Making WAF ML models go brrr: saving decades of processing time

In this post, we discussed how we optimized the performance of Cloudflare's Web Application Firewall (WAF) Machine Learning (ML) models by employing various techniques such as pre-processing optimization, model inference optimization, and caching. We achieved a significant reduction in WAF ML execution time, cutting it down from 1519 microseconds to 275 microseconds on average, which is approximately 81.90% faster. This optimization has allowed us to handle more traffic with the same resources, improving our overall system performance and scalability.