Trajectory Unlearning on AlfWorld
100Unlearn EfficacyNL
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| NLBackbone LLM=GPT2026.04 | 100 | 100 | 85 | 83 | |
| NLBackbone LLM=Claude2026.04 | 100 | 100 | 71 | 68 | |
| NLBackbone LLM=Llama2026.04 | 100 | 100 | 94 | 93 | |
| CodeBackbone LLM=GPT2026.04 | 100 | 98 | 88 | 71 | |
| CodeBackbone LLM=Claude2026.04 | 100 | 91 | 68 | 60 | |
| CodeBackbone LLM=Llama2026.04 | 100 | 97 | 93 | 91 | |
| Exam.Backbone LLM=GPT2026.04 | 100 | 52 | 30 | 34 | |
| Exam.Backbone LLM=Claude2026.04 | 100 | 100 | 11 | 19 | |
| Exam.Backbone LLM=Llama2026.04 | 100 | 100 | 36 | 42 | |
| NLBackbone=GPT2026.04 | 1 | 100 | 91 | 88 | |
| NLBackbone=Claude2026.04 | 1 | 85 | 75 | 72 | |
| NLBackbone=Llama2026.04 | 1 | 97 | 90 | 87 | |
| CodeBackbone=GPT2026.04 | 1 | 85 | 90 | 88 | |
| CodeBackbone=Claude2026.04 | 1 | 78 | 72 | 70 | |
| CodeBackbone=Llama2026.04 | 1 | 90 | 88 | 81 | |
| Example-basedBackbone=GPT2026.04 | 1 | 60 | 25 | 16 | |
| Example-basedBackbone=Claude2026.04 | 1 | 34 | 9 | 10 | |
| Example-basedBackbone=Llama2026.04 | 1 | 50 | 42 | 40 |