On July 6, it was announced that Gu Yuxian, a PhD graduate from the Department of Computer Science at Tsinghua University and recipient of the 2025 Graduate Special Award Scholarship, has officially joined DeepSeek, with his name now listed among the authors of the DeepSeek V4 paper. Gu Yuxian's research primarily focuses on optimizing the efficiency of large models during pre-training, model compression, and inference stages, with nearly 5,000 citations on Google Scholar. His notable works include the large model knowledge distillation method MiniLLM (adopted by platforms such as Google, Alibaba, and NVIDIA) and the hybrid architecture model Jet-Nemotron. Jet-Nemotron achieves a throughput acceleration of 53.6 times compared to traditional full attention models when processing 256K ultra-long contexts on H100 GPUs, outperforming larger parameter hybrid expert models in multiple benchmark tests.
All Comments