Accuracy Preservation
Cross-source consensus on Accuracy Preservation from 1 sources and 5 claims.
1 sources · 5 claims
Benefits
Comparisons
Evidence quality
Highlighted claims
- At 25% Threshold-L2 pruning, the pipeline reaches 81.7% top-1 accuracy and exceeds unpruned DeiT-S throughput while losing 0.5 percentage points of accuracy. — Dispatch-Aware Ragged Attention for Pruned Vision Transformers
- Numerical equivalence against padded PyTorch SDPA is verified across all ImageNet validation samples. — Dispatch-Aware Ragged Attention for Pruned Vision Transformers
- Top-1 predictions match exactly, with maximum absolute logit differences below 0.007 for all tested pruning methods. — Dispatch-Aware Ragged Attention for Pruned Vision Transformers
- At 50% pruning, throughput reaches 5,187 images per second with 78.7% top-1 accuracy. — Dispatch-Aware Ragged Attention for Pruned Vision Transformers
- The Triton backend gives higher throughput than padded PyTorch at the same accuracy on ImageNet-1K validation for DeiT-Small. — Dispatch-Aware Ragged Attention for Pruned Vision Transformers