CacheBlend: Fast Large Language Model Serving for RAG with Cached Knowledge Fusion
Reuse every retrieved chunk's KV cache anywhere, then recompute the ~15% of tokens that stitch cross-attention back.
Yao et al. · EuroSys 2025 · Serving. Read the paper ↗
A free, interactive, animated visual explainer of CacheBlend: Fast Large Language Model Serving for RAG with Cached Knowledge Fusion — every exhibit computed from the real formulas, with verbatim quotes from the source.
Questions
- What is CacheBlend: Fast Large Language Model Serving for RAG with Cached Knowledge Fusion?
- Reuse every retrieved chunk's KV cache anywhere, then recompute the ~15% of tokens that stitch cross-attention back.
- Who published CacheBlend: Fast Large Language Model Serving for RAG with Cached Knowledge Fusion, and where?
- Yao et al. — EuroSys 2025 (arXiv:2405.16444).
- Where can I find a visual explainer of CacheBlend: Fast Large Language Model Serving for RAG with Cached Knowledge Fusion?
- Right here — a free, interactive, animated walkthrough of the whole paper, with exhibits computed from the real formulas and verbatim quotes from the source.
Related explainers
- PagedAttention (vLLM)
- Efficiently Scaling Transformer Inference
- Mooncake: A KVCache-centric Disaggregated Architecture for LLM Serving
- Fast Inference from Transformers via Speculative Decoding
- EAGLE: Speculative Sampling Requires Rethinking Feature Uncertainty
- AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration
- DistServe: Disaggregating Prefill and Decoding for Goodput-optimized Large Language Model Serving
- Efficient Streaming Language Models with Attention Sinks