nutrientdocs/form-field-vlm-v2-benchmark

form-field-vlm-v2-benchmark A balanced held-out slice for evaluating form-field detect + extract — detection + fine type + text label + entered value — across the three render conditions a real pipeline meets: empty, filled (printed), and handwritten. Drawn from the commonforms-synth-v2 split (document-disjoint from training). 📊 Benchmark: nutrientdocs/form-field-vlm-v2-benchmark 🎯 Model: nutrientdocs/form-field-vlm-v2 · 🏆 Leaderboard Composition 354 pages —… See the full description on the dataset page: https://huggingface.co/datasets/nutrientdocs/form-field-vlm-v2-benchmark.