Commonsense Question Answering Suite on Commonsense QA
77.4BoolQ AccuracyFull Precision
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Full PrecisionModel=LLaMA-2-7B, Wbit/Abit=16/162026.04 | 77.4 | 78.8 | 77.2 | 69.2 | 75.2 | 45.9 | 58.6 | 68.9 | — | |
| DoRABudget=100%, Base Model=Llama2-7B2026.05 | 71.8 | 83.7 | 89.1 | 82.6 | 83.7 | 68.2 | 82.4 | 79.7 | 76 | |
| SVFTBudget=100%, Base Model=Llama2-7B2026.05 | 71.2 | 81 | 89.6 | 81.1 | 81.8 | 66.5 | 81.4 | 78.5 | 76.3 | |
| JACTUSBudget=80%, Base Model=Llama2-7B2026.05 | 70.7 | 82.9 | 93 | 83.3 | 85.4 | 71.7 | 79.6 | 80.9 | 80.6 | |
| LoRABudget=100%, Base Model=Llama2-7B2026.05 | 69.8 | 79.9 | 83.6 | 82.6 | 79.8 | 64.7 | 81 | 77.6 | 79.5 | |
| Dobi-SVD+LoRABudget=80%, Base Model=Llama2-7B2026.05 | 69.2 | 81.7 | 82.1 | 82.5 | 78.3 | 64.3 | 79.7 | 76.9 | 77.6 | |
| SVD-LLM+LoRABudget=80%, Base Model=Llama2-7B2026.05 | 68.9 | 80.2 | 81.6 | 82.1 | 79.6 | 63.2 | 79.1 | 76.5 | 76.9 | |
| JACTUSBudget=60%, Base Model=Llama2-7B2026.05 | 66.8 | 79.8 | 87.8 | 78.5 | 76.9 | 61.5 | 75.4 | 75.5 | 77.4 | |
| Dobi-SVD+LoRABudget=60%, Base Model=Llama2-7B2026.05 | 66 | 70.7 | 75.3 | 78 | 78.7 | 58.9 | 72.6 | 71.5 | 72 | |
| SVD-LLM+LoRABudget=60%, Base Model=Llama2-7B2026.05 | 65.2 | 69.4 | 68.3 | 74.5 | 75.6 | 53.2 | 70.4 | 68 | 67.4 | |
| Dobi-SVD+LoRABudget=40%, Base Model=Llama2-7B2026.05 | 64.4 | 61.8 | 50.5 | 59.3 | 47 | 30.8 | 40 | 52.3 | 64.3 | |
| BWTA (Ours)Model=Bitnet-b1.58-3B, Wbit/Abit=1.4/7, quantized_layers=10%2026.04 | 63.7 | 71.9 | 42 | 59.9 | 65.5 | 29.4 | 26.6 | 51.3 | — | |
| JACTUSBudget=40%, Base Model=Llama2-7B2026.05 | 62.6 | 64.4 | 61.1 | 60 | 56.9 | 43 | 56.6 | 58.9 | 66.6 | |
| PB-LLMModel=LLaMA-2-7B, Wbit/Abit=1.7/162026.04 | 62.3 | 53.8 | 27.7 | 49.3 | 28 | 25 | 30.2 | 39.5 | — | |
| SVD-LLM+LoRABudget=40%, Base Model=Llama2-7B2026.05 | 62.2 | 64.3 | 53.1 | 59.1 | 44.2 | 32.9 | 40.7 | 51.7 | 56.8 | |
| BitnetModel=Bitnet-b1.58-1.3B, Wbit/Abit=1.5/82026.04 | 62 | 68.9 | 37.5 | 55.6 | 59.3 | 26.2 | 20.6 | 47.2 | — | |
| BiLLMModel=LLaMA-2-7B, Wbit/Abit=1.1/162026.04 | 61.8 | 60.6 | 34.8 | 52.4 | 36.2 | 24.4 | 33.2 | 43.3 | — | |
| BWTA (Ours)Model=Bitnet-b1.58-1.3B, Wbit/Abit=1.4/7, quantized_layers=10%2026.04 | 61.5 | 68.4 | 37.6 | 55.5 | 59 | 26.2 | 22.4 | 47.2 | — | |
| BWNModel=Bitnet-b1.58-1.3B, Wbit/Abit=1.4/7, quantized_layers=10%2026.04 | 59.9 | 69.3 | 37.1 | 55 | 57.8 | 26.3 | 20.8 | 46.6 | — | |
| BWNModel=Bitnet-b1.58-1.3B, Wbit/Abit=1.3/6, quantized_layers=30%2026.04 | 59.8 | 62.8 | 32.8 | 51.5 | 48 | 22.4 | 20.4 | 42.5 | — | |
| BitnetModel=Bitnet-b1.58-3B, Wbit/Abit=1.5/82026.04 | 58.9 | 72.3 | 42.8 | 60.8 | 65 | 30 | 26.6 | 50.9 | — | |
| BWTA (Ours)Model=Bitnet-b1.58-0.7B, Wbit/Abit=1.3/6, quantized_layers=30%2026.04 | 58.2 | 65 | 33 | 52.1 | 50.2 | 23.5 | 22 | 43.4 | — | |
| BWTA (Ours)Model=Bitnet-b1.58-1.3B, Wbit/Abit=1.3/6, quantized_layers=30%2026.04 | 57.5 | 66.5 | 35.4 | 53.7 | 54 | 25.2 | 21 | 44.8 | — | |
| BitnetModel=Bitnet-b1.58-0.7B, Wbit/Abit=1.5/82026.04 | 54 | 67.9 | 35 | 53.9 | 55.4 | 24.7 | 20 | 44.4 | — | |
| BWNModel=Bitnet-b1.58-0.7B, Wbit/Abit=1.3/6, quantized_layers=30%2026.04 | 53.7 | 62.5 | 32.1 | 49.8 | 43.9 | 21.5 | 19 | 40.4 | — | |
| BWNModel=Bitnet-b1.58-0.7B, Wbit/Abit=1.4/7, quantized_layers=10%2026.04 | 52.8 | 65.9 | 32.9 | 51.2 | 52.5 | 23.6 | 19 | 42.6 | — | |
| BWTA (Ours)Model=Bitnet-b1.58-0.7B, Wbit/Abit=1.4/7, quantized_layers=10%2026.04 | 52.6 | 68.3 | 34.7 | 54 | 54.2 | 25.4 | 22 | 44.5 | — | |
| GPTQModel=LLaMA-2-7B, Wbit/Abit=2/162026.04 | 43.9 | 51.1 | 26.3 | 50.8 | 26.6 | 28.5 | 29 | 36.6 | — | |
| BWNModel=Bitnet-b1.58-3B, Wbit/Abit=1.4/7, quantized_layers=10%2026.04 | 37.8 | 49.6 | 25 | 49.6 | 25.7 | 22.9 | 27 | 33.9 | — |