Error Step Localization on TrajBench In-Distribution (test)
53.75Joint Exact Match (JEM)TrajAD
Evaluation Results
| Method | Links | |
|---|---|---|
| TrajADParams=4B, Method=LoRA finetune2026.02 | 53.75 | |
| Gemma-3-4B-InstructParams=4B, Method=Zero-shot2026.02 | 9.07 | |
| Qwen3-8BParams=8B, Method=Zero-shot2026.02 | 5.81 | |
| Qwen3-4BParams=4B, Method=Zero-shot2026.02 | 5.54 | |
| Phi-3-MiniParams=4B, Method=Zero-shot2026.02 | 3.28 |