Multiple-choice Question Answering on MMLU-Pro Zipf 1.1
81.8AccuracyMemBoost
Evaluation Results
| Method | Links | |
|---|---|---|
| MemBoostMeta-Controller=Qwen3.5-2B2026.03 | 81.8 | |
| Qwen3-14B-FP8-dynamic (Oracle)Model Type=Baseline / Oracle2026.03 | 79.9 | |
| MemBoostMeta-Controller=Qwen3-4B-Instruct-2507-FP82026.03 | 79.8 | |
| MemBoostMeta-Controller=Ministral-3-3B-Instruct-25122026.03 | 79.7 | |
| Qwen3-4B-Instruct-2507-FP8Model Type=Baseline2026.03 | 75.6 | |
| Ministral-3-3B-Instruct-2512Model Type=Baseline2026.03 | 46.4 | |
| Qwen3.5-2BModel Type=Baseline2026.03 | 43.5 |