Medical Question Answering on MedQA (Acc, Token usage, Speed)
78.3AccuracyLatentMAS
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| LatentMASBackbone=Qwen3-14B, MAS Setting=Hierarchical2025.11 | 78.3 | 899 | 1,250 | — | |
| TextMASBackbone=Qwen3-14B, MAS Setting=Hierarchical2025.11 | 78 | 5,473 | 7,591 | — | |
| LatentMASBackbone=Qwen3-8B, MAS Setting=Hierarchical2025.11 | 77 | 1,007 | 964 | — | |
| TextMASBackbone=Qwen3-8B, MAS Setting=Hierarchical2025.11 | 76.3 | 6,893 | 3,387 | — | |
| LatentMASBackbone=Llama-3.1-8B-Instruct, MAS Setting=Hierarchical2025.11 | 67.7 | 191 | — | 204 | |
| LatentMASBackbone=Qwen3-4B, MAS Setting=Hierarchical2025.11 | 67.3 | 1,015 | 557 | — | |
| LatentMASBackbone=Llama-3.1-8B-Instruct, MAS Setting=Sequential2025.11 | 65.3 | 316 | — | 182 | |
| TextMASBackbone=Qwen3-4B, MAS Setting=Hierarchical2025.11 | 65 | 6,702 | 1,495 | — | |
| TextMASBackbone=Llama-3.1-8B-Instruct, MAS Setting=Sequential2025.11 | 65 | 2,354 | — | 589 | |
| BiRG-LoRASetting=Qwen, CMExam train2026.06 | 65 | — | — | — | |
| SingleBackbone=Qwen3-14B, MAS Setting=Hierarchical2025.11 | 64.7 | 1,746 | 1,360 | — | |
| TextMASBackbone=Llama-3.1-8B-Instruct, MAS Setting=Hierarchical2025.11 | 64.3 | 1,847 | — | 464 | |
| MoELoRASetting=Qwen, CMExam train2026.06 | 64.14 | — | — | — | |
| DoRASetting=Qwen, CMExam train2026.06 | 63.71 | — | — | — | |
| BiRG-LoRASetting=Llama, CMB train2026.06 | 62.71 | — | — | — | |
| DoRASetting=Llama, CMB train2026.06 | 61.86 | — | — | — | |
| MoELoRASetting=Llama, CMB train2026.06 | 61.57 | — | — | — | |
| SingleBackbone=Llama-3.1-8B-Instruct, MAS Setting=Sequential2025.11 | 59 | 382 | — | 112 | |
| SingleBackbone=Llama-3.1-8B-Instruct, MAS Setting=Hierarchical2025.11 | 59 | 382 | — | 112 | |
| SingleBackbone=Qwen3-8B, MAS Setting=Hierarchical2025.11 | 53 | 2,098 | 476 | — | |
| TextMASBackbone=Llama-3.2-3B-Instruct, MAS Setting=Hierarchical2025.11 | 50.3 | 1,729 | — | 282 | |
| LatentMASBackbone=Llama-3.2-3B-Instruct, MAS Setting=Hierarchical2025.11 | 50 | 359 | — | 73 | |
| TextMASBackbone=Llama-3.2-3B-Instruct, MAS Setting=Sequential2025.11 | 48 | 1,595 | — | 236 | |
| LatentMASBackbone=Llama-3.2-3B-Instruct, MAS Setting=Sequential2025.11 | 48 | 403 | — | 85 | |
| SingleBackbone=Qwen3-4B, MAS Setting=Hierarchical2025.11 | 47.7 | 2,134 | 236 | — | |
| SingleBackbone=Llama-3.2-3B-Instruct, MAS Setting=Sequential2025.11 | 46.7 | 405 | — | 51 | |
| SingleBackbone=Llama-3.2-3B-Instruct, MAS Setting=Hierarchical2025.11 | 46.7 | 405 | — | 51 |