Zero-shot LLM Evaluation Suite (AE, AC, SciQ, MM, MM-P, HS, OBQA, PIQA, RACE, WG, CSQA, AGI)
67.97AE ScoreAMO
Evaluation Results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AMOBackbone=Llama3.1-1.4B, Pre-training token budget=2x Chinchilla optimal ratio2026.05 | 67.97 | 32.13 | 86.4 | 24.66 | 6.84 | 39.1 | 24.6 | 70.95 | 34 | 53.59 | 21.7 | 17.72 | 39.97 | |
| Muon-KJBackbone=Llama3.1-1.4B, Pre-training token budget=2x Chinchilla optimal ratio2026.05 | 65.32 | 30.41 | 85.2 | 25.47 | 6.94 | 37.32 | 24.6 | 69.36 | 32.15 | 53.67 | 19.99 | 17.4 | 38.99 |