Benchmarks Mean Business

The basic job of an eval is let you judge how good your model is on a task. If enough people use the same eval we can use it to benchmark the relative performance of multiple models on a level playing field. All good, no drama. But building good benchmarks is hard! ImageNet was a […]