evolution process.
Tsinghua University has open-sourced OpenMAIC, a multi-agent AI interactive classroom platform. Users only need to input a topic or document to automatically generate a complete classroom featuring AI teacher explanations, AI classmate discussions, quizzes, and whiteboard drawing. It is fully open-source, free, and supports local deployment.
Nvidia CEO Jensen Huang is joining the board of Tsinghua University in Beijing.
Dolphin-CN-Dialect is a streaming-capable ASR model that improves dialect recognition through temperature-based sampling and redesigned tokenization, achieving competitive performance with a smaller model size.
Researchers from Tsinghua University have developed a new shortest-path algorithm with O(m log^{2/3} n) complexity, surpassing Dijkstra's algorithm, which had been considered theoretically optimal for 41 years.