gpustack

Performance-optimized AI inference on your GPUs. Unlock superior throughput by selecting and tuning engines like vLLM or SGLang.

Stars: 4544

Visit

GPUStack is an open-source GPU cluster manager designed for running large language models (LLMs). It supports a wide variety of hardware, scales with GPU inventory, offers lightweight Python package with minimal dependencies, provides OpenAI-compatible APIs, simplifies user and API key management, enables GPU metrics monitoring, and facilitates token usage and rate metrics tracking. The tool is suitable for managing GPU clusters efficiently and effectively.

README:

English | 简体中文 | 日本語

Overview

GPUStack is an open-source GPU cluster manager designed for efficient AI model deployment. It lets you run models efficiently on your own GPU hardware by choosing the best inference engines, scheduling GPU resources, analyzing model architectures, and automatically configuring deployment parameters.

The following figure shows how GPUStack delivers improved inference throughput over the unoptimized vLLM baseline:

For detailed benchmarking methods and results, visit our Inference Performance Lab.

Tested Inference Engines, GPUs, and Models

GPUStack uses a plug-in architecture that makes it easy to add new AI models, inference engines, and GPU hardware. We work closely with partners and the open-source community to test and optimize emerging models across different inference engines and GPUs. Below is the current list of supported inference engines, GPUs, and models, which will continue to expand over time.

Tested Inference Engines:

vLLM
SGLang
TensorRT-LLM
MindIE

Tested GPUs:

NVIDIA A100
NVIDIA H100/H200
Ascend 910B

Tuned Models:

Qwen3
gpt-oss
GLM-4.5-Air
GLM-4.x
DeepSeek-R1
DeepSeek-V3.2

Architecture

GPUStack enables development teams, IT organizations, and service providers to deliver Model-as-a-Service at scale. It supports industry-standard APIs for LLM, voice, image, and video models. The platform includes built-in user authentication and access control, real-time monitoring of GPU performance and utilization, and detailed metering of token usage and API request rates.

The figure below illustrates how a single GPUStack server can manage multiple GPU clusters across both on-premises and cloud environments. The GPUStack scheduler allocates GPUs to maximize resource utilization and selects the appropriate inference engines for optimal performance. Administrators also gain full visibility into system health and metrics through integrated Grafana and Prometheus dashboards.

GPUStack provides a powerful framework for deploying AI models. Its core features include:

Multi-Cluster GPU Management. Manages GPU clusters across multiple environments. This includes on-premises servers, Kubernetes clusters, and cloud providers.
Pluggable Inference Engines. Automatically configures high-performance inference engines such as vLLM, SGLang, and TensorRT-LLM. You can also add custom inference engines as needed.
Performance-Optimized Configurations. Offers pre-tuned modes for low latency or high throughput. GPUStack supports extended KV cache systems like LMCache and HiCache to reduce TTFT. It also includes built-in support for speculative decoding methods such as EAGLE3, MTP, and N-grams.
Enterprise-Grade Operations. Offers support for automated failure recovery, load balancing, monitoring, authentication, and access control.

Quick Start

Prerequisites

A node with at least one NVIDIA GPU. For other GPU types, please check the guidelines in the GPUStack UI when adding a worker, or refer to the Installation documentation for more details.
Ensure the NVIDIA driver, Docker and NVIDIA Container Toolkit are installed on the worker node.
(Optional) A CPU node for hosting the GPUStack server. The GPUStack server does not require a GPU and can run on a CPU-only machine. Docker must be installed. Docker Desktop (for Windows and macOS) is also supported. If no dedicated CPU node is available, the GPUStack server can be installed on the same machine as a GPU worker node.
Only Linux is supported for GPUStack worker nodes. If you use Windows, consider using WSL2 and avoid using Docker Desktop. macOS is not supported for GPUStack worker nodes.

Install GPUStack

Run the following command to install and start the GPUStack server using Docker:

sudo docker run -d --name gpustack \
    --restart unless-stopped \
    -p 80:80 \
    --volume gpustack-data:/var/lib/gpustack \
    gpustack/gpustack

Alternative: Use Quay Container Registry Mirror

If you cannot pull images from Docker Hub or the download is very slow, you can use our Quay.io mirror by pointing your registry to quay.io:

sudo docker run -d --name gpustack \
    --restart unless-stopped \
    -p 80:80 \
    --volume gpustack-data:/var/lib/gpustack \
    quay.io/gpustack/gpustack \
    --system-default-container-registry quay.io

Check the GPUStack startup logs:

sudo docker logs -f gpustack

After GPUStack starts, run the following command to get the default admin password:

sudo docker exec gpustack cat /var/lib/gpustack/initial_admin_password

Open your browser and navigate to http://your_host_ip to access the GPUStack UI. Use the default username admin and the password you retrieved above to log in.

Set Up a GPU Cluster

On the GPUStack UI, navigate to the Clusters page.
Click the Add Cluster button.
Select Docker as the cluster provider.
Fill in the Name and Description fields for the new cluster, then click the Save button.

Follow the UI guidelines to configure the new worker node. You will need to run a Docker command on the worker node to connect it to the GPUStack server. The command will look similar to the following:

sudo docker run -d --name gpustack-worker \
      --restart=unless-stopped \
      --privileged \
      --network=host \
      --volume /var/run/docker.sock:/var/run/docker.sock \
      --volume gpustack-data:/var/lib/gpustack \
      --runtime nvidia \
      gpustack/gpustack \
      --server-url http://your_gpustack_server_url \
      --token your_worker_token \
      --advertise-address 192.168.1.2

Execute the command on the worker node to connect it to the GPUStack server.
After the worker node connects successfully, it will appear on the Workers page in the GPUStack UI.

Deploy a Model

Navigate to the Catalog page in the GPUStack UI.
Select the Qwen3 0.6B model from the list of available models.
After the deployment compatibility checks pass, click the Save button to deploy the model.

GPUStack will start downloading the model files and deploying the model. When the deployment status shows Running, the model has been deployed successfully.

Click Playground - Chat in the navigation menu, check that the model qwen3-0.6b is selected from the top-right Model dropdown. Now you can chat with the model in the UI playground.

Use the model via API

Hover over the user avatar and navigate to the API Keys page, then click the New API Key button.
Fill in the Name and click the Save button.
Copy the generated API key and save it somewhere safe. Please note that you can only see it once on creation.
You can now use the API key to access the OpenAI-compatible API endpoints provided by GPUStack. For example, use curl as the following:

# Replace `your_api_key` and `your_gpustack_server_url`
# with your actual API key and GPUStack server URL.
export GPUSTACK_API_KEY=your_api_key
curl http://your_gpustack_server_url/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $GPUSTACK_API_KEY" \
  -d '{
    "model": "qwen3-0.6b",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "Tell me a joke."
      }
    ],
    "stream": true
  }'

Documentation

Please see the official docs site for complete documentation.

Build

Install Python (version 3.10 to 3.12).
Run make build.

You can find the built wheel package in dist directory.

Contributing

Please read the Contributing Guide if you're interested in contributing to GPUStack.

Join Community

Any issues or have suggestions, feel free to join our Community for support.

License

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at LICENSE file for details.

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

For Tasks:

Click tags to check more tools for each tasks

manage gpu clusters monitor gpu performance track token usage scale gpu operations chat with language models

For Jobs:

data scientist machine learning engineer ai researcher system administrator software developer

Alternative AI tools for gpustack

Similar Open Source Tools

gpustack

github

: 4.5k

torchchat

torchchat is a codebase showcasing the ability to run large language models (LLMs) seamlessly. It allows running LLMs using Python in various environments such as desktop, server, iOS, and Android. The tool supports running models via PyTorch, chatting, generating text, running chat in the browser, and running models on desktop/server without Python. It also provides features like AOT Inductor for faster execution, running in C++ using the runner, and deploying and running on iOS and Android. The tool supports popular hardware and OS including Linux, Mac OS, Android, and iOS, with various data types and execution modes available.

github

: 3.5k

SWELancer-Benchmark

SWE-Lancer is a benchmark repository containing datasets and code for the paper 'SWE-Lancer: Can Frontier LLMs Earn $1 Million from Real-World Freelance Software Engineering?'. It provides instructions for package management, building Docker images, configuring environment variables, and running evaluations. Users can use this tool to assess the performance of language models in real-world freelance software engineering tasks.

github

: 1.1k

bedrock-claude-chatbot

Bedrock Claude ChatBot is a Streamlit application that provides a conversational interface for users to interact with various Large Language Models (LLMs) on Amazon Bedrock. Users can ask questions, upload documents, and receive responses from the AI assistant. The app features conversational UI, document upload, caching, chat history storage, session management, model selection, cost tracking, logging, and advanced data analytics tool integration. It can be customized using a config file and is extensible for implementing specialized tools using Docker containers and AWS Lambda. The app requires access to Amazon Bedrock Anthropic Claude Model, S3 bucket, Amazon DynamoDB, Amazon Textract, and optionally Amazon Elastic Container Registry and Amazon Athena for advanced analytics features.

github

: 95

mosec

Mosec is a high-performance and flexible model serving framework for building ML model-enabled backend and microservices. It bridges the gap between any machine learning models you just trained and the efficient online service API. * **Highly performant** : web layer and task coordination built with Rust 🦀, which offers blazing speed in addition to efficient CPU utilization powered by async I/O * **Ease of use** : user interface purely in Python 🐍, by which users can serve their models in an ML framework-agnostic manner using the same code as they do for offline testing * **Dynamic batching** : aggregate requests from different users for batched inference and distribute results back * **Pipelined stages** : spawn multiple processes for pipelined stages to handle CPU/GPU/IO mixed workloads * **Cloud friendly** : designed to run in the cloud, with the model warmup, graceful shutdown, and Prometheus monitoring metrics, easily managed by Kubernetes or any container orchestration systems * **Do one thing well** : focus on the online serving part, users can pay attention to the model optimization and business logic

github

: 834

geti-sdk

The Intel® Geti™ SDK is a python package that enables teams to rapidly develop AI models by easing the complexities of model development and enhancing collaboration between teams. It provides tools to interact with an Intel® Geti™ server via the REST API, allowing for project creation, downloading, uploading, deploying for local inference with OpenVINO, setting project and model configuration, launching and monitoring training jobs, and media upload and prediction. The SDK also includes tutorial-style Jupyter notebooks demonstrating its usage.

github

: 74

geti-sdk

The Intel® Geti™ SDK is a python package that enables teams to rapidly develop AI models by easing the complexities of model development and fostering collaboration. It provides tools to interact with an Intel® Geti™ server via the REST API, allowing for project creation, downloading, uploading, deploying for local inference with OpenVINO, configuration management, training job monitoring, media upload, and prediction. The repository also includes tutorial-style Jupyter notebooks demonstrating SDK usage.

github

: 78

vector-inference

This repository provides an easy-to-use solution for running inference servers on Slurm-managed computing clusters using vLLM. All scripts in this repository run natively on the Vector Institute cluster environment. Users can deploy models as Slurm jobs, check server status and performance metrics, and shut down models. The repository also supports launching custom models with specific configurations. Additionally, users can send inference requests and set up an SSH tunnel to run inference from a local device.

github

: 77

unitycatalog

Unity Catalog is an open and interoperable catalog for data and AI, supporting multi-format tables, unstructured data, and AI assets. It offers plugin support for extensibility and interoperates with Delta Sharing protocol. The catalog is fully open with OpenAPI spec and OSS implementation, providing unified governance for data and AI with asset-level access control enforced through REST APIs.

github

: 2.8k

generative-ai-sagemaker-cdk-demo

This repository showcases how to deploy generative AI models from Amazon SageMaker JumpStart using the AWS CDK. Generative AI is a type of AI that can create new content and ideas, such as conversations, stories, images, videos, and music. The repository provides a detailed guide on deploying image and text generative AI models, utilizing pre-trained models from SageMaker JumpStart. The web application is built on Streamlit and hosted on Amazon ECS with Fargate. It interacts with the SageMaker model endpoints through Lambda functions and Amazon API Gateway. The repository also includes instructions on setting up the AWS CDK application, deploying the stacks, using the models, and viewing the deployed resources on the AWS Management Console.

github

: 65

contoso-chat

Contoso Chat is a Python sample demonstrating how to build, evaluate, and deploy a retail copilot application with Azure AI Studio using Promptflow with Prompty assets. The sample implements a Retrieval Augmented Generation approach to answer customer queries based on the company's product catalog and customer purchase history. It utilizes Azure AI Search, Azure Cosmos DB, Azure OpenAI, text-embeddings-ada-002, and GPT models for vectorizing user queries, AI-assisted evaluation, and generating chat responses. By exploring this sample, users can learn to build a retail copilot application, define prompts using Prompty, design, run & evaluate a copilot using Promptflow, provision and deploy the solution to Azure using the Azure Developer CLI, and understand Responsible AI practices for evaluation and content safety.

github

: 431

ray-llm

RayLLM (formerly known as Aviary) is an LLM serving solution that makes it easy to deploy and manage a variety of open source LLMs, built on Ray Serve. It provides an extensive suite of pre-configured open source LLMs, with defaults that work out of the box. RayLLM supports Transformer models hosted on Hugging Face Hub or present on local disk. It simplifies the deployment of multiple LLMs, the addition of new LLMs, and offers unique autoscaling support, including scale-to-zero. RayLLM fully supports multi-GPU & multi-node model deployments and offers high performance features like continuous batching, quantization and streaming. It provides a REST API that is similar to OpenAI's to make it easy to migrate and cross test them. RayLLM supports multiple LLM backends out of the box, including vLLM and TensorRT-LLM.

github

: 1.1k

ai-starter-kit

SambaNova AI Starter Kits is a collection of open-source examples and guides designed to facilitate the deployment of AI-driven use cases for developers and enterprises. The kits cover various categories such as Data Ingestion & Preparation, Model Development & Optimization, Intelligent Information Retrieval, and Advanced AI Capabilities. Users can obtain a free API key using SambaNova Cloud or deploy models using SambaStudio. Most examples are written in Python but can be applied to any programming language. The kits provide resources for tasks like text extraction, fine-tuning embeddings, prompt engineering, question-answering, image search, post-call analysis, and more.

github

: 215

torchtitan

Torchtitan is a PyTorch native platform designed for rapid experimentation and large-scale training of generative AI models. It provides a flexible foundation for developers to build upon with extension points for creating custom extensions. The tool showcases PyTorch's latest distributed training features and supports pretraining Llama 3.1 LLMs of various sizes. It offers key features like multi-dimensional parallelisms, FSDP2 with per-parameter sharding, Tensor Parallel, Pipeline Parallel, and more. Users can contribute to the tool through the experiments folder and core contributions guidelines. Installation can be done from source, nightly builds, or stable releases. The tool also supports training Llama 3.1 models and provides guidance on starting a training run and multi-node training. Citation information is available for referencing the tool in academic work, and the source code is under a BSD 3 license.

github

: 5.1k

WindowsAgentArena

Windows Agent Arena (WAA) is a scalable Windows AI agent platform designed for testing and benchmarking multi-modal, desktop AI agents. It provides researchers and developers with a reproducible and realistic Windows OS environment for AI research, enabling testing of agentic AI workflows across various tasks. WAA supports deploying agents at scale using Azure ML cloud infrastructure, allowing parallel running of multiple agents and delivering quick benchmark results for hundreds of tasks in minutes.

github

: 147

verifAI

VerifAI is a document-based question-answering system that addresses hallucinations in generative large language models and search engines. It retrieves relevant documents, generates answers with references, and verifies answers for accuracy. The engine uses generative search technology and a verification model to ensure no misinformation. VerifAI supports various document formats and offers user registration with a React.js interface. It is open-source and designed to be user-friendly, making it accessible for anyone to use.

github

: 54

For similar tasks

gpustack

github

: 4.5k

tensor-fusion

Tensor Fusion is a state-of-the-art GPU virtualization and pooling solution designed to optimize GPU cluster utilization. It offers features like fractional virtual GPU, remote GPU sharing, GPU-first scheduling, GPU oversubscription, GPU pooling, monitoring, live migration, and more. The tool aims to enhance GPU utilization efficiency and streamline AI infrastructure management for organizations.

github

: 69

CursorLens

Cursor Lens is an open-source tool that acts as a proxy between Cursor and various AI providers, logging interactions and providing detailed analytics to help developers optimize their use of AI in their coding workflow. It supports multiple AI providers, captures and logs all requests, provides visual analytics on AI usage, allows users to set up and switch between different AI configurations, offers real-time monitoring of AI interactions, tracks token usage, estimates costs based on token usage and model pricing. Built with Next.js, React, PostgreSQL, Prisma ORM, Vercel AI SDK, Tailwind CSS, and shadcn/ui components.

github

: 73

FDAbench

FDABench is a benchmark tool designed for evaluating data agents' reasoning ability over heterogeneous data in analytical scenarios. It offers 2,007 tasks across various data sources, domains, difficulty levels, and task types. The tool provides ready-to-use data agent implementations, a DAG-based evaluation system, and a framework for agent-expert collaboration in dataset generation. Key features include data agent implementations, comprehensive evaluation metrics, multi-database support, different task types, extensible framework for custom agent integration, and cost tracking. Users can set up the environment using Python 3.10+ on Linux, macOS, or Windows. FDABench can be installed with a one-command setup or manually. The tool supports API configuration for LLM access and offers quick start guides for database download, dataset loading, and running examples. It also includes features like dataset generation using the PUDDING framework, custom agent integration, evaluation metrics like accuracy and rubric score, and a directory structure for easy navigation.

github

: 53

easyclaw

EasyClaw is a desktop application that simplifies the usage of OpenClaw, a powerful agent runtime, by providing a user-friendly interface for non-programmers. Users can write rules in plain language, configure multiple LLM providers and messaging channels, manage API keys, and interact with the agent through a local web panel. The application ensures data privacy by keeping all information on the user's machine and offers features like natural language rules, multi-provider LLM support, Gemini CLI OAuth, proxy support, messaging integration, token tracking, speech-to-text, file permissions control, and more. EasyClaw aims to lower the barrier of entry for utilizing OpenClaw by providing a user-friendly cockpit for managing the engine.

github

: 67

For similar jobs

weave

Weave is a toolkit for developing Generative AI applications, built by Weights & Biases. With Weave, you can log and debug language model inputs, outputs, and traces; build rigorous, apples-to-apples evaluations for language model use cases; and organize all the information generated across the LLM workflow, from experimentation to evaluations to production. Weave aims to bring rigor, best-practices, and composability to the inherently experimental process of developing Generative AI software, without introducing cognitive overhead.

github

: 1.1k

LLMStack

LLMStack is a no-code platform for building generative AI agents, workflows, and chatbots. It allows users to connect their own data, internal tools, and GPT-powered models without any coding experience. LLMStack can be deployed to the cloud or on-premise and can be accessed via HTTP API or triggered from Slack or Discord.

github

: 1.5k

VisionCraft

The VisionCraft API is a free API for using over 100 different AI models. From images to sound.

github

: 94

kaito

Kaito is an operator that automates the AI/ML inference model deployment in a Kubernetes cluster. It manages large model files using container images, avoids tuning deployment parameters to fit GPU hardware by providing preset configurations, auto-provisions GPU nodes based on model requirements, and hosts large model images in the public Microsoft Container Registry (MCR) if the license allows. Using Kaito, the workflow of onboarding large AI inference models in Kubernetes is largely simplified.

github

: 405

PyRIT

PyRIT is an open access automation framework designed to empower security professionals and ML engineers to red team foundation models and their applications. It automates AI Red Teaming tasks to allow operators to focus on more complicated and time-consuming tasks and can also identify security harms such as misuse (e.g., malware generation, jailbreaking), and privacy harms (e.g., identity theft). The goal is to allow researchers to have a baseline of how well their model and entire inference pipeline is doing against different harm categories and to be able to compare that baseline to future iterations of their model. This allows them to have empirical data on how well their model is doing today, and detect any degradation of performance based on future improvements.

github

: 3.5k

tabby

Tabby is a self-hosted AI coding assistant, offering an open-source and on-premises alternative to GitHub Copilot. It boasts several key features: * Self-contained, with no need for a DBMS or cloud service. * OpenAPI interface, easy to integrate with existing infrastructure (e.g Cloud IDE). * Supports consumer-grade GPUs.

github

: 32.9k

spear

SPEAR (Simulator for Photorealistic Embodied AI Research) is a powerful tool for training embodied agents. It features 300 unique virtual indoor environments with 2,566 unique rooms and 17,234 unique objects that can be manipulated individually. Each environment is designed by a professional artist and features detailed geometry, photorealistic materials, and a unique floor plan and object layout. SPEAR is implemented as Unreal Engine assets and provides an OpenAI Gym interface for interacting with the environments via Python.

github

: 224

Magick

Magick is a groundbreaking visual AIDE (Artificial Intelligence Development Environment) for no-code data pipelines and multimodal agents. Magick can connect to other services and comes with nodes and templates well-suited for intelligent agents, chatbots, complex reasoning systems and realistic characters.

github

: 675