Inference Performance and Latency on GSM8K (Math Word Problem Solving)
81.97Accuracy (GSM8K)Vanilla (OpenPangu)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Vanilla (OpenPangu)Base Model=OpenPangu-Embedded-7B2026.05 | 81.97 | 100 | 0.273 | 1 | |
| SPprunerBase Model=OpenPangu-Embedded-7B, Token Reduction Rate=88.0%2026.05 | 80.61 | 98.3 | 0.217 | 1.26 | |
| SPprunerBase Model=OpenPangu-Embedded-7B, Token Reduction Rate=84.0%2026.05 | 80.53 | 98.2 | 0.217 | 1.26 | |
| SPprunerBase Model=OpenPangu-Embedded-7B, Token Reduction Rate=91.6%2026.05 | 76.29 | 93.1 | 0.204 | 1.34 | |
| SPprunerBase Model=OpenPangu-Embedded-7B, Token Reduction Rate=92.0%2026.05 | 75.68 | 92.3 | 0.203 | 1.34 |