Textbook of AI

1980s–, Computer scientist

Also known as: Jianlin Su

Su Jianlin (苏剑林) is a Chinese computer scientist who in 2021 introduced RoPE (Rotary Position Embedding) in the paper RoFormer: Enhanced Transformer with Rotary Position Embedding (with Yu Lu, Shengfeng Pan, Bo Wen, and Yunfeng Liu at Zhuiyi Technology). RoPE encodes position by applying rotations in pairs of embedding dimensions, with rotation angle proportional to position. The mechanism naturally extrapolates to longer sequences than seen during training and has become the standard positional encoding in modern LLMs, LLaMA, PaLM, Mistral, GPT-NeoX, Falcon and many others use RoPE.

Su has been a prolific Chinese-language ML blogger ("kexue.fm") whose technical posts on attention, embeddings and Transformer architecture have shaped the practice of an entire generation of Chinese ML practitioners.