About

Tao Ge is a Research Scientist at Apodex, training Apodex foundation models. Prior to Apodex, Tao was a research scientist at Microsoft and Tencent.

Tao pioneered Speculative Decoding and persona-driven synthetic data creation – both are now widely used in LLM inference and training:

  • Speculative Decoding. Tao was the first (in 2022) to propose the method and name it Speculative Decoding for accelerating Transformer inference. It is now widely used in LLM inference.

  • Persona-driven synthetic data creation. Tao was the first to propose using personas to generate synthetic data at scale (June 2024 paper). The method is widely used to train frontier LLMs, including OLMo, Nemotron, SmolLM, and many others.

Publications (*: equal contributions; ✉: corresponding author)

Tech Report

Peer-reviewed