Question Answering on QA Suite Zero-shot (PIQA, ARC-E, ARC-C, BoolQ, HellaSwag, WinoGrande)
82.48PIQA AccuracySDLLM
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SDLLMSpike=✓, Bit=6-1.58, T × D=1 × 32, R=0.215, Backbone=Qwen2.5-14B2026.04 | 82.48 | 79.04 | 57.76 | 84.83 | 82.85 | 75.22 | 77.03 | 8.73 | 7.85 | — | — | |
| Qwen2.5-14BSpike=✗, Backbone=Qwen2.5-14B2026.04 | 82.1 | 79.12 | 58.87 | 85.26 | 82.91 | 75.3 | 77.26 | 13.53 | 62.23 | — | — | |
| SDLLMSpike=✓, Bit=4-1.58, T × D=1 × 16, R=0.213, Backbone=Qwen2.5-14B2026.04 | 81.28 | 80.43 | 55.29 | 82.64 | 81.41 | 74.82 | 76.15 | 2.88 | 2.59 | — | — | |
| FP16Model=LLAMA-3-8B, Type=N/A2026.06 | 81 | 81 | 57.7 | 83.6 | 79.5 | 73.9 | 76.12 | — | — | — | 100 | |
| FP16Model=LLaMA3-8B, Quantization=FP16, Protocol=Zero-shot2026.04 | 80.85 | 77.78 | 53.41 | 81.28 | 79.16 | 72.84 | 74.22 | — | — | — | — | |
| FP16Backbone=LLaMA3-8B, Quantization=None, Evaluation Protocol=Zero-shot2026.04 | 80.85 | 77.78 | 53.41 | 81.28 | 79.16 | 72.84 | 74.22 | — | — | — | — | |
| LLaMA3-8BSpike=✗, Bit=-, T x D=-, ρ=-, FLOPs(T)=7.97, Power(J)=36.672026.04 | 80.85 | 77.78 | 53.41 | 81.28 | 79.16 | 72.84 | 74.22 | — | — | — | — | |
| FP16Backbone=LLaMA1-65B, Quantization=FP162024.06 | 80.79 | 58.71 | 46.24 | 82.29 | 80.72 | 77.5 | 71.04 | — | — | — | — | |
| LLaMA2-13BSpike=✗, Bit=-, T x D=-, R=-2026.04 | 80.63 | 77.48 | 49.23 | 80.73 | 79.37 | 71.74 | 73.2 | 13.42 | 61.74 | — | — | |
| RTNSpike=✗, Bit=6-6, Backbone=Qwen2.5-14B2026.04 | 80.41 | 81.4 | 56.57 | 84.19 | 81.52 | 71.27 | 75.89 | 3.81 | 17.5 | — | — | |
| RUQuantModel=LLaMA3-8B, Quantization=W6A6, Protocol=Zero-shot2026.04 | 80.25 | 77.44 | 52.73 | 80.4 | 78.99 | 73.24 | 73.84 | — | — | — | — | |
| DuQuantModel=LLaMA3-8B, Quantization=W6A6, Protocol=Zero-shot2026.04 | 80.2 | 77.27 | 52.05 | 80.12 | 79.14 | 72.77 | 73.59 | — | — | — | — | |
| DuQuantSpike=✗, Bit=6-6, T x D=-, ρ=-, FLOPs(T)=2.24, Power(J)=10.312026.04 | 80.2 | 77.27 | 52.05 | 80.12 | 79.14 | 72.77 | 73.59 | — | — | — | — | |
| SDLLMSpike=✓, Bit=6-1.58, T x D=1 × 32, ρ=0.213, FLOPs(T)=5.09, Power(J)=4.582026.04 | 80.2 | 77.23 | 52.22 | 82.05 | 79.01 | 73.56 | 74.04 | — | — | — | — | |
| FP16Backbone=LLaMA1-30B, Quantization=FP162024.06 | 80.08 | 58.92 | 45.47 | 68.44 | 79.21 | 72.53 | 67.44 | — | — | — | — | |
| DuQuant+LWCBackbone=LLaMA1-65B, Quantization=W4A42024.06 | 79.98 | 58.29 | 44.8 | 77.89 | 79.22 | 72.21 | 68.73 | — | — | — | — | |
| DuQuantBackbone=LLaMA1-65B, Quantization=W4A42024.06 | 79.71 | 57.95 | 45.05 | 79.82 | 78.66 | 72.29 | 68.91 | — | — | — | — | |
| GPTQSpike=✗, Bit=6-6, Backbone=Qwen2.5-14B2026.04 | 79.71 | 76.85 | 52.82 | 80.24 | 80.11 | 70.17 | 73.32 | 3.81 | 17.5 | — | — | |
| SmoothQuantSpike=✗, Bit=6-6, Backbone=Qwen2.5-14B2026.04 | 79.33 | 78.96 | 55.03 | 80.89 | 79.12 | 68.67 | 73.66 | 3.81 | 17.5 | — | — | |
| PARETOQModel=LLAMA-3-8B, Type=QAT, Training Tokens=30B2026.06 | 79.2 | 78.5 | 54.5 | 76.4 | 73.8 | 70 | 72.07 | — | — | — | 95 | |
| RUQuantModel=LLaMA2-13B, Quantization=W6A6, Protocol=Zero-shot2026.04 | 79.16 | 57.74 | 44.45 | 67.83 | 76.45 | 69.77 | 65.9 | — | — | — | — | |
| FP16Backbone=LLaMA1-13B, Quantization=FP162024.06 | 79.1 | 59.89 | 44.45 | 68.01 | 76.21 | 70.31 | 66.33 | — | — | — | — | |
| FP16Model=LLAMA1-13B, Bits=FP162025.04 | 79.05 | 59.89 | 44.71 | 68.47 | 76.23 | 70.24 | 66.43 | — | — | — | — | |
| FP16Model=LLaMA2-13B, Quantization=FP16, Protocol=Zero-shot2026.04 | 79.05 | 57.91 | 44.2 | 69.02 | 76.6 | 69.69 | 66.08 | — | — | — | — | |
| FP16Backbone=LLaMA2-13B, Quantization=None, Evaluation Protocol=Zero-shot2026.04 | 79.05 | 57.91 | 44.2 | 69.02 | 76.6 | 69.69 | 66.08 | — | — | — | — | |
| LLAMA2-13BBackbone=LLaMA2-13B, Spike=✗, Bit=-, T x D=-, r=-, FLOPs(T)=13.42, Power(J)=65.772026.04 | 79.05 | 57.91 | 44.2 | 69.02 | 76.6 | 69.69 | 66.08 | — | — | — | — | |
| SDLLMBackbone=LLaMA2-13B, Spike=✓, Bit=6-1.58, T x D=1 x 32, r=0.217, FLOPs(T)=8.74, Power(J)=7.862026.04 | 79.05 | 57.66 | 44.2 | 67.83 | 76.42 | 69.93 | 65.85 | — | — | — | — | |
| LLaMA2-13BBackbone=LLaMA2-13B, Spike-driven=false, Bit=-, T x D=-, R=-2026.04 | 79.05 | 57.91 | 44.2 | 69.02 | 76.6 | 69.69 | 66.08 | 13.42 | 61.74 | — | — | |
| FP16Model=LLAMA2-13B, Bits=FP162025.04 | 79 | 57.95 | 44.28 | 69.02 | 76.58 | 69.69 | 66.09 | — | — | — | — | |
| SDLLMSpike=✓, Bit=4-1.58, T × D=1 × 8, R=0.212, Backbone=Qwen2.5-14B2026.04 | 79 | 77.82 | 52.13 | 80 | 78.28 | 67.8 | 72.51 | 1.43 | 1.29 | — | — | |
| SmoothQuantModel=LLaMA3-8B, Quantization=W6A6, Protocol=Zero-shot2026.04 | 78.94 | 75.88 | 49.49 | 77.58 | 77.39 | 70.8 | 71.68 | — | — | — | — | |
| SmoothQuantSpike=✗, Bit=6-6, T x D=-, ρ=-, FLOPs(T)=2.24, Power(J)=10.312026.04 | 78.94 | 75.88 | 49.49 | 77.58 | 77.39 | 70.8 | 71.68 | — | — | — | — | |
| OmniQuantModel=LLaMA3-8B, Quantization=W6A6, Protocol=Zero-shot2026.04 | 78.9 | 73.95 | 47.35 | 74.95 | 76.77 | 70.56 | 70.41 | — | — | — | — | |
| OmniQuantSpike=✗, Bit=6-6, T x D=-, ρ=-, FLOPs(T)=2.24, Power(J)=10.312026.04 | 78.9 | 73.95 | 47.35 | 74.95 | 76.77 | 70.56 | 70.41 | — | — | — | — | |
| LLaMA2-7BSpike=✗, Bit=-, T x D=-, R=-2026.04 | 78.84 | 74.54 | 46.33 | 77.74 | 75.97 | 69.22 | 70.44 | 6.91 | 31.77 | — | — | |
| QLLMModel=LLaMA2-13B, Quantization=W6A6, Protocol=Zero-shot2026.04 | 78.78 | 58.29 | 43.77 | — | 75.1 | 68.43 | 64.87 | — | — | — | — | |
| QLLMBackbone=LLaMA2-13B, Spike=✗, Bit=6-6, T x D=-, r=-, FLOPs(T)=3.77, Power(J)=17.362026.04 | 78.78 | 58.29 | 43.77 | — | 75.1 | 68.43 | 64.87 | — | — | — | — | |
| DuQuant+LWCBackbone=LLaMA1-30B, Quantization=W4A42024.06 | 78.73 | 56.52 | 43.17 | 68.84 | 77.53 | 70.96 | 65.96 | — | — | — | — | |
| AffineQuantSpike=✗, Bit=6-6, T x D=-, ρ=-, FLOPs(T)=2.24, Power(J)=10.312026.04 | 78.73 | 73.32 | 46.08 | 74.59 | 77.08 | 70.88 | 70.11 | — | — | — | — | |
| DuQuantModel=LLaMA2-13B, Quantization=W6A6, Protocol=Zero-shot2026.04 | 78.62 | 56.94 | 43.43 | 68.35 | 76.19 | 69.22 | 65.46 | — | — | — | — | |
| DuQuantBackbone=LLaMA2-13B, Spike=✗, Bit=6-6, T x D=-, r=-, FLOPs(T)=3.77, Power(J)=17.362026.04 | 78.62 | 56.94 | 43.43 | 68.35 | 76.19 | 69.22 | 65.46 | — | — | — | — | |
| LC-QATModel=LLAMA-3-8B, Type=QAT, Training Tokens=4B2026.06 | 78.62 | 82.95 | 57.94 | 77.46 | 76.67 | 66.85 | 73.42 | — | — | — | 96 | |
| DuQuantBackbone=LLaMA1-30B, Quantization=W4A42024.06 | 78.56 | 56.99 | 42.32 | 66.73 | 76.7 | 69.61 | 65.15 | — | — | — | — | |
| RUQuantBackbone=LLaMA2-13B, Quantization=W4A4, Evaluation Protocol=Zero-shot, fine-tune=true2026.04 | 78.51 | 56.14 | 41.98 | 68.41 | 73.78 | 65.98 | 64.13 | — | — | — | — | |
| SDLLMBase Model=LLaMA2-13B, Spike=✓, Bit=4-1.58, T x D=1 × 8, R=0.2092026.04 | 78.51 | 74.12 | 46.16 | 78.26 | 76.36 | 69.85 | 70.54 | 1.4 | 1.26 | — | — | |
| AffineQuantBackbone=LLaMA2-13B, Spike=✗, Bit=6-6, T x D=-, r=-, FLOPs(T)=3.77, Power(J)=17.362026.04 | 78.35 | 57.58 | 43.34 | 66.73 | 74.71 | 68.59 | 64.88 | — | — | — | — | |
| SmoothQuantModel=LLaMA2-13B, Quantization=W6A6, Protocol=Zero-shot2026.04 | 78.29 | 57.41 | 43.86 | 69.5 | 75.02 | 66.93 | 65.17 | — | — | — | — | |
| SmoothQuantBackbone=LLaMA2-13B, Spike=✗, Bit=6-6, T x D=-, r=-, FLOPs(T)=3.77, Power(J)=17.362026.04 | 78.29 | 57.41 | 43.86 | 69.5 | 75.02 | 66.93 | 65.17 | — | — | — | — | |
| OS+Backbone=LLaMA2-13B, Spike=✗, Bit=6-6, T x D=-, r=-, FLOPs(T)=3.77, Power(J)=17.362026.04 | 78.29 | 59.13 | 43.34 | — | 75.37 | 67.56 | 64.74 | — | — | — | — | |
| OmniQuantModel=LLaMA2-13B, Quantization=W6A6, Protocol=Zero-shot2026.04 | 78.24 | 57.58 | 43.86 | 71.1 | 75.52 | 68.35 | 65.78 | — | — | — | — | |
| RUQuantBackbone=LLaMA2-13B, Quantization=W4A4, Evaluation Protocol=Zero-shot2026.04 | 78.24 | 55.47 | 41.3 | 68.2 | 74.07 | 67.64 | 64.15 | — | — | — | — | |
| OmniQuantBackbone=LLaMA2-13B, Spike=✗, Bit=6-6, T x D=-, r=-, FLOPs(T)=3.77, Power(J)=17.362026.04 | 78.24 | 57.58 | 43.86 | 71.1 | 75.52 | 68.35 | 65.78 | — | — | — | — | |
| FP16Model=LLAMA-3-3B, Type=N/A2026.06 | 78.2 | 72.6 | 50.7 | 74.6 | 74.3 | 69.2 | 69.93 | — | — | — | 100 | |
| QUIP#Model=LLama3-8B-Instruct, Quantization Type=Vector2026.06 | 78.13 | 73.27 | 47.44 | 80.21 | 70.7 | 69.53 | 69.72 | — | — | 94 | — | |
| FP16Model=LLama3-8B-Instruct2026.06 | 77.8 | 80.93 | 56.97 | 86.61 | 74.94 | 67.88 | 74.12 | — | — | 100 | — | |
| AtomModel=LLAMA1-13B, Bits=W4A42025.04 | 77.64 | 58.38 | 41.81 | 68.5 | 73.75 | 65.98 | 64.64 | — | — | — | — | |
| DuQuant+LWCBackbone=LLaMA1-13B, Quantization=W4A42024.06 | 77.64 | 57.32 | 41.21 | 66.79 | 74.12 | 65.98 | 63.84 | — | — | — | — | |
| QLLMModel=LLaMA2-7B, Quantization=W6A6, Protocol=Zero-shot2026.04 | 77.48 | 52.99 | 39.33 | — | 71.38 | 65.98 | 61.43 | — | — | — | — | |
| QLLMBackbone=LLaMA-2-7B, Spike=✗, Bit=6-6, T x D=-, r=-, FLOPs(T)=1.94, Power(J)=8.942026.04 | 77.48 | 52.99 | 39.33 | — | 71.38 | 65.98 | 61.43 | — | — | — | — | |
| FP16QUANTIZATION TYPE=N/A, Quantization Bits=16-bit, Model Backbone=Qwen-3-8B, Evaluation Protocol=Zero-shot2026.06 | 77.48 | 80.89 | 56.4 | 86.64 | 74.96 | 68.35 | 74.12 | — | — | — | 100 | |
| FP16Model=QWEN-3-8B, Type=N/A, Zero-shot=true2026.06 | 77.48 | 80.89 | 56.4 | 86.64 | 74.96 | 68.35 | 74.12 | — | — | — | 100 | |
| FP16Backbone=LLaMA1-7B, Quantization=FP162024.06 | 77.47 | 52.48 | 41.46 | 73.08 | 73 | 67.07 | 64.09 | — | — | — | — | |
| YAQAModel=QWEN-3-8B, Type=VQ, Zero-shot=true2026.06 | 77.42 | 78.41 | 53.16 | 83.27 | 71.41 | 69.14 | 72.14 | — | — | — | 97 | |
| AtomModel=LLAMA2-13B, Bits=W4A42025.04 | 77.37 | 56.73 | 42.32 | 67.62 | 74.07 | 68.27 | 65.46 | — | — | — | — | |
| DuQuantBackbone=LLaMA2-13B, Quantization=W4A4, Evaluation Protocol=Zero-shot2026.04 | 77.31 | 55.6 | 41.55 | 66.61 | 73.68 | 66.06 | 63.47 | — | — | — | — | |
| DuQuantBackbone=LLaMA2-13B, Spike-driven=false, Bit=4-4, T x D=-, R=-2026.04 | 77.31 | 55.6 | 41.55 | 66.61 | 73.68 | 66.06 | 63.47 | 1.68 | 7.72 | — | — | |
| DuQuantBackbone=LLaMA1-13B, Quantization=W4A42024.06 | 77.26 | 58.04 | 41.55 | 67.55 | 73.62 | 66.69 | 64.12 | — | — | — | — | |
| SDLLMBackbone=LLaMA2-13B, Spike-driven=true, Bit=4-1.58, T x D=1 × 8, R=0.2092026.04 | 77.26 | 57.41 | 41.55 | 66.67 | 73.33 | 66.69 | 63.82 | 1.4 | 1.26 | — | — | |
| QTIPModel=QWEN-3-8B, Type=VQ, Zero-shot=true2026.06 | 77.26 | 77.31 | 52.82 | 85.24 | 70.85 | 68.75 | 72.04 | — | — | — | 97 | |
| RUQuantModel=LLaMA2-7B, Quantization=W6A6, Protocol=Zero-shot2026.04 | 77.04 | 52.97 | 40.61 | 70.4 | 72.71 | 67.88 | 63.6 | — | — | — | — | |
| QTIPQUANTIZATION TYPE=VECTOR, Quantization Bits=2-bit, Model Backbone=Qwen-3-8B, Evaluation Protocol=Zero-shot2026.06 | 77.04 | 77.06 | 52.05 | 84.95 | 68.14 | 67.25 | 71.08 | — | — | — | 96 | |
| DuQuantModel=LLaMA2-7B, Quantization=W6A6, Protocol=Zero-shot2026.04 | 76.99 | 52.99 | 40.87 | 70.4 | 72.49 | 67.32 | 63.51 | — | — | — | — | |
| DuQuantBackbone=LLaMA-2-7B, Spike=✗, Bit=6-6, T x D=-, r=-, FLOPs(T)=1.94, Power(J)=8.942026.04 | 76.99 | 52.99 | 40.87 | 70.4 | 72.49 | 67.32 | 63.51 | — | — | — | — | |
| SDLLMBackbone=LLaMA-2-7B, Spike=✓, Bit=6-1.58, T x D=1 x 32, r=0.222, FLOPs(T)=4.60, Power(J)=4.142026.04 | 76.99 | 53.75 | 41.04 | 70.64 | 72.84 | 67.25 | 63.75 | — | — | — | — | |
| LC-QATModel=QWEN-3-8B, Type=VQ-QAT, Zero-shot=true2026.06 | 76.99 | 78.82 | 53.75 | 82.29 | 71.37 | 69.85 | 72.18 | — | — | — | 97 | |
| QuaRotModel=LLAMA2-13B, Bits=W4A42025.04 | 76.93 | 52.1 | 40.7 | 68.1 | 72.7 | 62.51 | 62.99 | — | — | — | — | |
| RUQuantBackbone=LLaMA3-8B, Quantization=W4A4, Evaluation Protocol=Zero-shot2026.04 | 76.93 | 70.45 | 44.97 | 75.2 | 73.36 | 66.14 | 67.84 | — | — | — | — | |
| FP16Model=LLaMA2-7B, Quantization=FP16, Protocol=Zero-shot2026.04 | 76.88 | 53.54 | 40.53 | 71.13 | 72.96 | 67.25 | 63.72 | — | — | — | — | |
| FP16Backbone=LLaMA2-7B, Quantization=None, Evaluation Protocol=Zero-shot2026.04 | 76.88 | 53.54 | 40.53 | 71.13 | 72.96 | 67.25 | 63.72 | — | — | — | — | |
| LLAMA-2-7BBackbone=LLaMA-2-7B, Spike=✗, Bit=-, T x D=-, r=-, FLOPs(T)=6.91, Power(J)=31.772026.04 | 76.88 | 53.54 | 40.53 | 71.13 | 72.96 | 67.25 | 63.72 | — | — | — | — | |
| LLaMA2-7BBackbone=LLaMA2-7B, Spike-driven=false, Bit=-, T x D=-, R=-2026.04 | 76.88 | 53.54 | 40.53 | 71.13 | 72.96 | 67.25 | 63.72 | 6.91 | 31.77 | — | — | |
| PV-TUNINGModel=QWEN-3-8B, Type=VQ-QAT, Zero-shot=true2026.06 | 76.88 | 75.46 | 51.79 | 83.82 | 70.44 | 67.56 | 70.99 | — | — | — | 96 | |
| QuaRotModel=LLAMA1-13B, Bits=W4A42025.04 | 76.61 | 55.3 | 41.64 | 67.09 | 73.02 | 65.59 | 64.24 | — | — | — | — | |
| OmniQuantModel=LLaMA2-7B, Quantization=W6A6, Protocol=Zero-shot2026.04 | 76.55 | 53.83 | 40.96 | 68.75 | 55.89 | 65.59 | 60.26 | — | — | — | — | |
| OmniQuantBackbone=LLaMA-2-7B, Spike=✗, Bit=6-6, T x D=-, r=-, FLOPs(T)=1.94, Power(J)=8.942026.04 | 76.55 | 53.83 | 40.96 | 68.75 | 55.89 | 65.59 | 60.26 | — | — | — | — | |
| DuQuantBackbone=LLaMA1-7B, Quantization=W4A42024.06 | 76.44 | 50.04 | 38.99 | 70.98 | 69.39 | 64.72 | 61.76 | — | — | — | — | |
| EFFICIENTQATModel=LLAMA-3-8B, Type=QAT, Training Tokens=24M2026.06 | 76.4 | 69.3 | 46.8 | 75.05 | 69 | 66.3 | 67.14 | — | — | — | 88 | |
| PARETOQModel=LLAMA-3-3B, Type=QAT, Training Tokens=30B2026.06 | 76.4 | 73.09 | 49 | 68.8 | 69.2 | 64.4 | 66.82 | — | — | — | 96 | |
| UniSVQModel=LLama3-8B-Instruct, Quantization Type=Vector2026.06 | 76.27 | 69.52 | 44.79 | 81.77 | 67.2 | 66.14 | 67.62 | — | — | 91 | — | |
| DuQuant+LWCBackbone=LLaMA1-7B, Quantization=W4A42024.06 | 76.22 | 50.04 | 38.31 | 70.09 | 69.82 | 62.59 | 61.18 | — | — | — | — | |
| OS+Backbone=LLaMA-2-7B, Spike=✗, Bit=6-6, T x D=-, r=-, FLOPs(T)=1.94, Power(J)=8.942026.04 | 76.22 | 52.74 | 40.7 | — | 71.89 | 65.19 | 61.35 | — | — | — | — | |
| LC-QATModel=LLAMA-3-3B, Type=QAT, Training Tokens=4B2026.06 | 76.22 | 78.75 | 52.39 | 66.51 | 69.83 | 66.3 | 68.33 | — | — | — | 98 | |
| RUQuantBackbone=LLaMA3-8B, Quantization=W4A4, Evaluation Protocol=Zero-shot, fine-tune=true2026.04 | 76.12 | 69.53 | 42.92 | 71.93 | 74.13 | 67.01 | 66.94 | — | — | — | — | |
| RUQuantBackbone=LLaMA2-7B, Quantization=W4A4, Evaluation Protocol=Zero-shot2026.04 | 75.84 | 52.82 | 39.51 | 69.57 | 70.06 | 63.61 | 61.9 | — | — | — | — | |
| SDLLMBase Model=LLaMA2-7B, Spike=✓, Bit=4-1.58, T x D=1 × 8, R=0.2162026.04 | 75.84 | 69.65 | 41.21 | 74.01 | 71.75 | 66.14 | 66.43 | 0.75 | 0.67 | — | — | |
| SpikeLLMBase Model=LLaMA2-13B, Spike=✓, Bit=4-4, T x D=1.2 × 1, R=-2026.04 | 75.79 | 69.53 | 41.21 | 74.31 | 71.51 | 65.51 | 66.31 | 2.01 | 9.26 | — | — | |
| DuQuantBackbone=LLaMA3-8B, Quantization=W4A4, Evaluation Protocol=Zero-shot2026.04 | 75.68 | 68.48 | 41.81 | 71.99 | 73.07 | 66.22 | 66.21 | — | — | — | — | |
| SmoothQuantModel=LLaMA2-7B, Quantization=W6A6, Protocol=Zero-shot2026.04 | 75.57 | 53.62 | 39.93 | 69.54 | 71.76 | 66.14 | 62.76 | — | — | — | — | |
| SmoothQuantBackbone=LLaMA-2-7B, Spike=✗, Bit=6-6, T x D=-, r=-, FLOPs(T)=1.94, Power(J)=8.942026.04 | 75.57 | 53.62 | 39.93 | 69.54 | 71.76 | 66.14 | 62.76 | — | — | — | — | |
| DuQuantBackbone=LLaMA2-7B, Quantization=W4A4, Evaluation Protocol=Zero-shot2026.04 | 75.24 | 51.89 | 36.77 | 67.86 | 69.54 | 62.12 | 60.57 | — | — | — | — |