Multi-turn tool-use on AppWorld (test challenge)
27Task Goal Completion (TGC)SGCD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SGCDBackbone=Qwen3.5-4B, Decoding strategy=greedy pass@12026.06 | 27 | 8.5 | |
| GRPOBackbone=Qwen3.5-4B, Decoding strategy=greedy pass@12026.06 | 24.7 | 7.2 | |
| Base Qwen3.5-4BBackbone=Qwen3.5-4B, Decoding strategy=greedy pass@12026.06 | 11.3 | 2.2 |