I recently released an inference API company (scalattice. com), but the process building it required a huge amount of benchmarking, so we built some internal tools that we felt gave accurate and reliable standards to work off.
Source: [Hacker News](https://scalattice.com/blog/modelmeter/)