full-stack secure inference benchmark
Benchmark suite for evaluating encrypted ML inference stacks.
This project provides an end-to-end benchmark framework for secure inference workloads. The suite captures algorithm-level, compiler-level, and hardware-level metrics with a consistent methodology.
Highlights
- Unified benchmark traces for CKKS workloads
- Reproducible scripts for architecture-level profiling
- Practical comparison baseline for accelerator proposals
This benchmark helps answer a practical question: which optimization opportunities actually translate to meaningful system-level gains in secure ML pipelines?