If you're interested in rigorously evaluating existing VLMs, check our evaluation codebase that bundles together 12 different battle-tested vision-and-language benchmarks through a clean, automated ...