Research map / Architectures and efficiency / Pruning and efficient tuning
GRIM: A General, Real-Time Deep Learning Inference Framework for Mobile Devices Based on Fine-Grained Structured Weight Sparsity
Wei Niu, Zhengang Li, Xiaolong Ma, Peiyan Dong, Gang Zhou, Xuehai Qian, Xue Lin, Yanzhi Wang, Bin Ren
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 |
|---|---|---|---|
| replace pattern-based pruning (PatDNN) with BCR pruning at higher rate | VGG-16 · CIFAR-10 | accuracy: 93.7 -> 93.8 (19.8x -> 35.7x) | Improved |
| replace pattern-based pruning (PatDNN) with BCR pruning | ResNet-18 · CIFAR-10 27.0x | accuracy: 92.1 -> 93.9 | Improved |
| replace filter pruning (DCP) with BCR pruning | ResNet-18 · CIFAR-10 27.0x | accuracy: 68.2 -> 93.9 | Improved |
| raise BCR pruning rate from 35.7x to 71.3x | GRIM VGG-16 · CIFAR-10 | accuracy: 93.8 -> 93.1 | Hurt |
| raise BCR pruning rate from 4.0x to 8.0x | GRIM ResNet-18 · ImageNet | top-1 accuracy: 69.6 -> 67.2 | Hurt |
| raise BCR pruning rate from 19.5x to 245.5x | GRIM GRU · TIMIT | phone error rate: 18.8 -> 24.2 | Hurt |
| use BCR pruning instead of block-circulant (C-LSTM) at 16x | GRU · TIMIT | phone error rate: 25.48 -> 18.8 | Improved |
| add matrix reorder to BCR-pruned code (CPU) (ablation) | GRIM · VGG-16 ImageNet layers | speedup: 1.21x to 1.88x speedup | Improved |
| add matrix reorder to BCR-pruned code (GPU) (ablation) | GRIM · VGG-16 ImageNet layers | speedup: 1.30x to 2.88x speedup | Improved |
| add register load redundancy elimination (CPU) (ablation) | GRIM · VGG-16 ImageNet layers | speedup: 1.11x to 3.51x extra speedup | Improved |
| add register load redundancy elimination (GPU) (ablation) | GRIM · VGG-16 ImageNet layers | speedup: 0.89x to 1.90x extra speedup | Mixed |
| add auto-tuning (CPU) (ablation) | GRIM · VGG-16 ImageNet layers | speedup: 0.31x to 1.45x additional speedup | Mixed |
Closest papers by meaning
- PCONV: The Missing but Desirable Sparsity in DNN Weight Pruning for Real-Time Execution on Mobile Devices (AAAI 2020 · similarity 0.92)
- An Image Enhancing Pattern-based Sparsity for Real-time Inference on Mobile Devices (ECCV 2020 · similarity 0.92)
- RT3D: Achieving Real-Time Execution of 3D Convolutional Neural Networks on Mobile Devices (AAAI 2021 · similarity 0.90)
- Multi-Granularity Pruning for Model Acceleration on Mobile Devices (ECCV 2022 · similarity 0.89)
- R-Sparse: Rank-Aware Activation Sparsity for Efficient LLM Inference (ICLR 2025 · similarity 0.85)
- DuoGPT: Training-free Dual Sparsity through Activation-aware Pruning in LLMs (NeurIPS 2025 · similarity 0.85)
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.
