# dstack

> dstack is a unified control plane for GPU provisioning and orchestration that works with any GPU cloud, Kubernetes, or on-prem clusters.

## Getting started

- [Overview](https://dstack.ai/docs/index.md): Introduction to dstack and how it works
- [Installation](https://dstack.ai/docs/installation.md): How to install the dstack server and CLI
- [Quickstart](https://dstack.ai/docs/quickstart.md): Quick guide to creating fleets and submitting runs

## Concepts

- [Backends](https://dstack.ai/docs/concepts/backends.md): Configuring cloud providers and Kubernetes clusters
- [Fleets](https://dstack.ai/docs/concepts/fleets.md): Managing pools of compute instances
- [Dev environments](https://dstack.ai/docs/concepts/dev-environments.md): Provisioning remote instances for cloud-based development
- [Tasks](https://dstack.ai/docs/concepts/tasks.md): Running commands for training and batch processing
- [Services](https://dstack.ai/docs/concepts/services.md): Deploying models and web apps as endpoints
- [Volumes](https://dstack.ai/docs/concepts/volumes.md): Managing persistent data storage

### More

- [Gateways](https://dstack.ai/docs/concepts/gateways.md): Managing ingress traffic and endpoints for services
- [Secrets](https://dstack.ai/docs/concepts/secrets.md): Managing sensitive values and credentials
- [Projects](https://dstack.ai/docs/concepts/projects.md): Organizing teams and isolating resources
- [Metrics](https://dstack.ai/docs/concepts/metrics.md): Tracking and monitoring system metrics
- [Events](https://dstack.ai/docs/concepts/events.md): Auditing resource state changes and operations
- [Exports](https://dstack.ai/docs/concepts/exports.md): Exporting resources across projects

## Guides

- [CLI & API](https://dstack.ai/docs/guides/cli-api.md): How to use the dstack CLI and HTTP API
- [Server deployment](https://dstack.ai/docs/guides/server-deployment.md): Deploying the dstack server
- [Troubleshooting](https://dstack.ai/docs/guides/troubleshooting.md): Common issues and how to resolve them

### More

- [Protips](https://dstack.ai/docs/guides/protips.md): Tips and tricks for using dstack efficiently
- [Tenant isolation](https://dstack.ai/docs/guides/tenant-isolation.md): Restricting access to hosts managed by dstack
- [Upgrade](https://dstack.ai/docs/guides/upgrade.md): Upgrading to newer versions of dstack

### Migration

- [Slurm](https://dstack.ai/docs/guides/migration/slurm.md): This guide compares Slurm and dstack, and shows how to orchestrate equivalent GPU-based workloads using dstack.

## Examples


### Training

- [TRL](https://dstack.ai/docs/examples/training/trl.md): Fine-tuning Llama with TRL — single-node SFT with QLoRA, or distributed across multiple nodes with FSDP and DeepSpeed
- [Axolotl](https://dstack.ai/docs/examples/training/axolotl.md): Fine-tuning Llama models with Axolotl — single-node SFT with FSDP and QLoRA, or distributed across multiple nodes
- [Ray+RAGEN](https://dstack.ai/docs/examples/training/ray-ragen.md): Multi-node agent fine-tuning using RAGEN with Ray and verl for reinforcement learning
- [Miles](https://dstack.ai/docs/examples/training/miles.md): RL post-training Qwen2.5-32B with Miles, SGLang, Megatron-LM, and Ray across two 8xH100 nodes

### Clusters

- [AWS](https://dstack.ai/docs/examples/clusters/aws.md): High-performance distributed training on AWS using Elastic Fabric Adapter (EFA)
- [GCP](https://dstack.ai/docs/examples/clusters/gcp.md): Creating and using GPU clusters on GCP with GPUDirect-TCPX and RoCE support
- [Lambda](https://dstack.ai/docs/examples/clusters/lambda.md): Setting up Lambda clusters using Kubernetes or 1-Click Clusters with fast interconnect
- [Crusoe](https://dstack.ai/docs/examples/clusters/crusoe.md): Using Crusoe clusters with InfiniBand support via VMs or Kubernetes
- [Nebius](https://dstack.ai/docs/examples/clusters/nebius.md): Using Nebius clusters with InfiniBand support via VMs or Kubernetes
- [NCCL/RCCL tests](https://dstack.ai/docs/examples/clusters/nccl-rccl-tests.md): Running NCCL and RCCL tests to validate cluster network bandwidth

### Inference

- [SGLang](https://dstack.ai/docs/examples/inference/sglang.md): Deploying Qwen3.6-27B using SGLang on NVIDIA and AMD GPUs
- [Dynamo](https://dstack.ai/docs/examples/inference/dynamo.md): Deploying zai-org/GLM-4.5-Air-FP8 using NVIDIA Dynamo with Prefill-Decode disaggregation.
- [vLLM](https://dstack.ai/docs/examples/inference/vllm.md): Deploying Qwen3.6-27B using vLLM on NVIDIA and AMD GPUs
- [NIM](https://dstack.ai/docs/examples/inference/nim.md): Deploying Nemotron-3-Super-120B-A12B using NVIDIA NIM
- [TensorRT-LLM](https://dstack.ai/docs/examples/inference/trtllm.md): Deploying Qwen3-235B-A22B-FP8 using NVIDIA TensorRT-LLM on NVIDIA GPUs

### Models

- [DeepSeek V4](https://dstack.ai/docs/examples/models/deepseek-v4.md): Deploying DeepSeek-V4-Pro using SGLang on NVIDIA B200:8
- [Qwen 3.6](https://dstack.ai/docs/examples/models/qwen36.md): Deploying Qwen3.6-27B using SGLang on NVIDIA and AMD GPUs

### Accelerators

- [AMD](https://dstack.ai/docs/examples/accelerators/amd.md): Running dev environments, tasks, and services on AMD GPUs
- [Tenstorrent](https://dstack.ai/docs/examples/accelerators/tenstorrent.md): Running inference, training, and dev environments on Tenstorrent Wormhole and Blackhole accelerators
- [TPU](https://dstack.ai/docs/examples/accelerators/tpu.md): Deploying and fine-tuning models on Google Cloud TPUs using Optimum TPU and vLLM