Code Generation on HumanEval+ (Accuracy)
81.5AccuracyDiffMAS
Evaluation Results
| Method | Links | |
|---|---|---|
| DiffMASModel=Qwen3-8B, Communication Category=Trained Latent Communication2026.04 | 81.5 | |
| TextMASModel=Qwen3-8B, Communication Category=Text Communication2026.04 | 80.7 | |
| LatentMASModel=Qwen3-8B, Communication Category=Training-free Latent Communication2026.04 | 80.7 | |
| DiffMASModel=Qwen3-4B, Communication Category=Trained Latent Communication2026.04 | 79.8 | |
| LatentMASModel=Qwen3-4B, Communication Category=Training-free Latent Communication2026.04 | 78.9 | |
| TextMASModel=Qwen3-4B, Communication Category=Text Communication2026.04 | 76.3 | |
| SingleModel=Qwen3-4B, Communication Category=Text Communication2026.04 | 74.5 | |
| SingleModel=Qwen3-8B, Communication Category=Text Communication2026.04 | 74.5 | |
| C2CModel=Qwen3-8B, Communication Category=Trained Latent Communication2026.04 | 58.8 | |
| C2CModel=Qwen3-4B, Communication Category=Trained Latent Communication2026.04 | 33.3 |