We've published the comprehensive set of tools and scripts for evaluating VLM models and benchmarks in the UniBench repo

→ View original post on X — @aiatmeta,