microsoft / RetrievalAttentionLinks
Scalable long-context LLM decoding that leverages sparsity—by treating the KV cache as a vector storage system.
☆69Updated this week
Alternatives and similar repositories for RetrievalAttention
Users that are interested in RetrievalAttention are comparing it to the libraries listed below
Sorting:
- [SIGMOD 2025] PQCache: Product Quantization-based KVCache for Long Context LLM Inference☆64Updated last month
- ☆125Updated 3 weeks ago
- [OSDI'24] Serving LLM-based Applications Efficiently with Semantic Variable☆170Updated 10 months ago
- kvcached: Elastic KV cache for dynamic GPU sharing and efficient multi-LLM inference.☆35Updated last week
- ☆109Updated 8 months ago
- Stateful LLM Serving☆77Updated 4 months ago
- [NeurIPS 2024] Efficient LLM Scheduling by Learning to Rank☆51Updated 9 months ago
- ☆67Updated last year
- InfiniGen: Efficient Generative Inference of Large Language Models with Dynamic KV Cache Management (OSDI'24)☆147Updated last year
- A framework for generating realistic LLM serving workloads☆51Updated last month
- ☆75Updated 9 months ago
- ☆209Updated this week
- Paper-reading notes for Berkeley OS prelim exam.☆14Updated 11 months ago
- ☆115Updated 9 months ago
- ☆78Updated 3 months ago
- NEO is a LLM inference engine built to save the GPU memory crisis by CPU offloading☆49Updated last month
- 16-fold memory access reduction with nearly no loss☆103Updated 4 months ago
- High performance Transformer implementation in C++.☆129Updated 6 months ago
- PyTorch library for cost-effective, fast and easy serving of MoE models.☆215Updated 3 weeks ago
- Modular and structured prompt caching for low-latency LLM inference☆97Updated 8 months ago
- ☆15Updated last month
- ArkVale: Efficient Generative LLM Inference with Recallable Key-Value Eviction (NIPS'24)☆42Updated 7 months ago
- A tiny yet powerful LLM inference system tailored for researching purpose. vLLM-equivalent performance with only 2k lines of code (2% of …☆238Updated last month
- PipeRAG: Fast Retrieval-Augmented Generation via Algorithm-System Co-design (KDD 2025)☆22Updated last year
- [ICLR 2025] TidalDecode: A Fast and Accurate LLM Decoding with Position Persistent Sparse Attention☆44Updated 3 months ago
- FlexFlow Serve: Low-Latency, High-Performance LLM Serving☆48Updated last week
- A lightweight design for computation-communication overlap.☆155Updated last month
- ☆54Updated last year
- nnScaler: Compiling DNN models for Parallel Training☆114Updated this week
- [DAC'25] Official implement of "HybriMoE: Hybrid CPU-GPU Scheduling and Cache Management for Efficient MoE Inference"☆64Updated last month