Research map / Large language models / Retrieval and knowledge
AdaCache: Adaptive Caching and Context Augmentation for Efficient LLM Serving
Zeng Zihao, Siyi Li, Xinyu Yan, Lei Xiao, Wei Yang Bryan Lim
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 adaptive context augmentation (ablation) | Prefix Caching · RAG QA, 3 models | TTFT: 1.65× | Improved |
| add adaptive context augmentation (ablation) | CacheBlend · RAG QA, 3 models | TTFT: 1.22× | Improved |
| add cache-aware selective recomputation (hierarchical cache) to ACA (ablation) | Prefix Caching + ACA · RAG QA, 3 models | TTFT: 1.76× | Improved |
| add cache-aware selective recomputation (hierarchical cache) to ACA (ablation) | CacheBlend + ACA · RAG QA, 3 models | TTFT: 1.23× | Improved |
| increase top-k retrieval from 2 to 8 (ablation) | Prefix Caching · RAG TTFT vs full recomputation | TTFT improvement: 1.76× -> 1.13× | Hurt |
| increase top-k retrieval from 2 to 8 (ablation) | AdaCache · RAG TTFT vs full recomputation | TTFT improvement: 2.93× -> 4.67× | Improved |
| skewed (head-heavy) context-length distribution (MMLU, TriviaQA) vs flatter (MMLU-Pro, SuperGPQA) (ablation) | AdaCache vs CacheBlend · MMLU/TriviaQA vs MMLU-Pro/SuperGPQA | TTFT reduction: 1.95× and 1.62× vs 1.25× and 1.14× | Improved |
| adaptive retrieval depth instead of static top-6 | ACA · top-6 retrieval cost | prefill tokens: 3,200 -> 1,280 | Improved |
| use AdaCache instead of full recomputation | LLM RAG serving · 6 QA datasets | accuracy | No clear effect |
Closest papers by meaning
- Training Free Exponential Context Extension via Cascading KV Cache (ICLR 2025 · similarity 0.89)
- Accelerating Inference of Retrieval-Augmented Generation via Sparse Context Selection (ICLR 2025 · similarity 0.89)
- Inference Scaling for Long-Context Retrieval Augmented Generation (ICLR 2025 · similarity 0.89)
- Long-Context LLMs Meet RAG: Overcoming Challenges for Long Inputs in RAG (ICLR 2025 · similarity 0.89)
- RazorAttention: Efficient KV Cache Compression Through Retrieval Heads (ICLR 2025 · similarity 0.89)
- SCBench: A KV Cache-Centric Analysis of Long-Context Methods (ICLR 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.
