Research map / Architectures and efficiency / Attention and transformers
FASA: FREQUENCY-AWARE SPARSE ATTENTION
Yifei Wang, Yueqi Wang, Zhenrui Yue, Huimin Zeng, Yong Wang, Ismini Lourentzou, Zhengzhong Tu, Xiangxiang Chu, Julian McAuley
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 |
|---|---|---|---|
| use FASA frequency-chunk token selection instead of SnapKV at 256 budget | KV-cache eviction (Llama3.2-3B) · LongBench-V1 | average score: 37.0 -> 41.5 | Improved |
| use FASA instead of Quest | KV-cache eviction (Qwen2.5-7B) · LongBench-V1 | average score: 31.4 -> 47.9 | Improved |
| use FASA instead of full KV cache | Qwen2.5-7B / Mistral-7B · LongBench-V1 | average score: 47.8 -> 47.9 (Qwen), 47.4 -> 47.8 (Mistral) | No clear effect |
| use FASA instead of full KV cache | Llama3.2-3B / Llama3.1-8B / Qwen2.5-14B-1M · LongBench-V1 | average score: 42.2 -> 41.5, 48.7 -> 48.2, 50.3 -> 49.2 | Hurt |
| use FASA instead of RKV on long-CoT reasoning | DeepSeek-R1-Distill-Llama-8B · MATH500 (budget 300) | accuracy: 24.0 -> 62.2 | Improved |
| use FASA at 10% budget vs FKV | DeepSeek-R1-Distill-Qwen-32B · MATH500 (budget 300) | accuracy: 92.6 -> 86.4 | Hurt |
| vary calibration window K (ablation) | FASA · LongBench Qasper | average score: 44.5 (K=128) vs 43.9 (K=512) | No clear effect |
| vary offline calibration dataset (ablation) | FASA · LongBench 2Wiki/Musi/Hqa/Qasp/MF-en/Nqa | score: CV .007 to .014 | No clear effect |
| add PyramidKV layer-wise budget allocation (ablation) | FASA · LongBench Qasper | score: 43.7 -> 44.4 (budget 256) | Improved |
| apply FASA to ALiBi model | Baichuan-13B-Chat · LongBench Qasper | score: 9.11 -> 7.80 | Hurt |
| apply FASA to Partial-RoPE model | DeepSeek-V2-Lite-Chat · LongBench | score: Qasper 33.18 -> 33.46, Lcc 63.40 -> 62.49 | Mixed |
| use Triton FASA-C kernel at 64K context | FASA · decoding latency | speedup: up to 2.56x | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Quest — QUEST: Query-Aware Sparsity for Efficient Long-Context LLM Inference (ICML 2024) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- OrthoRank: Token Selection via Sink Token Orthogonality for Efficient LLM inference (ICML 2025 · similarity 0.89)
- Punctuation-aware Hybrid Trainable Sparse Attention for Large Language Models (NeurIPS 2026 · similarity 0.88)
- QUEST: Query-Aware Sparsity for Efficient Long-Context LLM Inference (ICML 2024 · similarity 0.88)
- Block Sparse Flash Attention (NeurIPS 2026 · similarity 0.88)
- FlexPrefill: A Context-Aware Sparse Attention Mechanism for Efficient Long-Sequence Inference (ICLR 2025 · similarity 0.87)
- A Training-Free Sub-quadratic Cost Transformer Model Serving Framework with Hierarchically Pruned Attention (ICLR 2025 · 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.
