Constrained Text Generation on 70 JSON schema benchmarks
79.3SaturationGPT2-Small-GCD
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GPT2-Small-GCDRegex strategy=non-padded (schema only)2026.02 | 79.3 | 98 | |
| GPT2-Large-GCDRegex strategy=non-padded (schema only)2026.02 | 77 | 98.6 | |
| GPT2-Medium-GCDRegex strategy=non-padded (schema only)2026.02 | 75.3 | 97.1 | |
| DIFFINITYGuidance scale=2.5, Regex strategy=padded (.*schema.*)2026.02 | 68.4 | 91.4 |