Research map / Architectures and efficiency / Pruning and efficient tuning
Hardware-Aware Compression with Random Operation Access Specific Tile (ROAST) Hashing
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Global memory sharing (GMS) instead of local memory sharing (LMS) at 10x compression (ablation) | BERT-12-12 · amazon-polarity | accuracy: 94.64 vs 93.49 | Helped |
| Global memory sharing (GMS) instead of local memory sharing (LMS) at 100x compression (ablation) | BERT-12-12 · amazon-polarity | accuracy: 93.9 vs 92.9 | Helped |
| ROAST 100x compression vs pruning 100x on BERT-2-2 | BERT-2-2 · yelp-polarity | accuracy: 90.8 vs 89 | Helped |
| ROAST compression vs pruning on BERT-2-2 at 100x | BERT-2-2 · amazon-polarity | accuracy: 93.8 vs 91.36 | Helped |
| Pruning 10x applied on top of ROAST 10x array on ResNet-9 (ablation) | ROAST-10x ResNet-9 · cifar-10 | accuracy: 94.06 vs 94.15 (ROAST-10x) | No clear effect |
| ROAST-MM kernel instead of HashedNet kernel for matrix multiplication (inference) | linear layers of compressed models · Inference, square matrices 512 to 20480 (Table 5) | average inference time (ms): avg 1.24 ms vs 56.03 ms for HashedNet at 512MB | Helped |
| ROAST-MM kernel instead of PyTorch full matmul | linear layers · Inference, square matrices (Table 5) | average inference time (ms): ROAST avg 1.24 ms vs PyTorch-MM 0.91 ms (1.34x slower) | Hurt |
Closest papers by meaning
- EchoKV: Efficient KV Cache Compression via Similarity-Based Reconstruction (NeurIPS 2026 · similarity 0.84)
- Network Memory Footprint Compression Through Jointly Learnable Codebooks and Mappings (ICLR 2024 · similarity 0.84)
- Spotlight Attention: Towards Efficient LLM Generation via Non-linear Hashing-based KV Cache Retrieval (NeurIPS 2025 · similarity 0.84)
- LSH-MoE: Communication-efficient MoE Training via Locality-Sensitive Hashing (NeurIPS 2024 · similarity 0.84)
- SqueezeAttention: 2D Management of KV-Cache in LLM Inference via Layer-wise Optimal Budget (ICLR 2025 · similarity 0.84)
- ZipCache: Accurate and Efficient KV Cache Quantization with Salient Token Identification (NeurIPS 2024 · similarity 0.83)
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.