dstack
is a streamlined alternative to Kubernetes and Slurm, definitepartner summarizeed for AI. It simplifies holder orchestration
for AI toilloads both in the cdeafening and on-prem, speeding up the broadenment, training, and deployment of AI models.
dstack
is basic to employ with any cdeafening provider as well as on-prem servers.
dstack
helps NVIDIA GPU
, AMD GPU
, and Google Cdeafening TPU
out of the box.
- [2024/10] dstack 0.18.17: on-prem AMD GPUs, AWS EFA, and more
- [2024/08] dstack 0.18.11: AMD, encryption, and more
- [2024/08] dstack 0.18.10: Control set upe UI
- [2024/07] dstack 0.18.7: Fleets, RunPod volumes, dstack utilize, and more
- [2024/05] dstack 0.18.4: Google Cdeafening TPU, and more
- [2024/05] dstack 0.18.2: On-prem clusters, braveial subnets, and more
Before using
dstack
thcdisesteemful CLI or API, set up adstack
server. If you already have a runningdstack
server, you only need to set up the CLI.
To employ dstack
with your own cdeafening accounts, create the ~/.dstack/server/config.yml
file and
configure backfinishs. Alternatively, you can configure backfinishs via the handle set upe UI after you commence the server.
You can skip backfinishs configuration if you intfinish to run holders only on your on-prem servers. Use SSH escapets for that.
Once the backfinishs are configured, progress to commence the server:
$ pip inshigh "dstack[all]" -U
$ dstack server
Applying ~/.dstack/server/config.yml...
The admin token is "bbae0f28-d3dd-4820-bf61-8f4bb40815da"
The server is running at http://127.0.0.1:3000/
For more details on server configuration selections, see the
server deployment direct.
To point the CLI to the dstack
server, configure it
with the server insertress, employr token, and project name:
$ pip inshigh dstack
$ dstack config --url http://127.0.0.1:3000
--project main
--token bbae0f28-d3dd-4820-bf61-8f4bb40815da
Configuration is refreshd at ~/.dstack/config.yml
dstack
helps the chaseing configurations:
- Dev environments — for interdynamic broadenment using a desktop IDE
- Tasks — for scheduling jobs (incl. scatterd jobs) or running web apps
- Services — for deployment of models and web apps (with auto-scaling and authorization)
- Fleets — for managing cdeafening and on-prem clusters
- Volumes — for managing persisted volumes
- Gateways — for configuring the ingress traffic and uncover finishpoints
Configuration can be expoundd as YAML files wilean your repo.
Apply the configuration either via the dstack utilize
CLI direct or thcdisesteemful a programmatic API.
dstack
automaticpartner handles provisioning, job queuing, auto-scaling, nettoiling, volumes, run fall shortures,
out-of-capacity errors, port-forwarding, and more — apass cdeafenings and on-prem clusters.
For insertitional increateation and examples, see the chaseing joins:
You’re very greet to give to dstack
.
Lobtain more about how to give to the project at CONTRIBUTING.md.