Mathematical Reasoning on MATH-500 (mean peak GPU memory (MB))
15,508Mean Peak GPU Memory (MB)thinKV
Evaluation Results
| Method | Links | |
|---|---|---|
| thinKVKV cache budget=5122026.06 | 15,508 | |
| lag-kvKV cache budget=5122026.06 | 15,524 | |
| EPIKVKV cache budget=5122026.06 | 15,525 | |
| band-adaptiveKV cache budget=5122026.06 | 15,525 | |
| kv-keyKV cache budget=5122026.06 | 15,525 | |
| hybrid-seg-hsKV cache budget=5122026.06 | 15,525 | |
| h2oKV cache budget=5122026.06 | 15,560 | |
| raasKV cache budget=5122026.06 | 15,560 | |
| attn×hsKV cache budget=5122026.06 | 15,568 | |
| h2oKV cache budget=10242026.06 | 15,574 | |
| thinKVKV cache budget=10242026.06 | 15,643 | |
| hybrid-seg-hsKV cache budget=10242026.06 | 15,655 | |
| EPIKVKV cache budget=10242026.06 | 15,692 | |
| kv-keyKV cache budget=10242026.06 | 15,697 | |
| lag-kvKV cache budget=10242026.06 | 15,697 | |
| band-adaptiveKV cache budget=10242026.06 | 15,698 | |
| h2oKV cache budget=20482026.06 | 15,714 | |
| raasKV cache budget=10242026.06 | 15,743 | |
| attn×hsKV cache budget=10242026.06 | 15,747 | |
| h2oKV cache budget=40962026.06 | 15,814 | |
| thinKVKV cache budget=20482026.06 | 15,836 | |
| hybrid-seg-hsKV cache budget=20482026.06 | 15,857 | |
| EPIKVKV cache budget=20482026.06 | 15,880 | |
| band-adaptiveKV cache budget=20482026.06 | 15,880 | |
| kv-keyKV cache budget=20482026.06 | 15,880 | |
| lag-kvKV cache budget=20482026.06 | 15,880 | |
| raasKV cache budget=20482026.06 | 15,933 | |
| attn×hsKV cache budget=20482026.06 | 15,968 | |
| thinKVKV cache budget=40962026.06 | 16,036 | |
| EPIKVKV cache budget=40962026.06 | 16,063 | |
| band-adaptiveKV cache budget=40962026.06 | 16,064 | |
| kv-keyKV cache budget=40962026.06 | 16,064 | |
| lag-kvKV cache budget=40962026.06 | 16,064 | |
| hybrid-seg-hsKV cache budget=40962026.06 | 16,065 | |
| raasKV cache budget=40962026.06 | 16,136 | |
| attn×hsKV cache budget=40962026.06 | 16,201 | |
| noneKV cache budget=none2026.06 | 16,221 |