Commonsense Reasoning on ARC-C, ARC-E, PIQA, StoryCloze
55.58Average AccuracyProjQ
Evaluation Results
| Method | Links | |
|---|---|---|
| ProjQModel Family=LLaMA2, Parameter Scale=13B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 55.58 | |
| ProjQModel Family=Qwen2.5-Ins, Parameter Scale=32B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 55.05 | |
| CALDERAModel Family=LLaMA2, Parameter Scale=13B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 54.02 | |
| LoftQModel Family=LLaMA2, Parameter Scale=13B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 53.82 | |
| LoftQModel Family=Qwen2.5-Ins, Parameter Scale=32B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 53.42 | |
| GPTQ+SVD-LLMModel Family=LLaMA2, Parameter Scale=13B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 52.97 | |
| GPTQ+SVD-LLMModel Family=Qwen2.5-Ins, Parameter Scale=32B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 49.77 | |
| ProjQModel Family=LLaMA2, Parameter Scale=7B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 47.82 | |
| CALDERAModel Family=Qwen2.5-Ins, Parameter Scale=14B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 47.78 | |
| ProjQModel Family=Qwen2.5-Ins, Parameter Scale=14B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 47.68 | |
| ProjQModel Family=Qwen3, Parameter Scale=32B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 47.23 | |
| CALDERAModel Family=LLaMA2, Parameter Scale=7B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 46.85 | |
| GPTQ+SVD-LLMModel Family=Qwen2.5-Ins, Parameter Scale=14B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 46.28 | |
| LoftQModel Family=Qwen2.5-Ins, Parameter Scale=14B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 46.1 | |
| GPTQ+SVD-LLMModel Family=LLaMA2, Parameter Scale=7B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 45.33 | |
| LoftQModel Family=LLaMA2, Parameter Scale=7B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 45.19 | |
| LoftQModel Family=Qwen3, Parameter Scale=32B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 44.88 | |
| ProjQModel Family=Qwen2.5-Ins, Parameter Scale=7B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 44.44 | |
| GPTQ+SVD-LLMModel Family=Qwen3, Parameter Scale=32B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 44.09 | |
| LoftQModel Family=Qwen2.5-Ins, Parameter Scale=7B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 43.9 | |
| CALDERAModel Family=Qwen2.5-Ins, Parameter Scale=7B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 42.5 | |
| ProjQModel Family=Qwen3, Parameter Scale=8B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 41.96 | |
| CALDERAModel Family=Qwen3, Parameter Scale=8B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 41.15 | |
| LoftQModel Family=Qwen3, Parameter Scale=8B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 40.98 | |
| GPTQ+SVD-LLMModel Family=Qwen2.5-Ins, Parameter Scale=7B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 40.96 | |
| GPTQ+SVD-LLMModel Family=Qwen3, Parameter Scale=8B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 39.55 | |
| ProjQModel Family=Qwen3, Parameter Scale=4B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 39.28 | |
| CALDERAModel Family=Qwen3, Parameter Scale=4B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 38.9 | |
| LoftQModel Family=Qwen3, Parameter Scale=4B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 38.81 | |
| GPTQ+SVD-LLMModel Family=Qwen3, Parameter Scale=4B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 38.38 | |
| AWQ+SVD-LLMModel Family=LLaMA2, Parameter Scale=13B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 37.82 | |
| AWQ+SVD-LLMModel Family=Qwen2.5-Ins, Parameter Scale=7B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 37.51 | |
| AWQ+SVD-LLMModel Family=Qwen2.5-Ins, Parameter Scale=14B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 36.67 | |
| AWQ+SVD-LLMModel Family=LLaMA2, Parameter Scale=7B, Quantization Bit=2-bit, Rank (ra = rd)=64, Zero-shot evaluation protocol=true2026.05 | 36.61 |