Research map / Architectures and efficiency / Attention and transformers
ManifoldKV: Training-Free KV Cache Compression via Euclidean Outlier Detection
Debajyoti Datta, Trishala Neeraj, Bibek Paudel, Vyom Sharma, Subhabrata Mukherjee
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 cosine scoring with L2 distance from centroid (ablation) | KeyDiff-style scorer · RULER standalone | accuracy (%): 52.8% -> 92.7% | Improved |
| replace L2 with L1 distance (ablation) | ManifoldKV · RULER standalone | accuracy (%): 92.7% -> 78.5% | Hurt |
| replace L2 with L-infinity distance (ablation) | ManifoldKV · RULER standalone | accuracy (%): 92.7% -> 71.2% | Hurt |
| use local sliding-window centroids (4K) instead of global centroid (ablation) | ManifoldKV · RULER 64K | accuracy (%): 35.2% -> 84.3% | Improved |
| shrink window from 16K to 4K (ablation) | WindowedManifoldKV · RULER 64K | accuracy (%): 82.4% -> 84.3% | Improved |
| shrink window from 4K to 2K (ablation) | WindowedManifoldKV · RULER 64K | accuracy (%): 84.3% -> 83.8% | Hurt |
| L2 scoring instead of cosine (KeyDiff) on 3-key retrieval | ManifoldKV · RULER multikey_3 (8K, 50% compression) | accuracy (%): 77.0 -> 92.4 | Improved |
| L2 scoring instead of cosine (KeyDiff) on 2-key retrieval | ManifoldKV · RULER multikey_2 (8K, 50% compression) | accuracy (%): 92.6 -> 99.8 | Improved |
| L2 scoring instead of cosine under AdaKV on overall RULER | ManifoldKV · RULER (4K-16K, AdaKV) | accuracy (%): 95.66% -> 95.73% | No clear effect |
| geometric scorer instead of attention-based SnapKV under AdaKV | ManifoldKV · RULER (4K-16K, AdaKV) | accuracy (%): 83.97% -> 95.73% | Improved |
| blend L2 with cosine (Hybrid 0.5) instead of windowing | ManifoldKV · RULER 64K | accuracy (%): 81.09% -> 79.14% | Hurt |
| increase context length from 4K to 64K with global centroid | Global ManifoldKV · RULER (Llama-3.1-8B) | accuracy (%): 95.7% -> 35.2% | Hurt |
Closest papers by meaning
- EchoKV: Efficient KV Cache Compression via Similarity-Based Reconstruction (NeurIPS 2026 · similarity 0.90)
- CompilerKV: Risk-Adaptive KV Cache Compression via Offline Experience Compilation (NeurIPS 2026 · similarity 0.89)
- PatchKV: Weight Space Compensation of KV Cache (NeurIPS 2026 · similarity 0.89)
- RazorAttention: Efficient KV Cache Compression Through Retrieval Heads (ICLR 2025 · similarity 0.89)
- ChunkKV: Semantic-Preserving KV Cache Compression for Efficient Long-Context LLM Inference (NeurIPS 2025 · similarity 0.89)
- Homogeneous Keys, Heterogeneous Values: Exploiting Local KV Cache Asymmetry for Long-Context LLMs (NeurIPS 2025 · similarity 0.89)
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.
