Common-sense Reasoning Average
90.4Average AccuracyFRAME
Evaluation Results
| Method | Links | |
|---|---|---|
| FRAMEBackbone=Qwen2.5, Model Scale=14B2026.06 | 90.4 | |
| LoRABackbone=Qwen2.5, Model Scale=14B2026.06 | 88.1 | |
| FRAMEBackbone=LLaMA-3.1, Model Scale=8B2026.06 | 87.1 | |
| LoRABackbone=LLaMA-3.1, Model Scale=8B2026.06 | 84.5 | |
| FRAMEBackbone=Qwen2.5, Model Scale=3B2026.06 | 81.7 | |
| LoRABackbone=Qwen2.5, Model Scale=3B2026.06 | 78.6 | |
| FRAMEBackbone=Qwen2.5, Model Scale=1.5B2026.06 | 73.6 | |
| FP16Bit=16, Backbone=Qwen2.5-14B-Ins2026.05 | 73.44 | |
| ProjQBit=3, Backbone=Qwen2.5-14B-Ins2026.05 | 71.61 | |
| FP16Bit=16, Backbone=LLaMA2-13B2026.05 | 70.61 | |
| LoRABackbone=Qwen2.5, Model Scale=1.5B2026.06 | 70.2 | |
| ProjQBit=3, Backbone=LLaMA2-13B2026.05 | 69.91 | |
| LoftQBit=3, Backbone=Qwen2.5-14B-Ins2026.05 | 69.89 | |
| FP16Bit=16, Backbone=Qwen2.5-7B-Ins2026.05 | 69.84 | |
| LoftQBit=3, Backbone=LLaMA2-13B2026.05 | 69.1 | |
| ProjQBit=3, Backbone=Qwen2.5-7B-Ins2026.05 | 68.33 | |
| FP16Bit=16, Backbone=LLaMA2-7B2026.05 | 68.07 | |
| ProjQBit=3, Backbone=LLaMA2-7B2026.05 | 67.13 | |
| QLoRA(RTN)Bit=3, Backbone=LLaMA2-7B2026.05 | 66.56 | |
| LoftQBit=3, Backbone=LLaMA2-7B2026.05 | 66.38 | |
| QLoRA(GPTQ)Bit=3, Backbone=LLaMA2-7B2026.05 | 66.15 | |
| QLoRA(GPTQ)Bit=3, Backbone=Qwen2.5-7B-Ins2026.05 | 65.08 | |
| QLoRA(RTN)Bit=3, Backbone=Qwen2.5-7B-Ins2026.05 | 64.91 | |
| LoftQBit=3, Backbone=Qwen2.5-7B-Ins2026.05 | 64.67 | |
| ProjQBit=2, Backbone=LLaMA2-13B2026.05 | 61.84 | |
| LoftQBit=2, Backbone=LLaMA2-13B2026.05 | 60.25 | |
| ProjQBit=2, Backbone=Qwen2.5-7B-Ins2026.05 | 59.88 | |
| ProjQBit=2, Backbone=Qwen2.5-14B-Ins2026.05 | 59.85 | |
| QLoRA(GPTQ)Bit=2, Backbone=Qwen2.5-7B-Ins2026.05 | 59.66 | |
| LoftQBit=2, Backbone=Qwen2.5-14B-Ins2026.05 | 59.08 | |
| LoftQBit=2, Backbone=Qwen2.5-7B-Ins2026.05 | 58.86 | |
| MLRA-4Evaluation Mode=Zero-shot2026.03 | 58.84 | |
| MLAEvaluation Mode=Zero-shot2026.03 | 58.75 | |
| MLRA-2Evaluation Mode=Zero-shot2026.03 | 58.68 | |
| MFAEvaluation Mode=Zero-shot2026.03 | 58.53 | |
| GLA-2Evaluation Mode=Zero-shot2026.03 | 58.3 | |
| GLA-4Evaluation Mode=Zero-shot2026.03 | 58.19 | |
| GQAEvaluation Mode=Zero-shot2026.03 | 57.89 | |
| TPAEvaluation Mode=Zero-shot2026.03 | 57.87 | |
| MHAEvaluation Mode=Zero-shot2026.03 | 57.81 | |
| GTAEvaluation Mode=Zero-shot2026.03 | 57.8 | |
| ProjQBit=2, Backbone=LLaMA2-7B2026.05 | 57.59 | |
| MQAEvaluation Mode=Zero-shot2026.03 | 57.55 | |
| QLoRA(GPTQ)Bit=2, Backbone=LLaMA2-7B2026.05 | 57.06 | |
| QLoRA(RTN)Bit=2, Backbone=Qwen2.5-7B-Ins2026.05 | 56.64 | |
| LoftQBit=2, Backbone=LLaMA2-7B2026.05 | 54.78 | |
| QLoRA(RTN)Bit=2, Backbone=LLaMA2-7B2026.05 | 54.64 |