Xinyang Geng
San Francisco, California, United States
Summary
🤩
RockstarXinyang Geng is a research scientist at Google DeepMind with 12 years of experience blending academic rigor and production ML engineering. He completed a PhD at UC Berkeley under Sergey Levine after early research roles in BAIR and a Google AI Residency, and has since held positions at OpenAI and Google. His work concentrates on infrastructure, data, and modeling for large-scale self-supervised learning, spanning language models, vision-language systems, AI for science, and reinforcement learning. He contributes to open-source ML tooling—e.g., implementing a GPT-J model and improving JAX/Flax LLM pipelines in the EasyLM repo—bringing practical systems thinking to research problems. Based in Berkeley, he combines deep theoretical training with hands-on experience shipping and debugging model checkpoints, tokenizers, and training objectives. Colleagues describe him as someone who bridges prototype research and reliable, scalable ML infrastructure.
12 years of coding experience
Github Skills (10)
deeplearning-ai10
nlp10
language-model10
deep-learning10
jax10
large-language-models10
python10
natural-language-processing10
flax10
transformer9
Programming languages (6)
Github contributions (5)
young-geng/EasyLM
Nov 2022 - Mar 2023
Large language models (LLMs) made easy, EasyLM is a one stop solution for pre-training, finetuning, evaluating and serving LLMs in JAX/Flax.
Role in this project:
ML Engineer
ML EngineerContributions:113 commits, 12 PRs, 187 pushes in 4 months
Contributions summary:Xinyang contributed to the implementation of a GPTJ model within the EasyLM framework, adding the necessary files for model definition. They updated the Readme and made several improvements to the jax_utils module. Additionally, the user made changes related to checkpointing and logging, and subsequently fixed LM objective and tokenizer issues.
fine-tuningjaxlarge-language-modelsdeep-learningflax
young-geng/CQL
Aug 2021 - Oct 2022
Conservative Q Learning on top of SAC
Contributions:26 commits, 20 pushes, 1 branch in 1 year 2 months
q-learningpytorchreinforcement-learning
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.