Lifelong Learning for LLM Agents on LifelongAgentBench
82DB AccuracyLifeSkill
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| LifeSkillExperience=0, Training Category=Training-Based, Backbone=Llama-3.1-8B-Instruct2026.06 | 82 | 64 | 32 | 59 | |
| RFTExperience=0, Training Category=Training-Based, Backbone=Llama-3.1-8B-Instruct2026.06 | 70 | 51 | 35 | 52 | |
| SkillRLExperience=4, Training Category=Training-Based, Backbone=Llama-3.1-8B-Instruct2026.06 | 65 | 51 | 34 | 50 | |
| DAPOExperience=0, Training Category=Training-Based, Backbone=Llama-3.1-8B-Instruct2026.06 | 64 | 53 | 28 | 48 | |
| Group Self-ConsistencyExperience=4, Training Category=Training-Free, Backbone=Llama-3.1-8B-Instruct2026.06 | 63 | 49 | 36 | 49 | |
| Self-RefineExperience=0, Training Category=Training-Free, Backbone=Llama-3.1-8B-Instruct2026.06 | 25 | 52 | 31 | 36 | |
| Reasoning BankExperience=4, Training Category=Training-Free, Backbone=Llama-3.1-8B-Instruct2026.06 | 25 | 45 | 30 | 33 | |
| AWMExperience=4, Training Category=Training-Free, Backbone=Llama-3.1-8B-Instruct2026.06 | 20 | 56 | 33 | 36 | |
| VanillaExperience=0, Training Category=Training-Free, Backbone=Llama-3.1-8B-Instruct2026.06 | 16 | 44 | 27 | 29 |