Case Study: ROAST achieves 4.5x faster AI image inference with Pruna AI

ROAST Deploys 4.5x Faster AI Models on Modal in 7 Hours with Pruna

ROAST, an AI-powered service for improving dating profiles, faced challenges with the speed and cost of its AI image generation feature. Their workflow required retraining models on user-uploaded photos and generating numerous images, which was computationally expensive on their Modal cloud and A100 hardware setup. They sought a solution from Pruna AI to accelerate inference and improve the user experience.

By integrating Pruna's optimization service, the vendor achieved a significant performance boost. ROAST deployed Pruna and saw inference speed increase by 4.5x, from 1.77 to 5.48 steps per second, while cold start times were reduced to around 4 seconds. This resulted in a 2.5x to 3x improvement for the entire inference pipeline, leading to substantial cost savings from reduced GPU uptime and a better user experience that helped drive revenue. The entire process from initial contact to a successful production test was completed in just seven hours.

ROAST

Benoit Baylin

Co-founder