Physical Reasoning on PIQA (Accuracy, Accuracy Drop)
82.21AccuracyBF16
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BF16Model=Mistral-7B, A-W Quant Type=BF162026.02 | 82.21 | — | |
| HiF4+HiGPTQModel=Mistral-7B, A-W Quant Type=HiF4+HiGPTQ2026.02 | 81.91 | -0.3 | |
| BF16Model=Qwen2.5-14B, A-W Quant Type=BF162026.02 | 81.88 | — | |
| NVFP4Model=Qwen2.5-14B, A-W Quant Type=NVFP42026.02 | 81.72 | -0.16 | |
| HiF4+HiGPTQModel=Qwen2.5-14B, A-W Quant Type=HiF4+HiGPTQ2026.02 | 81.67 | -0.21 | |
| NVFP4+PTSModel=Qwen2.5-14B, A-W Quant Type=NVFP4+PTS2026.02 | 81.66 | -0.22 | |
| HiF4Model=Qwen2.5-14B, A-W Quant Type=HiF42026.02 | 81.29 | -0.59 | |
| HiF4Model=Mistral-7B, A-W Quant Type=HiF42026.02 | 81.01 | -1.2 | |
| BF16Model=Llama3-8B, A-W Quant Type=BF162026.02 | 80.85 | — | |
| NVFP4+PTSModel=Mistral-7B, A-W Quant Type=NVFP4+PTS2026.02 | 80.85 | -1.36 | |
| NVFP4+PTSModel=Llama3-8B, A-W Quant Type=NVFP4+PTS2026.02 | 79.76 | -1.09 | |
| HiF4+HiGPTQModel=Llama3-8B, A-W Quant Type=HiF4+HiGPTQ2026.02 | 79.6 | -1.25 | |
| BF16Model=Llama2-7B, A-W Quant Type=BF162026.02 | 79.11 | — | |
| HiF4Model=Llama3-8B, A-W Quant Type=HiF42026.02 | 79.09 | -1.76 | |
| NVFP4Model=Llama3-8B, A-W Quant Type=NVFP42026.02 | 78.56 | -2.29 | |
| HiF4Model=Llama2-7B, A-W Quant Type=HiF42026.02 | 78.4 | -0.71 | |
| NVFP4+PTSModel=Llama2-7B, A-W Quant Type=NVFP4+PTS2026.02 | 77.91 | -1.2 | |
| HiF4+HiGPTQModel=Llama2-7B, A-W Quant Type=HiF4+HiGPTQ2026.02 | 77.86 | -1.25 | |
| NVFP4Model=Llama2-7B, A-W Quant Type=NVFP42026.02 | 77.37 | -1.74 | |
| NVFP4Model=Mistral-7B, A-W Quant Type=NVFP42026.02 | 48.69 | -33.52 |