Meta-cognitive Assessment on Self-Knowledge Benchmark (test)
49.21Answerable AccuracyCDKC
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CDKCBackbone=Qwen2.5-7B2026.02 | 49.21 | 79.07 | 58.36 | 63.37 | 36.63 | 79.07 | 20.93 | 87.27 | 48.8 | 63.37 | 73.43 | 68.18 | |
| GRPOBackbone=Qwen2.5-7B2026.02 | 46.56 | 94.57 | 61.26 | 21.18 | 78.82 | 94.57 | 5.43 | 89.84 | 34.63 | 21.18 | 34.28 | 43.66 | |
| Qwen2.5-7B-InstructBackbone=Qwen2.5-7B2026.02 | 31.66 | 66.86 | 42.45 | 57.6 | 42.4 | 66.86 | 33.14 | 79.74 | 41.05 | 57.6 | 66.88 | 60.43 | |
| Know WhatBackbone=Qwen2.5-7B2026.02 | 29.1 | 98.16 | 50.25 | 2.05 | 97.95 | 98.16 | 1.84 | 71.64 | 30.68 | 2.05 | 3.99 | 31.49 | |
| BARRELBackbone=Qwen2.5-7B2026.02 | 27.77 | 76.45 | 42.68 | 24.78 | 75.22 | 76.45 | 23.55 | 70.44 | 30.98 | 24.78 | 36.66 | 40.61 | |
| CRew-DPOBackbone=Qwen2.5-7B2026.02 | 19.26 | 82.36 | 38.59 | 20.15 | 79.85 | 82.36 | 17.64 | 72.13 | 31.3 | 20.15 | 31.51 | 39.21 |