NVBench is a large dataset for complex and cross-domain NL2VIS tasks, which covers 105 domains, supports seven common types of visualizations, and contains 25,750 (NL, VIS) pairs.