Large Language Model Evaluation on MMLU, GSM8k, HellaSwag, WinoGrande
78.9Average ScoreFP16
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| FP16Format=FP162025.09 | 78.9 | — | — | — | — | — | |
| RTNFormat=INT82025.09 | 78.73 | — | — | — | — | 99.74 | |
| RTNFormat=FP82025.09 | 78.65 | — | — | — | — | 99.64 | |
| GPTQFormat=FP82025.09 | 78.58 | — | — | — | — | 99.55 | |
| GPTQFormat=INT82025.09 | 78.53 | — | — | — | — | 99.48 | |
| MR-GPTQFormat=NVINT42025.09 | 76.66 | — | — | — | — | 97.12 | |
| QAT+HadFormat=NVFP2025.09 | 76.1 | — | — | — | — | 96.5 | |
| RTN + HTFormat=NVINT42025.09 | 75.74 | — | — | — | — | 95.96 | |
| GPTQFormat=NVFP2025.09 | 75.6 | — | — | — | — | 95.7 | |
| GPTQ+HadFormat=NVFP2025.09 | 75.6 | — | — | — | — | 95.8 | |
| GPTQFormat=NVINT42025.09 | 75.32 | — | — | — | — | 95.42 | |
| QATFormat=NVFP2025.09 | 75.3 | — | — | — | — | 95.4 | |
| QAT+HadFormat=MXFP2025.09 | 75.3 | — | — | — | — | 95.4 | |
| RTNFormat=NVINT42025.09 | 75.13 | — | — | — | — | 95.18 | |
| GPTQ+Had128Format=NVFP2025.09 | 75.1 | — | — | — | — | 95.1 | |
| RTNFormat=NVFP2025.09 | 74.8 | — | — | — | — | 94.8 | |
| GPTQ+Had128Format=MXFP2025.09 | 74.2 | — | — | — | — | 94 | |
| RTN+Had128Format=NVFP2025.09 | 74.1 | — | — | — | — | 93.9 | |
| RTN+HadFormat=NVFP2025.09 | 74 | — | — | — | — | 93.8 | |
| GPTQ+HadFormat=MXFP2025.09 | 73.7 | — | — | — | — | 93.3 | |
| QATFormat=MXFP2025.09 | 72.9 | — | — | — | — | 92.3 | |
| MR-GPTQFormat=MXINT42025.09 | 72.51 | — | — | — | — | 91.86 | |
| RTN+Had128Format=MXFP2025.09 | 71.6 | — | — | — | — | 90.7 | |
| GPTQFormat=MXFP2025.09 | 70.78 | — | — | — | — | 89.66 | |
| RTN+HadFormat=MXFP2025.09 | 70.5 | — | — | — | — | 89.3 | |
| RTNFormat=MXFP2025.09 | 69.5 | — | — | — | — | 88.1 | |
| GPTQFormat=MXINT42025.09 | 68.91 | — | — | — | — | 87.3 | |
| RTN + HTFormat=MXINT42025.09 | 66.3 | — | — | — | — | 84 | |
| RTNFormat=MXINT42025.09 | 63.05 | — | — | — | — | 79.87 | |
| FP16Model=Llama-3.2-1B-Instruct, Format=FP162025.09 | 53.47 | 46.2 | 46.32 | 59.78 | 61.56 | — | |
| RTNModel=Llama-3.2-1B-Instruct, Format=INT8, Quantization=RTN2025.09 | 52.8 | 45.9 | 44.2 | 59.8 | 61.3 | 99.55 | |
| GPTQModel=Llama-3.2-1B-Instruct, Format=FP8, Quantization=GPTQ2025.09 | 52.63 | 45.8 | 45 | 59.1 | 60.6 | 99.22 | |
| GPTQModel=Llama-3.2-1B-Instruct, Format=INT8, Quantization=GPTQ2025.09 | 52.5 | 45.4 | 44.9 | 59.6 | 60.1 | 98.99 | |
| RTNModel=Llama-3.2-1B-Instruct, Format=FP8, Quantization=RTN2025.09 | 52.5 | 46.1 | 44.7 | 59.5 | 59.5 | 98.99 | |
| MR-GPTQModel=Llama-3.2-1B-Instruct, Format=NVINT4, Quantization=MR-GPTQ2025.09 | 46.74 | 36.69 | 33.36 | 57.95 | 58.96 | 87.42 | |
| GPTQ+Had16Model=Llama-3.2-1B-Instruct, Format=NVFP, Quantization=GPTQ+Had162025.09 | 46.7 | 38.99 | 32.98 | 56.66 | 58.17 | 87.35 | |
| QAT+Had16Model=Llama-3.2-1B-Instruct, Format=NVFP, Quantization=QAT+Had162025.09 | 46.57 | 32.72 | 37.6 | 57.53 | 58.41 | 87.09 | |
| QATModel=Llama-3.2-1B-Instruct, Format=NVFP, Quantization=QAT2025.09 | 46.05 | 27.85 | 38.51 | 57.52 | 60.3 | 86.12 | |
| GPTQModel=Llama-3.2-1B-Instruct, Format=NVFP, Quantization=GPTQ2025.09 | 45.82 | 37.79 | 29.8 | 55.48 | 60.22 | 85.71 | |
| GPTQ+Had128Model=Llama-3.2-1B-Instruct, Format=NVFP, Quantization=GPTQ+Had1282025.09 | 45.71 | 35.47 | 31.16 | 57.02 | 59.19 | 85.5 | |
| RTN+Had16Model=Llama-3.2-1B-Instruct, Format=NVINT4, Quantization=RTN+Had162025.09 | 45.56 | 33.41 | 32.52 | 57.12 | 59.19 | 85.21 | |
| RTN+Had128Model=Llama-3.2-1B-Instruct, Format=NVFP, Quantization=RTN+Had1282025.09 | 45.23 | 38.28 | 29.95 | 54.27 | 58.41 | 84.59 | |
| QAT+Had32Model=Llama-3.2-1B-Instruct, Format=MXFP, Quantization=QAT+Had322025.09 | 45.2 | 28.12 | 36.85 | 57.04 | 58.8 | 84.55 | |
| GPTQModel=Llama-3.2-1B-Instruct, Format=NVINT4, Quantization=GPTQ2025.09 | 45.01 | 37.15 | 27.6 | 55.94 | 59.35 | 84.19 | |
| RTNModel=Llama-3.2-1B-Instruct, Format=NVFP, Quantization=RTN2025.09 | 44.87 | 36.08 | 31.39 | 54.77 | 57.22 | 83.91 | |
| GPTQ+Had128Model=Llama-3.2-1B-Instruct, Format=MXFP, Quantization=GPTQ+Had1282025.09 | 44.28 | 35.68 | 28.13 | 54.6 | 58.72 | 82.83 | |
| RTNModel=Llama-3.2-1B-Instruct, Format=NVINT4, Quantization=RTN2025.09 | 43.65 | 37.33 | 26.08 | 52.62 | 58.56 | 81.64 | |
| RTN+Had16Model=Llama-3.2-1B-Instruct, Format=NVFP, Quantization=RTN+Had162025.09 | 43.28 | 32.8 | 25.02 | 56.24 | 59.04 | 80.94 | |
| RTN+Had128Model=Llama-3.2-1B-Instruct, Format=MXFP, Quantization=RTN+Had1282025.09 | 43.01 | 34.48 | 25.55 | 53.98 | 58.01 | 80.44 | |
| GPTQ+Had32Model=Llama-3.2-1B-Instruct, Format=MXFP, Quantization=GPTQ+Had322025.09 | 42.66 | 29.44 | 27.6 | 54.89 | 58.72 | 79.8 | |
| RTN+Had32Model=Llama-3.2-1B-Instruct, Format=MXFP, Quantization=RTN+Had322025.09 | 39.79 | 30.89 | 19.41 | 51.64 | 57.22 | 74.42 | |
| MR-GPTQModel=Llama-3.2-1B-Instruct, Format=MXINT4, Quantization=MR-GPTQ2025.09 | 38.83 | 21.81 | 23.12 | 54.96 | 55.41 | 72.62 | |
| GPTQModel=Llama-3.2-1B-Instruct, Format=MXFP, Quantization=GPTQ2025.09 | 36.6 | 26.84 | 13.5 | 49.29 | 56.75 | 68.45 | |
| QATModel=Llama-3.2-1B-Instruct, Format=MXFP, Quantization=QAT2025.09 | 36.44 | 15.6 | 20.32 | 53.34 | 56.51 | 68.16 | |
| RTNModel=Llama-3.2-1B-Instruct, Format=MXFP, Quantization=RTN2025.09 | 36.2 | 30.46 | 11.83 | 48.28 | 54.22 | 67.7 | |
| GPTQModel=Llama-3.2-1B-Instruct, Format=MXINT4, Quantization=GPTQ2025.09 | 35.65 | 23.42 | 13.27 | 50.02 | 55.88 | 66.67 | |
| RTNModel=Llama-3.2-1B-Instruct, Format=MXINT4, Quantization=RTN2025.09 | 31.74 | 21.85 | 4.55 | 45.07 | 55.49 | 59.37 | |
| RTN+Had32Model=Llama-3.2-1B-Instruct, Format=MXINT4, Quantization=RTN+Had322025.09 | 31.39 | 13.17 | 9.48 | 48.91 | 53.99 | 58.71 |