Image Classification on ImageNet-1K (val) (Accuracy and Computational Cost)
86.84Top-1 AccuracyQFlash
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| QFlashBackbone Architecture=ViT-B, FP32 Accuracy Baseline=85.10, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 86.84 | 34.3 | |
| I-ViTBackbone Architecture=ViT-B, FP32 Accuracy Baseline=85.10, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 85.02 | 34.3 | |
| INT-Flash-HalfBackbone Architecture=ViT-B, FP32 Accuracy Baseline=85.10, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=△, Key (K) Quantization Granularity=△2026.04 | 84.83 | 68.7 | |
| FQ-ViTBackbone Architecture=ViT-B, FP32 Accuracy Baseline=85.10, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 84.82 | 34.3 | |
| INT-Flash-FullBackbone Architecture=ViT-B, FP32 Accuracy Baseline=85.10, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=△, Key (K) Quantization Granularity=△, Value (V) Quantization Granularity=◦2026.04 | 84.74 | 34.3 | |
| QAttnBackbone Architecture=ViT-B, FP32 Accuracy Baseline=85.10, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 84.68 | 68.7 | |
| I-BERTBackbone Architecture=ViT-B, FP32 Accuracy Baseline=85.10, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 83.97 | 34.3 | |
| INT-Flash-HalfBackbone Architecture=Swin-S, FP32 Accuracy Baseline=83.20, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=△, Key (K) Quantization Granularity=△2026.04 | 83.2 | 43.9 | |
| FQ-ViTBackbone Architecture=Swin-S, FP32 Accuracy Baseline=83.20, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 83.15 | 22 | |
| I-BERTBackbone Architecture=Swin-S, FP32 Accuracy Baseline=83.20, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 83.09 | 22 | |
| I-ViTBackbone Architecture=Swin-S, FP32 Accuracy Baseline=83.20, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 82.81 | 22 | |
| QFlashBackbone Architecture=ViT-S, FP32 Accuracy Baseline=81.38, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 82.24 | 17.2 | |
| QAttnBackbone Architecture=Swin-S, FP32 Accuracy Baseline=83.20, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 82.24 | 43.9 | |
| INT-Flash-FullBackbone Architecture=Swin-S, FP32 Accuracy Baseline=83.20, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=△, Key (K) Quantization Granularity=△, Value (V) Quantization Granularity=◦2026.04 | 82.11 | 22 | |
| INT-Flash-HalfBackbone Architecture=DeiT-B, FP32 Accuracy Baseline=81.85, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=△, Key (K) Quantization Granularity=△2026.04 | 81.9 | 137 | |
| QFlashBackbone Architecture=Swin-S, FP32 Accuracy Baseline=83.20, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 81.86 | 22 | |
| INT-Flash-FullBackbone Architecture=DeiT-B, FP32 Accuracy Baseline=81.85, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=△, Key (K) Quantization Granularity=△, Value (V) Quantization Granularity=◦2026.04 | 81.85 | 68.7 | |
| QAttnBackbone Architecture=DeiT-B, FP32 Accuracy Baseline=81.85, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 81.85 | 137 | |
| FQ-ViTBackbone Architecture=DeiT-B, FP32 Accuracy Baseline=81.85, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 81.72 | 68.7 | |
| QFlashBackbone Architecture=DeiT-B, FP32 Accuracy Baseline=81.85, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 81.59 | 68.7 | |
| I-BERTBackbone Architecture=DeiT-B, FP32 Accuracy Baseline=81.85, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 81.59 | 68.7 | |
| I-ViTBackbone Architecture=DeiT-B, FP32 Accuracy Baseline=81.85, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 81.47 | 68.7 | |
| I-ViTBackbone Architecture=ViT-S, FP32 Accuracy Baseline=81.38, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 81.19 | 17.2 | |
| I-BERTBackbone Architecture=Swin-T, FP32 Accuracy Baseline=81.35, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 81.11 | 13.5 | |
| FQ-ViTBackbone Architecture=ViT-S, FP32 Accuracy Baseline=81.38, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Granularity=▽, Value (V) Granularity=▽2026.04 | 81.06 | 17.2 | |
| I-BERTBackbone Architecture=ViT-S, FP32 Accuracy Baseline=81.38, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 81 | 17.2 | |
| FQ-ViTBackbone Architecture=Swin-T, FP32 Accuracy Baseline=81.35, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 80.9 | 13.5 | |
| I-ViTBackbone Architecture=Swin-T, FP32 Accuracy Baseline=81.35, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 80.83 | 13.5 | |
| INT-Flash-HalfBackbone Architecture=ViT-S, FP32 Accuracy Baseline=81.38, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=△, Key (K) Quantization Granularity=△2026.04 | 80.62 | 34.3 | |
| INT-Flash-FullBackbone Architecture=ViT-S, FP32 Accuracy Baseline=81.38, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=△, Key (K) Quantization Granularity=△, Value (V) Quantization Granularity=◦2026.04 | 80.6 | 17.2 | |
| QAttnBackbone Architecture=ViT-S, FP32 Accuracy Baseline=81.38, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 80.23 | 34.3 | |
| QAttnBackbone Architecture=Swin-T, FP32 Accuracy Baseline=81.35, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 80.13 | 26.9 | |
| QFlashBackbone Architecture=Swin-T, FP32 Accuracy Baseline=81.35, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 80.06 | 13.5 | |
| INT-Flash-HalfBackbone Architecture=Swin-T, FP32 Accuracy Baseline=81.35, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=△, Key (K) Quantization Granularity=△2026.04 | 80.05 | 26.9 | |
| INT-Flash-FullBackbone Architecture=Swin-T, FP32 Accuracy Baseline=81.35, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=△, Key (K) Quantization Granularity=△, Value (V) Quantization Granularity=◦2026.04 | 80.04 | 13.5 | |
| INT-Flash-HalfBackbone Architecture=DeiT-S, FP32 Accuracy Baseline=79.85, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=△, Key (K) Quantization Granularity=△2026.04 | 79.68 | 68.7 | |
| INT-Flash-FullBackbone Architecture=DeiT-S, FP32 Accuracy Baseline=79.85, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=△, Key (K) Quantization Granularity=△, Value (V) Quantization Granularity=◦2026.04 | 79.64 | 34.3 | |
| I-ViTBackbone Architecture=DeiT-S, FP32 Accuracy Baseline=79.85, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 79.53 | 34.3 | |
| I-BERTBackbone Architecture=DeiT-S, FP32 Accuracy Baseline=79.85, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 79.48 | 34.3 | |
| QFlashBackbone Architecture=DeiT-S, FP32 Accuracy Baseline=79.85, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 79.46 | 34.3 | |
| QAttnBackbone Architecture=DeiT-S, FP32 Accuracy Baseline=79.85, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 79.43 | 68.7 | |
| FQ-ViTBackbone Architecture=DeiT-S, FP32 Accuracy Baseline=79.85, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 79.39 | 34.3 | |
| QFlashBackbone Architecture=DeiT-T, FP32 Accuracy Baseline=72.21, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 71.7 | 17.2 | |
| I-ViTBackbone Architecture=DeiT-T, FP32 Accuracy Baseline=72.21, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 71.7 | 17.2 | |
| INT-Flash-FullBackbone Architecture=DeiT-T, FP32 Accuracy Baseline=72.21, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=△, Key (K) Quantization Granularity=△, Value (V) Quantization Granularity=◦2026.04 | 71.64 | 17.2 | |
| INT-Flash-HalfBackbone Architecture=DeiT-T, FP32 Accuracy Baseline=72.21, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=△, Key (K) Quantization Granularity=△2026.04 | 71.64 | 34.3 | |
| FQ-ViTBackbone Architecture=DeiT-T, FP32 Accuracy Baseline=72.21, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 71.53 | 17.2 | |
| QAttnBackbone Architecture=DeiT-T, FP32 Accuracy Baseline=72.21, Integer-Only (Int-Only)=✗, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 71.25 | 34.3 | |
| I-BERTBackbone Architecture=DeiT-T, FP32 Accuracy Baseline=72.21, Integer-Only (Int-Only)=✓, Query (Q) Quantization Granularity=▽, Key (K) Quantization Granularity=▽, Value (V) Quantization Granularity=▽2026.04 | 71.2 | 17.2 |