Language Understanding on MMLU (0-shot)
70.46AccuracyToken Filtering
Evaluation results
| Method | Links | |||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Token FilteringPruning Ratio=20%, Model=Phi-4-14B, Zero-shot=true2025.12 | 70.46 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 63.4 | 63.34 | 82.09 | 76.86 | |
| NAMEX - FullRouting Strategy=Stable - MoE, Evaluation Protocol=Zero-shot2025.10 | 61.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NAMEX - FullRouting Strategy=Linear, Evaluation Protocol=Zero-shot2025.10 | 61.87 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NAMEX - FullRouting Strategy=Cosine, Evaluation Protocol=Zero-shot2025.10 | 61.85 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EP - CAMEXRouting Strategy=Stable - MoE, Evaluation Protocol=Zero-shot2025.10 | 61.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EP - CAMEXRouting Strategy=Linear, Evaluation Protocol=Zero-shot2025.10 | 61.54 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EP - CAMEXRouting Strategy=Cosine, Evaluation Protocol=Zero-shot2025.10 | 61.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen1.5 - MoERouting Strategy=Stable - MoE, Evaluation Protocol=Zero-shot2025.10 | 61.35 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen1.5 - MoERouting Strategy=Linear, Evaluation Protocol=Zero-shot2025.10 | 61.28 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen1.5 - MoERouting Strategy=Cosine, Evaluation Protocol=Zero-shot2025.10 | 61.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LORABackbone Model=Mistral-v0.1 7B, Precision=16-bit2024.02 | 60.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Pre-trainedBackbone Model=Mistral-v0.1 7B, Precision=16-bit2024.02 | 60.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| L4QModel=Mistral-v0.1 7B, Precision=4-bit2024.02 | 59 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FP16Model=Vicuna-v1.5-13B, Quantization=None2024.06 | 54.54 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 50.48 | 43.7 | 62.72 | 62.74 | |
| L4QBackbone Model=Mistral-v0.1 7B, Precision=3-bit2024.02 | 54.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| L4QModel=Mistral-v0.1 7B, Precision=3-bit2024.02 | 54.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LORABackbone Model=LLaMA-1 33B, Precision=16-bit2024.02 | 54.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OmniQBackbone Model=Mistral-v0.1 7B, Precision=3-bit2024.02 | 54.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| L4QModel=LLaMA-1 33B, Precision=4-bit2024.02 | 53.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Pre-trainedBackbone Model=LLaMA-1 33B, Precision=16-bit2024.02 | 53 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLaMA2-13bBit=16, Foundation Model=LLaMA2-13b2024.05 | 52.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 56.9 | 42.4 | 61 | 55.6 | |
| LORABackbone Model=LLaMA-2 13B, Precision=16-bit2024.02 | 52.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLM-QFABit=4, Foundation Model=LLaMA2-13b2024.05 | 52.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 57.4 | 41.3 | 60.4 | 55.8 | |
| QAT-LORABackbone Model=Mistral-v0.1 7B, Precision=3&16-bit2024.02 | 52.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QA-LoRABit=4, Foundation Model=LLaMA2-13b2024.05 | 52.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 56.9 | 41.5 | 60.4 | 54.9 | |
| QLORABackbone Model=Mistral-v0.1 7B, Precision=3&16-bit2024.02 | 52.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Pre-trainedBackbone Model=LLaMA-2 13B, Precision=16-bit2024.02 | 52.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| L4QModel=LLaMA-2 13B, Precision=4-bit2024.02 | 51.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLM-QFABit=3, Foundation Model=LLaMA2-13b2024.05 | 51.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 56.3 | 40.3 | 58.8 | 54.6 | |
| GPTQBit=4, Foundation Model=LLaMA2-13b2024.05 | 51.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 55.3 | 41.6 | 58.1 | 53.3 | |
| DuQuant+LWCModel=Vicuna-v1.5-13B, Quantization=W4A42024.06 | 51.08 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 47.48 | 40.13 | 58.86 | 57.83 | |
| DuQuantModel=Vicuna-v1.5-13B, Quantization=W4A42024.06 | 50.94 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 46.61 | 40.82 | 58.73 | 57.59 | |
| Token FilteringPruning Ratio=20%, Backbone=LLaMA-3-8B, Evaluation Protocol=zero-shot2025.12 | 50.59 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 45.65 | 43.2 | 56.42 | 59.8 | |
| L4QBackbone Model=LLaMA-1 33B, Precision=3-bit2024.02 | 50.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GPTQBackbone Model=Mistral-v0.1 7B, Precision=3-bit2024.02 | 50.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QA-LORABackbone Model=Mistral-v0.1 7B, Precision=3-bit2024.02 | 50.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| L4QModel=LLaMA-1 33B, Precision=3-bit2024.02 | 50.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OmniQBackbone Model=LLaMA-1 33B, Precision=3-bit2024.02 | 50.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QAT-LORABackbone Model=LLaMA-1 33B, Precision=3&16-bit2024.02 | 50.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GPTQBackbone Model=LLaMA-1 33B, Precision=3-bit2024.02 | 50 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QA-LoRABit=3, Foundation Model=LLaMA2-13b2024.05 | 49.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 54 | 40 | 57.1 | 52.5 | |
| SlimGPT w/oPruning Ratio=20%, Backbone=LLaMA-3-8B, Evaluation Protocol=zero-shot2025.12 | 47.44 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 43.51 | 40.88 | 54.63 | 52.91 | |
| L4QBackbone Model=LLaMA-2 13B, Precision=3-bit2024.02 | 47.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| L4QModel=LLaMA-2 13B, Precision=3-bit2024.02 | 47.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QLORABackbone Model=LLaMA-2 13B, Precision=3&16-bit2024.02 | 46.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QA-LORABackbone Model=LLaMA-2 13B, Precision=3-bit2024.02 | 46.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GPTQBackbone Model=LLaMA-2 13B, Precision=3-bit2024.02 | 46.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OmniQBackbone Model=LLaMA-2 13B, Precision=3-bit2024.02 | 46.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QA-LORABackbone Model=LLaMA-1 33B, Precision=3-bit2024.02 | 46.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FLAPPruning Ratio=20%, Backbone=LLaMA-3-8B, Evaluation Protocol=zero-shot2025.12 | 45.99 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 42.81 | 41.42 | 54.32 | 47.69 | |
| QAT-LORABackbone Model=LLaMA-2 13B, Precision=3&16-bit2024.02 | 45.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLM-QFABit=Avg. (2-4 bit), Foundation Model=LLaMA2-13b2024.05 | 45.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QA-LoRABit=4, Foundation Model=LLaMA2-7b2024.05 | 45.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 49.7 | 37.5 | 51.4 | 47.8 | |
| QLORABackbone Model=LLaMA-1 33B, Precision=3&16-bit2024.02 | 45.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QA-LoRABit=Avg. (2-4 bit), Foundation Model=LLaMA2-13b2024.05 | 45.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLM-QFABit=4, Foundation Model=LLaMA2-7b2024.05 | 45.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 50.3 | 37.4 | 49.8 | 46.8 | |
| LORABackbone Model=LLaMA-1 13B, Precision=16-bit2024.02 | 44.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LORABackbone Model=LLaMA-2 7B, Precision=16-bit2024.02 | 43.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Pre-trainedBackbone Model=LLaMA-1 13B, Precision=16-bit2024.02 | 43.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLaMA2-7bBit=16, Foundation Model=LLaMA2-7b2024.05 | 43.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 48.3 | 35.2 | 48.8 | 45.8 | |
| L4QModel=LLaMA-1 13B, Precision=4-bit2024.02 | 43.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Pre-trainedBackbone Model=LLaMA-2 7B, Precision=16-bit2024.02 | 41.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLM-QFABit=3, Foundation Model=LLaMA2-7b2024.05 | 41.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 42.3 | 34.4 | 48.1 | 42.9 | |
| AtomModel=Vicuna-v1.5-13B, Quantization=W4A42024.06 | 41.07 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 39.6 | 32.54 | 46.02 | 46.11 | |
| L4QModel=LLaMA-2 7B, Precision=4-bit2024.02 | 40.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DensePruning Ratio=0%, Model=Phi-4-14B, Zero-shot=true2025.12 | 40.81 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 39.21 | 33.17 | 45.99 | 45.41 | |
| DensePruning Ratio=0%, Backbone=LLaMA-3-8B, Evaluation Protocol=zero-shot2025.12 | 40.81 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 39.21 | 33.17 | 45.99 | 45.41 | |
| L4QBackbone Model=LLaMA-1 13B, Precision=3-bit2024.02 | 40.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| L4QModel=LLaMA-1 13B, Precision=3-bit2024.02 | 40.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QA-LoRABit=3, Foundation Model=LLaMA2-7b2024.05 | 40.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 43.3 | 33.7 | 44.8 | 42.9 | |
| GPTQBit=4, Foundation Model=LLaMA2-7b2024.05 | 39.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 40.4 | 33.7 | 45.9 | 42.2 | |
| LLM-QFABit=Avg. (2-4 bit), Foundation Model=LLaMA2-7b2024.05 | 39.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QA-LoRABit=Avg. (2-4 bit), Foundation Model=LLaMA2-7b2024.05 | 39.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GPTQBit=3, Foundation Model=LLaMA2-13b2024.05 | 39 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 42 | 31.8 | 43.6 | 41.3 | |
| QAT-LORABackbone Model=LLaMA-1 13B, Precision=3&16-bit2024.02 | 38.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GPTQBit=Avg. (2-4 bit), Foundation Model=LLaMA2-13b2024.05 | 37.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QA-LORABackbone Model=LLaMA-1 13B, Precision=3-bit2024.02 | 37.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LORABackbone Model=LLaMA-1 7B, Precision=16-bit2024.02 | 36.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QLORABackbone Model=LLaMA-1 13B, Precision=3&16-bit2024.02 | 36.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LoftQBackbone Model=Mistral-v0.1 7B, Precision=3&16-bit2024.02 | 35.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| L4QBackbone Model=LLaMA-2 7B, Precision=3-bit2024.02 | 34.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| L4QModel=LLaMA-1 7B, Precision=4-bit2024.02 | 34.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| L4QModel=LLaMA-2 7B, Precision=3-bit2024.02 | 34.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OmniQBackbone Model=LLaMA-1 13B, Precision=3-bit2024.02 | 34.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OmniQBackbone Model=LLaMA-2 7B, Precision=3-bit2024.02 | 34.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLM-QFABit=2, Foundation Model=LLaMA2-13b2024.05 | 33.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 34.5 | 30.3 | 33 | 37.3 | |
| GPTQBackbone Model=LLaMA-1 13B, Precision=3-bit2024.02 | 33.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Token FilteringPruning Ratio=50%, Model=Phi-4-14B, Zero-shot=true2025.12 | 33.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 29.69 | 31.43 | 34.68 | 38.2 | |
| Pre-trainedBackbone Model=LLaMA-1 7B, Precision=16-bit2024.02 | 32.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QLORABackbone Model=LLaMA-2 7B, Precision=3&16-bit2024.02 | 32.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LLM-QFABit=2, Foundation Model=LLaMA2-7b2024.05 | 32.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 33.7 | 28.7 | 36.3 | 32.9 | |
| QA-LoRABit=2, Foundation Model=LLaMA2-13b2024.05 | 31.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 32.6 | 28.9 | 31.4 | 35.3 | |
| QA-LoRABit=2, Foundation Model=LLaMA2-7b2024.05 | 31.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 32.6 | 27.2 | 35.6 | 33.2 | |
| QAT-LORABackbone Model=LLaMA-2 7B, Precision=3&16-bit2024.02 | 31.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GPTQBackbone Model=LLaMA-2 7B, Precision=3-bit2024.02 | 31.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QA-LORABackbone Model=LLaMA-2 7B, Precision=3-bit2024.02 | 31 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| QAT-LORABackbone Model=LLaMA-1 7B, Precision=3&16-bit2024.02 | 30.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| L4QBackbone Model=LLaMA-1 7B, Precision=3-bit2024.02 | 30.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| L4QModel=LLaMA-1 7B, Precision=3-bit2024.02 | 30.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GPTQBit=Avg. (2-4 bit), Foundation Model=LLaMA2-7b2024.05 | 30.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |