Common Sense Reasoning on Six Benchmarks Suite (ARC-e, PIQA, OpenbookQA, Winogrande, HellaSwag, MathQA)
61Average AccuracyOriginal
Evaluation results
| Method | Links | |
|---|---|---|
| OriginalModel=MISTRAL-7B, Compression Ratio=1.0, Evaluation Protocol=Zero-shot2026.04 | 61 | |
| OriginalModel=LLAMA 2-7B, Compression Ratio=1.0, Evaluation Protocol=Zero-shot2026.04 | 57 | |
| Swift-SVDModel=LLAMA 2-7B, Compression Ratio=0.8, Evaluation Protocol=Zero-shot2026.04 | 56 | |
| Swift-SVD*Model=LLAMA 2-7B, Compression Ratio=0.8, Evaluation Protocol=Zero-shot2026.04 | 56 | |
| Swift-SVD*Model=MISTRAL-7B, Compression Ratio=0.8, Evaluation Protocol=Zero-shot2026.04 | 55 | |
| Swift-SVDModel=MISTRAL-7B, Compression Ratio=0.8, Evaluation Protocol=Zero-shot2026.04 | 54 | |
| SVD-LLM (W)Model=LLAMA 2-7B, Compression Ratio=0.8, Evaluation Protocol=Zero-shot2026.04 | 53 | |
| OriginalModel=OPT-6.7B, Compression Ratio=1.0, Evaluation Protocol=Zero-shot2026.04 | 52 | |
| Swift-SVD*Model=OPT-6.7B, Compression Ratio=0.8, Evaluation Protocol=Zero-shot2026.04 | 51 | |
| Swift-SVDModel=OPT-6.7B, Compression Ratio=0.8, Evaluation Protocol=Zero-shot2026.04 | 50 | |
| SVD-LLM (W)Model=MISTRAL-7B, Compression Ratio=0.8, Evaluation Protocol=Zero-shot2026.04 | 42 | |
| SVD-LLM (W)Model=OPT-6.7B, Compression Ratio=0.8, Evaluation Protocol=Zero-shot2026.04 | 41 | |
| ASVDModel=LLAMA 2-7B, Compression Ratio=0.8, Evaluation Protocol=Zero-shot2026.04 | 36 | |
| ASVDModel=OPT-6.7B, Compression Ratio=0.8, Evaluation Protocol=Zero-shot2026.04 | 32 | |
| ASVDModel=MISTRAL-7B, Compression Ratio=0.8, Evaluation Protocol=Zero-shot2026.04 | 32 |