AF3Parallel: Memory-Profile-Driven Multi-GPU Scheduling for High-Throughput AlphaFold 3 Inference

A toolkit for running AlphaFold 3 inference at scale on multi-GPU clusters. The repository contains a profile-driven multi-GPU executor, two memory profilers (peak and time-series), GPU and CPU runtime estimators, an input-JSON manipulation utility, and a standalone GPU memory monitor.

The tools are designed to work together: profile your hardware once, use the resulting TSV to estimate batch runtime ahead of time, then run the batch in parallel across all available GPUs with VRAM-aware scheduling.