Question Answering on MMLU (EM and F1 Scores)
85EM ScoreAgent
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AgentBackbone=Qwen3-8B, Budget Strategy=Per-query Budget, B = 20, Avg. Cost=2.90, Avg. Psg=1.14, Avg. Steps=2.142026.06 | 85 | 85 | |
| AgentBackbone=Qwen3-8B, Budget Strategy=Shared Budget, Avg. Cost=2.56, Avg. Psg=0.86, Avg. Steps=1.862026.06 | 85 | 85 | |
| AgentBackbone=Llama3.1-8B-Instruct, Budget Strategy=Per-query Budget, B = 20, Avg. Cost=15.74, Avg. Psg=7.02, Avg. Steps=8.112026.06 | 72 | 72 | |
| AgentBackbone=Llama3.1-8B-Instruct, Budget Strategy=Shared Budget, Avg. Cost=16.04, Avg. Psg=6.28, Avg. Steps=7.352026.06 | 72 | 72 |