Boolean Question Answering on BoolQ (acc_norm)
88.7Acc (Normalized)Full model
Evaluation results
| Method | Links | |
|---|---|---|
| Full modelSparsity=0%, Backbone=Qwen3-30B-A3B, Evaluation Protocol=Zero-shot2026.05 | 88.7 | |
| RCOSparsity=25%, Backbone=Qwen3-30B-A3B, Evaluation Protocol=Zero-shot2026.05 | 88.5 | |
| EvoESAPSparsity=25%, Backbone=Qwen3-30B-A3B, Evaluation Protocol=Zero-shot2026.05 | 86.6 | |
| BaseModel=Llama-70B2026.04 | 85.3 | |
| BaseModel=Mistral-7B2026.04 | 84.3 | |
| BaseModel=Qwen2.5-7B2026.04 | 84.3 | |
| RCOSparsity=50%, Backbone=Qwen3-30B-A3B, Evaluation Protocol=Zero-shot2026.05 | 83.5 | |
| BaseModel=Llama-13B2026.04 | 82.5 | |
| EvoESAPSparsity=50%, Backbone=Qwen3-30B-A3B, Evaluation Protocol=Zero-shot2026.05 | 80.5 | |
| BaseModel=GPT-2 (774M)2026.04 | 60.5 | |
| GAINModel=Qwen2.5-7B2026.04 | -0.8 | |
| GAINModel=Mistral-7B2026.04 | -1.2 | |
| GAINModel=Llama-13B2026.04 | -1.4 | |
| GAINModel=GPT-2 (774M)2026.04 | -2.1 | |
| LoRAModel=Llama-13B2026.04 | -2.3 | |
| GAINModel=Llama-70B2026.04 | -2.5 | |
| LoRAModel=GPT-2 (774M)2026.04 | -7.8 | |
| LoRAModel=Mistral-7B2026.04 | -9.2 | |
| LoRAModel=Llama-70B2026.04 | -14.6 | |
| LoRAModel=Qwen2.5-7B2026.04 | -14.9 |