Mathematical Reasoning on GSM8K (100-question subset, test)
87AccuracyDALC-Id
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DALC-IdModel parameters=14B, n=1002026.04 | 87 | 1,181 | |
| DALC-SVDModel parameters=14B, n=1002026.04 | 86 | 1,200 | |
| SC (k=5)Model parameters=14B, n=1002026.04 | 84 | 1,589 | |
| DALC-GSModel parameters=14B, n=1002026.04 | 83 | 1,188 | |
| DALC-GSModel parameters=7B, n=1002026.04 | 83 | 1,213 | |
| SnapKV-DBase Model=R1-Distill-Qwen14B, Cache budget=2562025.12 | 82 | — | |
| SeerAttentionBase Model=R1-Distill-Qwen14B, Cache budget=3842025.12 | 82 | — | |
| SingleModel parameters=14B, n=1002026.04 | 82 | 319 | |
| SC (k=5)Model parameters=7B, n=1002026.04 | 82 | 1,648 | |
| DALC-SVDModel parameters=7B, n=1002026.04 | 82 | 1,211 | |
| SnapKV-DBase Model=R1-Distill-Qwen14B, Cache budget=3842025.12 | 81 | — | |
| SnapKV-DBase Model=R1-Distill-Qwen14B, Cache budget=1282025.12 | 80 | — | |
| SeerAttentionBase Model=R1-Distill-Qwen14B, Cache budget=2562025.12 | 80 | — | |
| SnapKV-DBase Model=R1-Distill-Qwen14B, Cache budget=5122025.12 | 78 | — | |
| SingleModel parameters=7B, n=1002026.04 | 77 | 328 | |
| SeerAttentionBase Model=R1-Distill-Qwen14B, Cache budget=5122025.12 | 70 | — | |
| SeerAttentionBase Model=R1-Distill-Qwen14B, Cache budget=1282025.12 | 66 | — | |
| H2OBase Model=R1-Distill-Qwen14B, Cache budget=5122025.12 | 64 | — | |
| H2OBase Model=R1-Distill-Qwen14B, Cache budget=3842025.12 | 62 | — | |
| H2OBase Model=R1-Distill-Qwen14B, Cache budget=2562025.12 | 56 | — | |
| H2OBase Model=R1-Distill-Qwen14B, Cache budget=1282025.12 | 33 | — |