Research map / Architectures and efficiency / Pruning and efficient tuning
FlexHiNM-GP: Flexible Hierarchical Pruning via Region Allocation and Channel Permutation
YI XIAODIE, Hayun Lee, Dongkun Shin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Gyro-Permutation (ablation) | FlexHiNM · Bert-Base QQP, 75% sparsity | F1: 84.78 -> 85.35 | Improved |
| add Gyro-Permutation (ablation) | FlexHiNM · Bert-Base SQuAD, 75% sparsity | F1: 88.04 -> 88.55 | Improved |
| add Gyro-Permutation (ablation) | FlexHiNM · Bert-Base SST-2, 75% sparsity | accuracy: 90.60 -> 91.65 | Improved |
| add adaptive boundary search (OptFlexHiNM) (ablation) | HiNM-GP · Deit-Base ImageNet-1K, 95% sparsity | top-1: 58.94 -> 61.43 | Improved |
| add adaptive boundary search (OptFlexHiNM) (ablation) | HiNM-GP · Deit-Base ImageNet-1K, 90% sparsity | top-1: 67.64 -> 69.26 | Improved |
| search boundaries adaptively instead of fixing ps=0.5 (ablation) | BalFlexHiNM · Deit-Base ImageNet-1K, 95% sparsity | top-1: 57.26 -> 61.43 | Improved |
| fix ps=0.5 (BalFlexHiNM) vs HiNM-GP (ablation) | HiNM-GP · Deit-Base ImageNet-1K, 95% sparsity | top-1: 58.94 -> 57.26 | Hurt |
| add flexible distribution with Hard Concrete (variant 1 vs 6) (ablation) | HiNM-GP · Deit-Base ImageNet-1K, 95% sparsity | top-1: 58.94 -> 61.43 | Improved |
| add flexible distribution with Hard Concrete (variant 1 vs 6) (ablation) | HiNM-GP · Deit-Base ImageNet-1K, 75% sparsity | top-1: 81.04 -> 81.13 | No clear effect |
| use Hard Concrete instead of Gumbel-Softmax with flexible distribution (variant 1 vs 2) (ablation) | FlexHiNM-GP · Deit-Base ImageNet-1K, 95% sparsity | top-1: 59.77 -> 61.43 | Improved |
| add flexible region allocation to HiNM-GP | HiNM-GP · Llama2-7B downstream, 75% sparsity | average accuracy: +1.39% | Improved |
| add Gyro-Permutation to HiNM-V | HiNM-V · Llama2-7B HellaSwag, 75% sparsity | accuracy: 40.63 -> 42.72 | Improved |
Closest papers by meaning
- DuoGPT: Training-free Dual Sparsity through Activation-aware Pruning in LLMs (NeurIPS 2025 · similarity 0.89)
- Balanced Token Pruning: Accelerating Vision Language Models Beyond Local Optimization (NeurIPS 2025 · similarity 0.88)
- Týr-the-Pruner: Structural Pruning LLMs via Global Sparsity Distribution Optimization (NeurIPS 2025 · similarity 0.88)
- DEPrune: Depth-wise Separable Convolution Pruning for Maximizing GPU Parallelism (NeurIPS 2024 · similarity 0.87)
- ICP: Immediate Compensation Pruning for Mid-to-high Sparsity (CVPR 2025 · similarity 0.87)
- REPrune: Channel Pruning via Kernel Representative Selection (AAAI 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
