Multiple-choice Question Answering on MMLU-Pro Zipf 0.8
76.7AccuracyMemBoost
Evaluation results
| Method | Links | |
|---|---|---|
| MemBoostMeta-Controller=Qwen3.5-2B2026.03 | 76.7 | |
| Qwen3-14B-FP8-dynamic (Oracle)Model Type=Baseline / Oracle2026.03 | 76.4 | |
| MemBoostMeta-Controller=Ministral-3-3B-Instruct-25122026.03 | 76.2 | |
| MemBoostMeta-Controller=Qwen3-4B-Instruct-2507-FP82026.03 | 76.1 | |
| Qwen3-4B-Instruct-2507-FP8Model Type=Baseline2026.03 | 74.5 | |
| Ministral-3-3B-Instruct-2512Model Type=Baseline2026.03 | 53.8 | |
| Qwen3.5-2BModel Type=Baseline2026.03 | 50 |