Exla

An SDK to run transformer models anywhere

Y CombinatorWinter 2025ActiveB2BEngineering, Product and DesignArtificial IntelligenceEdge Computing SemiconductorsComputer Vision

What they do

Problem

Running transformer models in production is often too slow and memory-intensive across diverse deployment environments.

Solution

Exla provides an SDK that aggressively quantizes transformer models to reduce memory usage and speed up inference with minimal code changes.

Who it's for

Developers and ML teams deploying LLMs/VLMs and other transformer-based models on-device, on servers, or across varied hardware.

Pranav Nair

Co-Founder

CTO at Exla. Previously an OS engineer at Apple leading sleep/hibernation for all Apple devices. B.S. Computer Science from Purdue.

Viraat Das

Founder

CEO @ Exla. Previously machine learning engineer @ Amazon.