Tsinghua University's Special Award winner YuXian Gu has joined DeepSeek to focus on large-scale model compression and efficiency optimization

According to Dongcha Beating monitoring, Tsinghua University Computer Science Ph.D. graduate and recipient of the 2025 Graduate Scholarship, YuXian Gu, has officially joined DeepSeek, and his name has appeared on the DeepSeek V4 paper's author list.

YuXian Gu's research mainly focuses on optimizing the efficiency of large-scale models in pre-training, model compression, and inference. His total citations on Google Scholar are close to 5000. YuXian Gu's notable works include the large-scale model knowledge distillation method MiniLLM (adopted by platforms such as Google, Alibaba, NVIDIA) and the hybrid architecture model Jet-Nemotron. Jet-Nemotron, when processing 256K ultra-long contexts on an H100 GPU, achieved a throughput 53.6 times faster than traditional full-attention models and outperformed larger-scale mixed expert models in multiple benchmark tests.