Language-to-Execution Grounding on LabGuard-Bench stress benchmark (Random 80/20 split)
100Tag F1Gold
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GoldGrounding Module=Gold2026.06 | 100 | 100 | — | 77.1 | 93.1 | 84.4 | |
| LabGuard-HybridGrounding Module=LabGuard-Hybrid2026.06 | 93.8 | 87.5 | — | 58.6 | 85.4 | 69.5 | |
| LoRA (Qwen3-8B)Grounding Module=LoRA, Backbone=Qwen3-8B2026.06 | 93.4 | 87.2 | — | 58.6 | 85.4 | 69.5 | |
| RegexGrounding Module=Regex2026.06 | 68.3 | 54.2 | — | 48.6 | 79.1 | 60.2 |