Kenji Watanabe

Kenji Watanabe

Researcher — pre-training, scaling, evaluation

Member of Technical Staff, Pre-training at OpenAISan Francisco, CA · America/Los_Angeles14 yrs · IC6

kenji.watanabe@example.com+1 (415) 555-6516LinkedIn GitHub

ScreeningWarmSourcedPassiveUS citizen

AI match score

96/ 100

Strong

Last touched1w ago

OwnerMei Lin

Comp expectation$520,000–$720,000

Flight riskmedium

Overview Experience5 Education2 Skills19 Projects2 Activity12

Projects

2 portfolio entries.

Sparse Mixture-of-Experts scaling study
First author
Sep 2023 — Present
Co-led a study on routing stability in MoE transformers at >100B parameter scale; proposed a gradient-balanced routing variant.
publicationStable routing for sparse MoE at scale — arXiv preprintcommitFollow-up blog postcommitLightning talk at internal summit
Open evals library
Maintainer
Feb 2022 — Present
Maintained an eval harness used by the broader team to track capability gains.
commitEval harness — OSS contributionspublicationConference poster