flageval-baai

Integrate new evaluation datasets into FlagEvalMM as benchmark tasks. Use when adding a dataset from HuggingFace or other sources to FlagEvalMM, creating task configs, writing data processors, building custom evaluators, setting up prompt templates, or running evaluation benchmarks on VLMs. Trigger on: "add dataset to FlagEvalMM", "create a new task", "integrate benchmark", "evaluate model on [dataset]", "write process.py", "write evaluator", or any request involving the tasks/ directory of FlagEvalMM.