Question Answering on MedQA (Acc, Token, Speed)
80.7AccuracyLatentMAS
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LatentMASBackbone=Qwen3-14B, Collaboration Method=Latent-based MAS, MAS Setting=Sequential2025.11 | 80.7 | 1,841 | 1,420 | |
| TextMASBackbone=Qwen3-14B, Collaboration Method=Text-based MAS, MAS Setting=Sequential2025.11 | 80.3 | 3,444 | 4,142 | |
| LatentMASBackbone=Qwen3-8B, Collaboration Method=Latent-based MAS, MAS Setting=Sequential2025.11 | 75.3 | 1,555 | 928 | |
| TextMASBackbone=Qwen3-8B, Collaboration Method=Text-based MAS, MAS Setting=Sequential2025.11 | 75 | 4,260 | 1,923 | |
| LatentMASBackbone=Qwen3-4B, Collaboration Method=Latent-based MAS, MAS Setting=Sequential2025.11 | 66.3 | 1,685 | 438 | |
| TextMASBackbone=Qwen3-4B, Collaboration Method=Text-based MAS, MAS Setting=Sequential2025.11 | 65.3 | 3,962 | 1,267 | |
| SingleBackbone=Qwen3-14B, Collaboration Method=Single-model, MAS Setting=Sequential2025.11 | 64.7 | 1,746 | 1,360 | |
| SingleBackbone=Qwen3-8B, Collaboration Method=Single-model, MAS Setting=Sequential2025.11 | 53 | 2,098 | 476 | |
| SingleBackbone=Qwen3-4B, Collaboration Method=Single-model, MAS Setting=Sequential2025.11 | 47.7 | 2,134 | 236 |