About

Hi! I’m Rohan. I’m an engineer currently at Meta AI working on large-scale training. Previously, I graduated with a BS in Computer Science from UCLA. I grew up in San Jose, California.

Work

I currently work at Meta AI on all things related to large scale training. Much of my work is open-source as part of the PyTorch framework. You can find more about my research interests at this link and some publications at this link

I worked nearly full-time on the PyTorch Distributed framework from 2019-2023, and continue to serve as a maintainer for the framework.

Previously, I’ve worked on building mortgage lending products at Blend, risk models at Affirm, and messaging applications for sports teams at Hudl.

Writing

This blog is mostly a place for me to collect random notes and tips that I’ve found interesting and useful over the years. Below is a reverse-chronological set of links to some things that I’ve written or helped write over the years!

2024

TorchTune library release

2023

PyTorch FSDP paper at VLDB 2023 PyTorch RPC paper at MLSys 2023 Mitigating stragglers in large-scale jobs with PyTorch DDP

2022

A tutorial on using PyTorch FSDP for large-scale deep learning training

Scaling foundation multimodal models using PyTorch FSDP

PyTorch Data Parallel on GCP

Scaling to 1 trillion parameters with PyTorch FSDP

2020

How to use PyTorch’s RPC Framework

2018

My Internship at Affirm: Crafting a Reliable Metrics and Alerting Framework

2017

Welcome to AI at UCLA

Getting Started in Machine Learning

Finance

I have a couple of writings around personal finance here that I don’t actively maintain anymore.

Misc

This section is still in progress!

Contact me

rvarm1 @ gmail dot com