Text-based agent interaction on TextWorld Cooking
76AccuracyDirect-Action (RL)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Direct-Action (RL)Backbone=Qwen3-4B-Instruct, Independent runs=32026.05 | 76 | 31.5 | |
| Base ModelBackbone=Qwen3-4B-Instruct, Independent runs=32026.05 | 69 | 33.9 | |
| Agent-BRACEBackbone=Qwen3-4B-Instruct, Independent runs=32026.05 | 68.7 | 44.7 | |
| PABUBackbone=Qwen3-4B-Instruct, Independent runs=32026.05 | 32.5 | 72.1 | |
| ReActBackbone=Qwen3-4B-Instruct, Independent runs=32026.05 | 13.2 | 24.4 | |
| ReAct (RL)Backbone=Qwen3-4B-Instruct, Independent runs=32026.05 | 13 | 40.6 |