Reading Comprehension on RACE (Generic Score)
91ScoreN-3-Super 120B-A12B-Base
Evaluation results
| Method | Links | |
|---|---|---|
| N-3-Super 120B-A12B-BaseShots=02026.04 | 91 | |
| Ling-flash base-2.0Shots=02026.04 | 90.1 | |
| GLM-4.5 Air-BaseShots=02026.04 | 89.5 | |
| DCLMPre-training Curation Strategy=Human Discovered, Model Parameters=3B, Training Token Budget=500B tokens2026.03 | 36.08 | |
| Nemotron-CC_ASIPre-training Curation Strategy=AI Discovered, Model Parameters=3B, Training Token Budget=500B tokens2026.03 | 35.63 | |
| Fineweb-EduPre-training Curation Strategy=Human Discovered, Model Parameters=3B, Training Token Budget=500B tokens2026.03 | 35.43 | |
| Nemotron-CCPre-training Curation Strategy=AI Discovered, Model Parameters=3B, Training Token Budget=500B tokens2026.03 | 35.04 | |
| Ultra-FinewebPre-training Curation Strategy=Human Discovered, Model Parameters=3B, Training Token Budget=500B tokens2026.03 | 34.28 | |
| Nemotron-CC_ASI+Pre-training Curation Strategy=AI Discovered, Model Parameters=3B, Training Token Budget=500B tokens2026.03 | 34.28 | |
| UniformBackbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=02026.04 | 27.85 | |
| RegMixBackbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=3.072 × 10^182026.04 | 27.85 | |
| ADAPT-BM25Backbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=≪ 1.0 × 10^142026.04 | 27.39 | |
| ADAPTBackbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=≪ 1.1 × 10^152026.04 | 26.6 | |
| LinUpperBackbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=02026.04 | 26.41 | |
| DoReMiBackbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=4.92 × 10^192026.04 | 26.41 |