Skip to content

Dataset · DatologyAI

DatBench: Benchmark for VLM Evaluation

DatologyAI/DatBench

Curated set for fair and efficient evaluation of vision-language models.

What's inside

DatBench is an evaluation set for vision-language models, gathered as an alternative to benchmarks suffering from multiple-choice bloat, linguistic shortcuts, and annotation noise. The card specifies images and text; the volume is between 10k and 100k examples. The exact breakdown by subtask is not disclosed in the available snippet.

What you can do with it

  • Evaluating VLM robustness to linguistic shortcuts
  • Comparing vision-language models against each other
  • Validating multimodal pipelines before release
Size
10K–100K rows
Downloads this month
2K
Likes
108

How to load it

from datasets import load_dataset

ds = load_dataset("DatologyAI/DatBench")