Mostafa Elhoushi

RESEARCH ENGINEER | TORONTO, CANADA

I am a research engineer in FAIR Systems and Machine Learning (SysML) group. I focus on reserach that makes deep learning inference or training faster. Prior to joining FAIR, I was in the compilers team of Meta's Training and Inference Accelerator (MTIA) chip. I graduated from Ain Shams University, Egypt with my Bachelors and Masters, and obtained my PhD from Queen's University, Canada.

Twitter

GitHub

Google Scholar

Research Areas

Computer Vision

Natural Language Processing (NLP)

Systems Research

Mostafa's Work

LayerSkip

Speedup LLMs by drafting with earlier layers and verifying with remaining layers

Sieve

Filtering datasets for CLIP models

Memory Optimizations for Deep Learning

Reduce memory footprint of deep learning models by reordering ops and defragmenting tensors

MTIA

Meta’s first-generation AI inference accelerator

CompilerGym

DeepShift

Replacing multiplications with bitwise shift and sign flips

Mostafa's Publications

June 14, 2024

NLP

SYSTEMS RESEARCH

LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding

Mostafa Elhoushi, Akshat Shrivastava, Diana Liskovich, Basil Hosmer, Bram Wasti, Liangzhen Lai, Nas Mahmoud, Bilge Acun, Saurabh Agarwal, Ahmed Roman, Ahmed Aly, Beidi Chen, Carole-Jean Wu