Sachin Mehta

Career

Mehta received a PhD from the University of Washington, supervised by Linda Shapiro and Hannaneh Hajishirzi, and defended the thesis in March 2021. Mehta's site records joining Apple as a Research Scientist in April 2021 and returning to the University of Washington as an Affiliate Assistant Professor in October 2021; the university's Electrical & Computer Engineering department lists Mehta as an Affiliate Assistant Professor. Mehta is listed with a FAIR at Meta affiliation on the 2026 paper "Compute Optimal Tokenization".¹²³

Research

Mehta's research focuses on efficient machine learning architectures that can run on devices such as smartphones. This work includes ESPNetv2 (2018), a light-weight convolutional network, DeLighT (2020), a deep and light-weight transformer, and MobileViT (2021), a mobile-friendly vision transformer. In 2021 Mehta released CVNets, a library for training computer vision networks.¹⁷⁸⁶

In 2024 Mehta was first author of the OpenELM paper, which released an efficient open language model family together with training and inference code through Apple's CoreNet repository.⁴⁵

Work and contributions

• OpenELM Open weights

  First author of the OpenELM paper, which describes a layer-wise scaling strategy for allocating parameters within each transformer layer and releases the models with an open training and inference framework.⁴⁵

• OpenELM 3B Instruct Open weights

  First-listed author in the OpenELM citation of the CoreNet project, which provides the instruction-tuning recipe and pretrained checkpoints for the OpenELM models.⁵

• MobileViT Published research

  First author, with Mohammad Rastegari, of a paper on a light-weight vision transformer designed for mobile devices.⁶

Timeline

1. Nov 2018Posted ESPNetv2, a light-weight general-purpose convolutional network, as first author.⁷
2. Aug 2020Posted DeLighT, a deep and light-weight transformer, as first author.⁸
3. Apr 19, 2021Joined Apple as a Research Scientist.¹
4. Oct 5, 2021Posted MobileViT and became an Affiliate Assistant Professor at the University of Washington.¹⁶
5. Oct 22, 2021Released CVNets, a library for training computer vision networks.¹
6. Apr 2024Posted the OpenELM paper on an efficient open language model family as first author.⁴
7. May 2026Co-authored "Compute Optimal Tokenization", listed with a FAIR at Meta affiliation.³

Other documented roles

Roles the sources document that are not confirmed as current.

• Research Scientist (joined April 2021, per personal site), Apple¹
• Listed with a FAIR at Meta affiliation on the Compute Optimal Tokenization paper (arXiv, May 2026), Meta³

Where they work

• University of Washington 19 people in the wiki

Related people

• Xiaodong HeAlso at University of Washington
• Ali FarhadiAlso at University of Washington
• Noah A. SmithAlso at University of Washington
• Aylin CaliskanAlso at University of Washington
• Emily M. BenderAlso at University of Washington
• Karén SimonyanComputer vision · Language models
• Yuxin WuComputer vision · Language models
• Aravind SrinivasLanguage models · Computer vision
• Stefano SoattoLanguage models · Computer vision
• Luke Melas-KyriaziLanguage models · Computer vision