Commonsense Reasoning on HellaSwag, PIQA, WinoGrande
33.9HellaSwag AccuracySpikeLLM-7B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SpikeLLM-7BTime Step=42026.04 | 33.9 | 53.4 | 51.5 | 46.3 | |
| Qwen-1.5BEnergy Consumption (mJ)=3398.32026.04 | 26.5 | 53.7 | 52.8 | 44.3 | |
| Adaptive Spiking NeuronEnergy Consumption (mJ)=245.1, Time Step=4, Architecture=WD-SpikingFormer-0.4B2026.04 | 26 | 54.6 | 49.4 | 43.3 | |
| NILIF (WD-SpikingFormer-0.4B)Energy Consumption (mJ)=238.4, Time Step=42026.04 | 25.9 | 53.4 | 50.2 | 43.2 |