Mathematical Reasoning on GSM8K (EM, Avg.)
64.22Exact Match (EM)EPnG
Evaluation results
| Method | Links | ||
|---|---|---|---|
| EPnGBackbone=Qwen1.5-MoE, Params. (%)=0.552026.07 | 64.22 | 58.98 | |
| ESFTBackbone=Qwen1.5-MoE, Params. (%)=15.912026.07 | 64.14 | 58.98 | |
| LoRA (static)Backbone=Qwen1.5-MoE, Params. (%)=0.552026.07 | 62.93 | 57.6 |