
InfiniStore
A distributed KV store for disaggregated LLM inference
Stars: 52

InfiniStore is an open-source high-performance KV store designed to support LLM Inference clusters. It provides high-performance and low-latency KV cache transfer and reuse among inference nodes. In addition to inference clusters, it can be used as a standalone KV store for integration with LLM training or inference services. InfiniStore is currently integrated with vLLM via LMCache and is in progress for integration with SGLang and other inference engines.
README:
InfiniStore is an open-source high-performance KV store. It's designed to support LLM Inference clusters, whether the cluster is in prefill-decoding disaggregation mode or not. InfiniStore provides high-performance and low-latency KV cache transfer and KV cache reuse among inference nodes in the cluster.
In addition to inference clusters, InfiniStore can also be used as a standalone KV store to integrate with any other LLM training or inference services.
There are two major scenarios how InfiniStore supports :
- Prefill-Decoding disaggregation clusters: in such mode inference workloads are separated into two node pools: prefill nodes and decoding nodes. InfiniStore enables KV cache transfer among these two types of nodes, and also KV cache reuse.
- Non-disaggregated clusters: in such mode prefill and decoding workloads are mixed on every node. Infinistore serves as an extra large KV cache pool in addition to GPU cache and local CPU cache, and also enables cross-node KV cache reuse.
Currently InfiniStore has been integrated with vLLM. The integration is done via LMCache for the flexibility purpose.
Integration with SGLang and other inference engines are in progress.
Most users just need to deploy and run InfiniStore, and they don't need to understand how InfiniStore works internally. For these users, PIP is the recommended way to install:
pip install InfiniStore
For users who need to understand how InfiniStore code works or make code contributions to InfiniStore, it's recommended to install from source code:
apt install libuv1-dev
apt install libflatbuffers-dev
apt install libspdlog-dev libfmt-dev
apt install ibverbs-utils libibverbs-dev
apt install libboost-dev libboost-stacktrace-dev
pip install -e .
pip install pre-commit
pre-commit install
After installation, either from PIP or from source code, run the following command to verify your installation is successful:
InfiniStore --manage-port 8088
curl http://127.0.0.1:8088/selftest
- Start InfiniStore Server
The first step is to start an InfiniStore server. The server can be running on a GPU machine or a CPU machine.
Your server machine may be equipped with TCP network or RDMA network. The command line to start a server varies depending on the network configurations:
For TCP/IP Network:
InfiniStore --service-port 12345
For RDMA(RoCE):
InfiniStore --service-port 12345 --dev-name mlx5_0 --link-type Ethernet
For RDMA(Infiniband):
InfiniStore --service-port 12345 --dev-name mlx5_0 --link-type IB
- Run InfiniStore Client
Check the following example code to run an InfiniStore client:
InfiniStore/example/client.py
InfiniStore/example/client_async.py
InfiniStore/example/client_async_single.py
As illustrated in the previous section, InfiniStore enables different functionalities in a vLLM cluster: KV cache transfer between prefill nodes and decoding nodes, extended KV cache pool, cross-node KV cache reuse, etc.
The setup will varies depending on the specific vLLM cluster configurations. But usually it requires the following installations:
- Install vLLM on all nodes
- Install LMCache on all nodes
- Install InfiniStore on all nodes
Because this setup is a complicated process, we've made a separate demo repo for the PD disaggregation setup.
Please refer to the repo for the details: https://github.com/bytedance-iaas/splitwise-demos
InfiniStore is an open-source project and community. We welcome anyone who is interested in helping improve InfiniStore, whether code contributions, document contributions or any other contributions.
If you are submitting a code change, run the following unit test and pre-commit check to ensure your code change doesn't break existing features before submitting the PR:
-
Make your code changes
Just clone this repo, make code changes according to your feature design.
-
Run Unit Tests
pytest InfiniStore/test_InfiniStore.py
- Run Pre-commit Checks
pre-commit run --all-files
- Submit PR
If you code change passes both unit tests and pre-commit checks, submit the PR.
For Tasks:
Click tags to check more tools for each tasksFor Jobs:
Alternative AI tools for InfiniStore
Similar Open Source Tools

InfiniStore
InfiniStore is an open-source high-performance KV store designed to support LLM Inference clusters. It provides high-performance and low-latency KV cache transfer and reuse among inference nodes. In addition to inference clusters, it can be used as a standalone KV store for integration with LLM training or inference services. InfiniStore is currently integrated with vLLM via LMCache and is in progress for integration with SGLang and other inference engines.

llama_deploy
llama_deploy is an async-first framework for deploying, scaling, and productionizing agentic multi-service systems based on workflows from llama_index. It allows building workflows in llama_index and deploying them seamlessly with minimal changes to code. The system includes services endlessly processing tasks, a control plane managing state and services, an orchestrator deciding task handling, and fault tolerance mechanisms. It is designed for high-concurrency scenarios, enabling real-time and high-throughput applications.

pathway
Pathway is a Python data processing framework for analytics and AI pipelines over data streams. It's the ideal solution for real-time processing use cases like streaming ETL or RAG pipelines for unstructured data. Pathway comes with an **easy-to-use Python API** , allowing you to seamlessly integrate your favorite Python ML libraries. Pathway code is versatile and robust: **you can use it in both development and production environments, handling both batch and streaming data effectively**. The same code can be used for local development, CI/CD tests, running batch jobs, handling stream replays, and processing data streams. Pathway is powered by a **scalable Rust engine** based on Differential Dataflow and performs incremental computation. Your Pathway code, despite being written in Python, is run by the Rust engine, enabling multithreading, multiprocessing, and distributed computations. All the pipeline is kept in memory and can be easily deployed with **Docker and Kubernetes**. You can install Pathway with pip: `pip install -U pathway` For any questions, you will find the community and team behind the project on Discord.

agentok
Agentok Studio is a visual tool built for AutoGen, a cutting-edge agent framework from Microsoft and various contributors. It offers intuitive visual tools to simplify the construction and management of complex agent-based workflows. Users can create workflows visually as graphs, chat with agents, and share flow templates. The tool is designed to streamline the development process for creators and developers working on next-generation Multi-Agent Applications.

ztachip
ztachip is a RISCV accelerator designed for vision and AI edge applications, offering up to 20-50x acceleration compared to non-accelerated RISCV implementations. It features an innovative tensor processor hardware to accelerate various vision tasks and TensorFlow AI models. ztachip introduces a new tensor programming paradigm for massive processing/data parallelism. The repository includes technical documentation, code structure, build procedures, and reference design examples for running vision/AI applications on FPGA devices. Users can build ztachip as a standalone executable or a micropython port, and run various AI/vision applications like image classification, object detection, edge detection, motion detection, and multi-tasking on supported hardware.

langchainjs-quickstart-demo
Discover the journey of building a generative AI application using LangChain.js and Azure. This demo explores the development process from idea to production, using a RAG-based approach for a Q&A system based on YouTube video transcripts. The application allows to ask text-based questions about a YouTube video and uses the transcript of the video to generate responses. The code comes in two versions: local prototype using FAISS and Ollama with LLaMa3 model for completion and all-minilm-l6-v2 for embeddings, and Azure cloud version using Azure AI Search and GPT-4 Turbo model for completion and text-embedding-3-large for embeddings. Either version can be run as an API using the Azure Functions runtime.

llm-engine
Scale's LLM Engine is an open-source Python library, CLI, and Helm chart that provides everything you need to serve and fine-tune foundation models, whether you use Scale's hosted infrastructure or do it in your own cloud infrastructure using Kubernetes.

langdrive
LangDrive is an open-source AI library that simplifies training, deploying, and querying open-source large language models (LLMs) using private data. It supports data ingestion, fine-tuning, and deployment via a command-line interface, YAML file, or API, with a quick, easy setup. Users can build AI applications such as question/answering systems, chatbots, AI agents, and content generators. The library provides features like data connectors for ingestion, fine-tuning of LLMs, deployment to Hugging Face hub, inference querying, data utilities for CRUD operations, and APIs for model access. LangDrive is designed to streamline the process of working with LLMs and making AI development more accessible.

vector-vein
VectorVein is a no-code AI workflow software inspired by LangChain and langflow, aiming to combine the powerful capabilities of large language models and enable users to achieve intelligent and automated daily workflows through simple drag-and-drop actions. Users can create powerful workflows without the need for programming, automating all tasks with ease. The software allows users to define inputs, outputs, and processing methods to create customized workflow processes for various tasks such as translation, mind mapping, summarizing web articles, and automatic categorization of customer reviews.

uwazi
Uwazi is a flexible database application designed for capturing and organizing collections of information, with a focus on document management. It is developed and supported by HURIDOCS, benefiting human rights organizations globally. The tool requires NodeJs, ElasticSearch, ICU Analysis Plugin, MongoDB, Yarn, and pdftotext for installation. It offers production and development installation guides, including Docker setup. Uwazi supports hot reloading, unit and integration testing with JEST, and end-to-end testing with Nightmare or Puppeteer. The system requirements include RAM, CPU, and disk space recommendations for on-premises and development usage.

generative-ai-application-builder-on-aws
The Generative AI Application Builder on AWS (GAAB) is a solution that provides a web-based management dashboard for deploying customizable Generative AI (Gen AI) use cases. Users can experiment with and compare different combinations of Large Language Model (LLM) use cases, configure and optimize their use cases, and integrate them into their applications for production. The solution is targeted at novice to experienced users who want to experiment and productionize different Gen AI use cases. It uses LangChain open-source software to configure connections to Large Language Models (LLMs) for various use cases, with the ability to deploy chat use cases that allow querying over users' enterprise data in a chatbot-style User Interface (UI) and support custom end-user implementations through an API.

lollms-webui
LoLLMs WebUI (Lord of Large Language Multimodal Systems: One tool to rule them all) is a user-friendly interface to access and utilize various LLM (Large Language Models) and other AI models for a wide range of tasks. With over 500 AI expert conditionings across diverse domains and more than 2500 fine tuned models over multiple domains, LoLLMs WebUI provides an immediate resource for any problem, from car repair to coding assistance, legal matters, medical diagnosis, entertainment, and more. The easy-to-use UI with light and dark mode options, integration with GitHub repository, support for different personalities, and features like thumb up/down rating, copy, edit, and remove messages, local database storage, search, export, and delete multiple discussions, make LoLLMs WebUI a powerful and versatile tool.

MARS5-TTS
MARS5 is a novel English speech model (TTS) developed by CAMB.AI, featuring a two-stage AR-NAR pipeline with a unique NAR component. The model can generate speech for various scenarios like sports commentary and anime with just 5 seconds of audio and a text snippet. It allows steering prosody using punctuation and capitalization in the transcript. Speaker identity is specified using an audio reference file, enabling 'deep clone' for improved quality. The model can be used via torch.hub or HuggingFace, supporting both shallow and deep cloning for inference. Checkpoints are provided for AR and NAR models, with hardware requirements of 750M+450M params on GPU. Contributions to improve model stability, performance, and reference audio selection are welcome.

vespa
Vespa is a platform that performs operations such as selecting a subset of data in a large corpus, evaluating machine-learned models over the selected data, organizing and aggregating it, and returning it, typically in less than 100 milliseconds, all while the data corpus is continuously changing. It has been in development for many years and is used on a number of large internet services and apps which serve hundreds of thousands of queries from Vespa per second.

FlowTest
FlowTestAI is the world’s first GenAI powered OpenSource Integrated Development Environment (IDE) designed for crafting, visualizing, and managing API-first workflows. It operates as a desktop app, interacting with the local file system, ensuring privacy and enabling collaboration via version control systems. The platform offers platform-specific binaries for macOS, with versions for Windows and Linux in development. It also features a CLI for running API workflows from the command line interface, facilitating automation and CI/CD processes.

Robyn
Robyn is an experimental, semi-automated and open-sourced Marketing Mix Modeling (MMM) package from Meta Marketing Science. It uses various machine learning techniques to define media channel efficiency and effectivity, explore adstock rates and saturation curves. Built for granular datasets with many independent variables, especially suitable for digital and direct response advertisers with rich data sources. Aiming to democratize MMM, make it accessible for advertisers of all sizes, and contribute to the measurement landscape.
For similar tasks

InfiniStore
InfiniStore is an open-source high-performance KV store designed to support LLM Inference clusters. It provides high-performance and low-latency KV cache transfer and reuse among inference nodes. In addition to inference clusters, it can be used as a standalone KV store for integration with LLM training or inference services. InfiniStore is currently integrated with vLLM via LMCache and is in progress for integration with SGLang and other inference engines.
For similar jobs

db2rest
DB2Rest is a modern low-code REST DATA API platform that simplifies the development of intelligent applications. It seamlessly integrates existing and new databases with language models (LMs/LLMs) and vector stores, enabling the rapid delivery of context-aware, reasoning applications without vendor lock-in.

mage-ai
Mage is an open-source data pipeline tool for transforming and integrating data. It offers an easy developer experience, engineering best practices built-in, and data as a first-class citizen. Mage makes it easy to build, preview, and launch data pipelines, and provides observability and scaling capabilities. It supports data integrations, streaming pipelines, and dbt integration.

airbyte
Airbyte is an open-source data integration platform that makes it easy to move data from any source to any destination. With Airbyte, you can build and manage data pipelines without writing any code. Airbyte provides a library of pre-built connectors that make it easy to connect to popular data sources and destinations. You can also create your own connectors using Airbyte's no-code Connector Builder or low-code CDK. Airbyte is used by data engineers and analysts at companies of all sizes to build and manage their data pipelines.

labelbox-python
Labelbox is a data-centric AI platform for enterprises to develop, optimize, and use AI to solve problems and power new products and services. Enterprises use Labelbox to curate data, generate high-quality human feedback data for computer vision and LLMs, evaluate model performance, and automate tasks by combining AI and human-centric workflows. The academic & research community uses Labelbox for cutting-edge AI research.

telemetry-airflow
This repository codifies the Airflow cluster that is deployed at workflow.telemetry.mozilla.org (behind SSO) and commonly referred to as "WTMO" or simply "Airflow". Some links relevant to users and developers of WTMO: * The `dags` directory in this repository contains some custom DAG definitions * Many of the DAGs registered with WTMO don't live in this repository, but are instead generated from ETL task definitions in bigquery-etl * The Data SRE team maintains a WTMO Developer Guide (behind SSO)

airflow
Apache Airflow (or simply Airflow) is a platform to programmatically author, schedule, and monitor workflows. When workflows are defined as code, they become more maintainable, versionable, testable, and collaborative. Use Airflow to author workflows as directed acyclic graphs (DAGs) of tasks. The Airflow scheduler executes your tasks on an array of workers while following the specified dependencies. Rich command line utilities make performing complex surgeries on DAGs a snap. The rich user interface makes it easy to visualize pipelines running in production, monitor progress, and troubleshoot issues when needed.

airbyte-platform
Airbyte is an open-source data integration platform that makes it easy to move data from any source to any destination. With Airbyte, you can build and manage data pipelines without writing any code. Airbyte provides a library of pre-built connectors that make it easy to connect to popular data sources and destinations. You can also create your own connectors using Airbyte's low-code Connector Development Kit (CDK). Airbyte is used by data engineers and analysts at companies of all sizes to move data for a variety of purposes, including data warehousing, data analysis, and machine learning.

chronon
Chronon is a platform that simplifies and improves ML workflows by providing a central place to define features, ensuring point-in-time correctness for backfills, simplifying orchestration for batch and streaming pipelines, offering easy endpoints for feature fetching, and guaranteeing and measuring consistency. It offers benefits over other approaches by enabling the use of a broad set of data for training, handling large aggregations and other computationally intensive transformations, and abstracting away the infrastructure complexity of data plumbing.