Factual Correction on Suppressed Samples 134 samples
58.2CRCDS
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CDSModel=Qwen3.5-9B, Prep.=✓, Latency=1.002×, Peak GPU Mem.=16.75 GB2026.05 | 58.2 | 0.007 | 0.957 | |
| TAEModel=LLaMA3.1-8B, Prep.=✓, Latency=1.379×, Peak GPU Mem.=15.05 GB2026.05 | 55.9 | 0.228 | 0.702 | |
| CDSModel=LLaMA3.1-8B, Prep.=✓, Latency=1.000×, Peak GPU Mem.=15.05 GB2026.05 | 53.7 | 0.077 | 0.833 | |
| DoLaModel=LLaMA3.1-8B, Prep.=✗, Latency=1.951×, Peak GPU Mem.=41.64 GB2026.05 | 50.8 | 0.044 | 0.897 | |
| DPAModel=LLaMA3.1-8B, Prep.=✗, Latency=1.022×, Peak GPU Mem.=15.05 GB2026.05 | 50 | 0.09 | 0.829 | |
| OriginalModel=LLaMA3.1-8B, Prep.=✗, Latency=1.000×, Peak GPU Mem.=15.05 GB2026.05 | 39.6 | 0.037 | 0.911 | |
| InstructionModel=LLaMA3.1-8B, Prep.=✗, Latency=1.000×, Peak GPU Mem.=15.05 GB2026.05 | 38.1 | 0.016 | 0.944 | |
| DoLaModel=Qwen3.5-9B, Prep.=✗, Latency=2.262×, Peak GPU Mem.=38.80 GB2026.05 | 34.3 | 0.001 | 0.988 | |
| DPAModel=Qwen3.5-9B, Prep.=✗, Latency=1.009×, Peak GPU Mem.=16.75 GB2026.05 | 32.8 | 0.003 | 0.963 | |
| TAEModel=Qwen3.5-9B, Prep.=✓, Latency=1.410×, Peak GPU Mem.=16.75 GB2026.05 | 23.1 | 0.06 | 0.911 | |
| InstructionModel=Qwen3.5-9B, Prep.=✗, Latency=1.000×, Peak GPU Mem.=16.75 GB2026.05 | 20.5 | 0.004 | 0.966 | |
| ITIModel=LLaMA3.1-8B, Prep.=✓, Latency=1.132×, Peak GPU Mem.=15.05 GB2026.05 | 20.1 | 0.191 | 0.743 | |
| ITIModel=Qwen3.5-9B, Prep.=✓, Latency=1.018×, Peak GPU Mem.=16.75 GB2026.05 | 12.7 | 0.001 | 0.978 | |
| OriginalModel=Qwen3.5-9B, Prep.=✗, Latency=1.000×, Peak GPU Mem.=16.75 GB2026.05 | 0 | 0.002 | 0.977 |