Invariant Capability on Inv. Capability
96.5Accuracy HighQwen-2.5-32B +CD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen-2.5-32B +CDBackbone=Qwen-2.5-32B, Method=CD, alpha=12026.01 | 96.5 | 44.5 | |
| Qwen-2.5-32B +CDBackbone=Qwen-2.5-32B, Method=CD, alpha=02026.01 | 95 | 94.5 | |
| Qwen-2.5-14B +CDBackbone=Qwen-2.5-14B, Method=CD, alpha=02026.01 | 92.5 | 91 | |
| Qwen-2.5-7B +CDBackbone=Qwen-2.5-7B, Method=CD, alpha=02026.01 | 91 | 88 | |
| Qwen-2.5-14B +CDBackbone=Qwen-2.5-14B, Method=CD, alpha=12026.01 | 90 | 31.5 | |
| Qwen-2.5-7B +CDBackbone=Qwen-2.5-7B, Method=CD, alpha=12026.01 | 89.5 | 46.5 | |
| Olmo-3.1-32B +CDBackbone=Olmo-3.1-32B, Method=CD, alpha=02026.01 | 89 | 19 | |
| Olmo-3.1-32B +CDBackbone=Olmo-3.1-32B, Method=CD, alpha=12026.01 | 88 | 13 | |
| Llama-3.1-8B +CDBackbone=Llama-3.1-8B, Method=CD, alpha=02026.01 | 83.5 | 43 | |
| Qwen-2.5-7B +CDBackbone=Qwen-2.5-7B, Method=CD, alpha=22026.01 | 83 | 30 | |
| Qwen-2.5-32B +CDBackbone=Qwen-2.5-32B, Method=CD, alpha=22026.01 | 83 | 21.5 | |
| Olmo-3.1-32B +CDBackbone=Olmo-3.1-32B, Method=CD, alpha=22026.01 | 81.5 | 10.5 | |
| Qwen-2.5-14B +CDBackbone=Qwen-2.5-14B, Method=CD, alpha=22026.01 | 75.5 | 11 | |
| Llama-3.1-8B +CDBackbone=Llama-3.1-8B, Method=CD, alpha=12026.01 | 75.5 | 12.5 | |
| Llama-3.1-8B +CDBackbone=Llama-3.1-8B, Method=CD, alpha=22026.01 | 65.5 | 12.5 |