Control plane for agents & engineers to provision compute and run training & inference across NVIDIA, AMD, and other chips — on clouds, Kubernetes, and on-prem.
github.com/dstackai/dstack
Based in Germany
- Tweets 1,263
- Following 56
- Followers 1,583
- Likes 1,080
Here's one example of using the presets toolkit to optimize Qwen3.8-27B on a single @AMD MI300X.
Through compound learning across linked sessions and source-level patches, performance improved from 311 to 495 tok/s: +59%.
The resulting preset can be deployed on any AMD cloud, Kubernetes cluster, or bare-metal fleet.
Inference serving is increasingly open source. But inference optimization still happens inside each provider, and the optimized deployment remains tied to its proprietary stack. This has to change.
Introducing Presets: an open-source toolkit for agent-based inference optimization and a portable format for the result.
Deploying optimized Kimi K3 to any cloud, Kubernetes cluster, or bare-metal fleet, whether on NVIDIA, AMD, or other silicon, should be as simple as deploying a Docker image.
dstack.ai/blog/presets/
3
3
402
dstack 0.21 is out 🎉
A major release: the codebase has migrated to @pydantic v2. Improved CLI and server performance, plus Python 3.14 support. This clears the way to accelerate the project much further.
Also in 0.21:
* A new version of Presets, our experimental inference optimization toolkit
* Replicated gateways for high availability for inference deployments
* A ton of other improvements across the CLI, offers, and backends
Full release notes 👇
github.com/dstackai/dstack/r…
1
2
4
373
AI infra builders in SF, we are meeting on July 23.
@CrusoeAI, @lmsysorg SGLang, and dstack are hosting a meetup on GPUs, training, inference, and open source AI systems.
12 short technical talks with speakers from @sgl_project, @ByteDanceOSS, @NVIDIA, @radixark, @CrusoeDev, and more to be announced soon.
RSVP: luma.com/rxsn0u0h
3
4
13
5,565
Crusoe has been doing great work making GPU infrastructure more accessible for AI teams.
Now @CrusoeAI has a new guide on using dstack to manage clusters, training, and inference on Crusoe Cloud: docs.crusoecloud.com/third-p…
3
6
547
dstack 0.20.26 is out 🚀
This release makes run submission noticeably faster. Starting a run on an idle instance now takes ~1–2s instead of ~7s.
The server reuses pooled SSH connections to instances instead of reopening one per operation.
Full release notes 👇
github.com/dstackai/dstack/r…
2
4
252
dstack is now in the official @jarvislabsai docs 🙌
docs.jarvislabs.ai/dstack/
Orchestration is becoming the default way GPUs get used. Thank you, @jarvislabsai for the integration!
If you're looking for on-demand GPUs, they offer H100, L4, H200, RTX PRO 6000.
2
5
274
Native @NVIDIA Dynamo support is now in dstack.
Deploy high-throughput inference with PD disaggregation, without custom orchestration glue.
Works with SGLang, vLLM, and TensorRT-LLM, across GPU clouds, Kubernetes, and on-prem fleets.
Another step toward simpler AI-native orchestration for production inference:
dstack.ai/blog/nvidia-dynamo…
2
3
15
410
The agentic engineering wave is coming to AI infra, but Kubernetes is too messy to keep up, and is a real bottleneck.
The founder of dstack, @andrey_cheptsov, sat down with @readsail at @nvidia GTC to talk about GPU orchestration.
Big thanks to @makora_ai and @SemiAnalysis_ for organizing the interview!
1
2
5
277
@KonstantinWille, the brain modeling lead at Metamorphic, gave a talk on the Enigma project at @Stanford building the foundation for the next generation of foundation models.
Slides: drive.google.com/file/d/1_xW…
🧵 8/10
1
1
3
131
Tanya Verma, a co-founder of @TinfoilAI, gave a talk on verifiably private AI and confidential inference with secure hardware enclaves.
🧵 7/10
1
1
4
116
@andrey_cheptsov, the founder of dstack, gave a talk on building a new orchestration standard for AI infrastructure across hardware, cloud, and software vendors.
GitHub repo: github.com/dstackai/dstack/
Slides: drive.google.com/file/d/1lfm…
🧵 6/10
1
3
7
275
DJ Gianduso, the product lead at @makora_ai, gave a talk on generating GPU kernels to improve inference performance.
Slides: drive.google.com/file/d/1mbt…
🧵 5/10
1
1
3
83
Tanmay Chopra, the founder of @withemissary, gave a talk on post-training LLMs beyond weights.
Slides: drive.google.com/file/d/1a9D…
🧵 4/10
1
1
3
87
Connor Guerrero, the founding member of @CrusoeAI's developer relations team gave a talk on serving LLMs on Crusoe with KServe.
Tutorial: crusoe.ai/resources/blog/ser…
Slides: drive.google.com/file/d/1ERl…
🧵 3/10
1
2
4
184
@thu_yushengsu, the founding member at @radixark and a core contributor to @lmsysorg SGLang, gave a talk on the efficiency and determinism in large-scale RL training using the Miles framework.
GitHub repo: github.com/radixark/miles
Slides: docs.google.com/presentation…
🧵 2/10
1
4
5
1,815
Tomorrow in SF, we're bringing together GPU experts, AI researchers, and infra engineers.
A curated meetup hosted with @CrusoeAI and @lmsysorg — with talks from @radixark, @makora_ai, @tinfoil, Metamorphic, and others.
RSVP: luma.com/ykt5b294
3
1
7
1,758
.@transformerlab has integrated with dstack!
Transformer Lab is an agent friendly ML research platform for training models with modern experiment tracking, automated hyperparameter sweeps, and persistent storage across ephemeral nodes.
With dstack, those workflows run across any GPU cloud or on-prem cluster.
Both projects are open source. Check it out 👇🏽
lab.cloud/blog/dstack-launch…
4
5
564
dstack 0.20.18 is out 🚀
What’s new:
⚡️ Fetching available compute offers in `dstack offer` and `dstack apply` is 1.5x–22x faster (depending on backend)
⚡️ CLI now shows Docker pull progress for VM backends and SSH fleets
⚡️ Added support for GeForce RTX 2/3/4/5 GPUs
github.com/dstackai/dstack/r…
3
2
274
A lot of teams still do not know what to use for GPU orchestration. Kubernetes has become the default, but it usually means stitching together too much glue just to get development and inference working.
@GraphsignalAI uses dstack as a more lightweight, GPU-native orchestration layer for both development and inference.
1
2
5
384
See you today at Beyond Summit by @TensorWave. Marc will be on the panel for "Building the Open Source Ecosystem" at 12:30pm.
Talking about AMD, GPU infrastructure, orchestration, serving, and hardware enablement — and the open, interoperable stack that removes bottlenecks for AI deployment at scale.
2
5
248
dstack 0.20.16 is out 🚀
Major server performance improvements — a single replica now handles ~10x more resources, with 2x–10x faster background processing in benchmarks.
This release also includes multiple other improvements, including @Runpod support for CPU instances.
Release notes: github.com/dstackai/dstack/r…
1
2
5
1,331
autodebug by @GraphsignalAI is a closed-loop system for inference optimization.
It uses @dstackai to provision GPUs and redeploy services on each pass through the loop:
benchmark → read profiling telemetry → tweak config → redeploy → repeat.
What's interesting here is the combination of agentic optimization and heterogeneous hardware: the system is not tuning a fixed deployment, it is continuously searching across infrastructure and configuration.
There's no manual step between iterations.
@dmitrimelikyan's writeup: graphsignal.com/blog/autodeb…
1
6
8
751
dstack 0.20.15 is out.
Nice milestone for the @AMD compute ecosystem: with dstack, you can now provision AMD MI350X GPUs via @CloudRiftAI, one of the first providers to offer them on-demand.
Also in this release: ROCm 7.x compatibility and fixes for SSH fleets.
Release notes: github.com/dstackai/dstack/r…
3
7
1,140
Now @GraphsignalAI integrates with dstack — add @sgl_project profiling, tracing, and GPU metrics to your inference services.
pip install 'graphsignal[cu12]' + wrap with graphsignal-run. That's it.
graphsignal.com/docs/integra…
6
6
456
🔥 GPU Night at GTC was a packed room of people building at every layer of the AI stack — from training to inference.
Great conversations about open-source infra, GPU orchestration, and what it actually takes to ship AI workloads at scale.
Thanks to our co-hosts @AMD, @tensorwave, and @makora_ai for making it happen. See you at the next one.
4
7
383
dstack 0.20.13 is out ⚡️
Big update: SSH fleets can now be exported and shared across projects, making it easier for teams to reuse GPUs across on-prem clusters instead of leaving capacity locked to a single project.
Also shipping in 0.20.13:
* @CrusoeAI H200/B200 support for InfiniBand clusters
* Per-project Launch Wizard templates
Release notes: github.com/dstackai/dstack/r…
1
4
11
484
Join our GPU Night meetup at NVIDIA GTC 🔥
We're organizing an invite-only gathering for engineers and researchers working on training infrastructure, inference stacks, and GPU clusters.
Hosted by the teams behind @dstackai, @makora_ai, and @tensorwave.
See you there! RSVP: luma.com/1sgm9q5d
4
10
1,944
dstack is a CLI-first tool for managing AI infra. But if you prefer the good old UI , we just added Launch templates.
Define templates in YAML, launch runs on any GPU cloud via a guided UI wizard.
Here's an example of launching a dev environment with in-browser VS Code on @CrusoeAI. 👇
1
3
5
566
dstack 0.20.9 is out 🚀
More improvements around GPU workload provisioning and visibility: easier to see what’s happening during provisioning/scheduling, plus better Kubernetes integration.
Release notes 👇 github.com/dstackai/dstack/r…
3
4
431
🔥 @LambdaAPI now offers B200 GPUs on demand.
With dstack, you can schedule provisioning. It allocates a B200 as soon as capacity becomes available within your time window.
Orchestrate development, training, and inference.
1CC supported too (via SSH fleets or K8s backends).
3
6
302
Tomorrow (Feb 6) at 11am PT we’re going live 👋
Messing around with AI infra, GPU workloads on @runpod using dstack and AI agents!
piped.video/watch?v=CO60cekS…
3
5
374
🚀 New dstack release is out!
CLI upgrades:
• Live event watching with `dstack event -w`
• Rich fleet details via `dstack fleet`
Plus: dstack skills are here. Install with
`npx skills add dstackai/dstack`
Agents can now write configs and run the CLI for you 🤯
Full details: github.com/dstackai/dstack/r…
3
6
518
Live Feb 6 · 11am PT
We’ll mess with agentic workflows for AI infra cloud GPUs on @runpod with dstack, containers, agents doing the messy stuff.
👉 piped.video/watch?v=t3hgkz0I…
With @andrey_cheptsov & @NERDDISCO
2
2
7
930
Still provisioning and orchestrating AI infra by hand?
Not with dstack.
Announcing dstack skills for AI agents.
Install in one command:
npx skills add dstackai/dstack
dstack.ai/docs/installation/…
Agents get direct control over dstack resources to create and edit configs, manage resources, etc via the CLI and YAML.
1
3
9
883
🚀 dstack 0.20.6 released!
* The server now uses much less memory even with lots of runs.
* You can now ingest run logs via Fluent Bit and forward them to Elasticsearch/OpenSearch.
* The UI now assists with creating a default fleet, including during project creation.
Learn more: github.com/dstackai/dstack/r…
1
2
4
675
Our first release of the year — dstack 0.20.3 is out 🥳
You can now use @windsurf with dstack to provision GPU machines and spin up containerized dev environments — code, iterate, and debug ML workloads without managing infra or local setups.
Release notes: github.com/dstackai/dstack/r…
1
3
5
275
Announcing a new cluster example: how to orchestrate distributed workloads on @CrusoeAI 🚀
Under the hood, the example shows how to connect dstack to Crusoe's GPU clusters using either:
* @kubernetesio (with @NVIDIAAI GPU Operator), or
* Virtual Machines (VMs)
A first step toward deeper dstack × Crusoe integration to automate provisioning and cluster management!
Example: dstack.ai/examples/clusters/…
4
5
2,232
dstack 0.20.2 is released.
This update includes bug fixes and a UI improvement. The UI now displays a warning when a project has no fleets created yet, aligning its behavior with the CLI and making this requirement explicit before submitting runs.
github.com/dstackai/dstack/r…
4
7
669
We tested @LambdaAPI's 1-Click Clusters and managed Kubernetes. The cluster interconnect worked out of the box with no extra setup, which was great to see.
If you're using Lambda's clusters, we published a guide on orchestrating distributed workloads on Lambda using dstack.
dstack.ai/examples/clusters/…
4
13
2,077
* @verdacloud (formerly DataCrunch) is partnering with dstack on GPU orchestration!🚀
The docs include a guide for using dstack to orchestrate GPU workloads:
* Unified GPU provisioning and job scheduling
* Support for dev, training, and inference workflows
Docs: docs.verda.com/integrations/…
1
2
8
368
🚀 dstack 0.20 RC1 is out!
The upcoming 0.20 release is one of our biggest updates yet, introducing major improvements across the platform.
Release notes: github.com/dstackai/dstack/r…
RC1 is now live in staging — test it, share your feedback, and help us get it ready for GA on Dec 17.
3
4
1,692
Read how @toffee_ai simplified their multi-cloud GPU stack with dstack and now ship LLM and image-gen inference across GPU clouds, all while reducing GPU spend by 2-3x.
Read the full case study: dstack.ai/blog/toffee/
5
7
518
A new case study is out: @toffee_ai breaks down how they use dstack to manage AI inference across neoclouds like @runpod and @vast_ai.
One control plane, consistent deployments, much less infra overhead.
Read it here: research.toffee.ai/blog/how-…
3
4
1,131
Early access: B300 (Blackwell Ultra) VMs are now available on @verdacloud (former DataCrunch) via dstack.
* Available on both on-demand and spot
* Up to 8× GPUs per VM
* Pricing: $4.95/h on-demand, $1.24/h spot
If you need Blackwell nodes for training, eval or inference, you can now orchestrate them through dstack.
1
2
5
1,673
Two weeks ago, @andrey_cheptsov and @AleksandrPatr talked about why multi-cloud matters for GPUs and how dstack, an open-source project, makes it easier to use.
Here’s a short demo from that session showing distributed training with dstack and @nebiusai.
piped.video/watch?v=X41Umi2l…
1
3
10
1,759
Since launching fleets last year, they’ve become one of the most impactful ways to simplify GPU provisioning across cloud and on-prem.
Over the past year, we've added:
* Native @kubernetesio support
* Elastic fleets that auto-scale as team needs change
Read more about how fleets work: dstack.ai/docs/concepts/flee…
2
8
1,713
dstack now supports multi-node scheduling on @runpod!
Run your jobs on B200, H200, H100, and A100 GPU instant clusters - InfiniBand up to 3200 Gbps 🚀
1
4
10
2,461
Last week’s @PyTorch Conf was packed with great discussions on GPU orchestration, model training, and inference🚀
Our meetup was a real highlight - tons of builders, deep technical chats on AI infra, and open-source models.
2
2
8
1,337
Multi-cloud AI infrastructure in action - with @nebiusai and dstack⚡️
Join @AleksandrPatr and @andrey_cheptsov for a deep dive on:
* Why GPU container orchestration matters, and what's beyond Kubernetes and Slurm
* Deploying distributed training job
* Cost-optimization an efficient GPU utilization
Oct 30 at 9am PDT / 5pm CEST, register here: nebius.com/events/webinar-mu…
3
4
354
At @PyTorch Conf and working with GPUs?
Whether you run on cloud or on-prem, swing by B14 to see dstack, open-source orchestration made for GPUs.
1
4
1,164
You can now create dev environments from the UI, not just the CLI.
Use any GPU on any cloud (or on-prem). It’s that simple:
1. Pick a GPU
2. Click Apply
3. Open in @code or @cursor_ai
The environment automatically launches on your selected GPU in a configured cloud.
4
10
721
Building AI infra or open-source models? Be here tomorrow at 6pm.
Makers from @nvidia, @scale_AI, @LambdaAPI, @Snowflake, @TinfoilAI, @mako_dev_ai, @axolotl_ai, and more are dropping what they've been working on.
2
3
12
1,562
If you're into AI infra or open-source models, this is the place to be on Oct 21 evening.
Folks from @mako_dev_ai, @scale_AI, @Snowflake, @PhalaNetwork, @BytedanceTalk, @axolotl_ai, @TinfoilAI, @LambdaAPI, and more will be sharing what they’re building.
4
6
26
12,005
We’re taking the next step in bringing GPU-native orchestration to the @kubernetesio ecosystem. 🔥
After adding NVIDIA GPU support two weeks ago, today’s release expands that to include @AMD GPUs too!
3
5
19
3,623
Big update for the dstack Sky GPU marketplace 🎉
@nebiusai joins as the first provider with on-demand and spot clusters. Until now, Nebius GPUs were accessible only through the open-source control plane.
👉🏼More on how dstack helps you manage clusters and accelerate training & development: dstack.ai/blog/nebius-in-dst…
1
5
2
813
Don’t want to self-host the open-source dstack? Try dstack Sky.
Today’s release adds a project wizard to choose between the GPU marketplace or your own cloud accounts.
New accounts also get free GPU credits.
1
5
7
2,368
🚀 dstack 0.19.29 is here!
This release takes another step toward simpler, fully declarative GPU fleet management. dstack now automatically reuses existing fleets when they fit.
We've also improved the Offers UI, which now supports grouping by backend for easier comparison.
More details: github.com/dstackai/dstack/r…
2
6
1,190
Multi-cloud GPUs, one control plane.
dstack Sky gives AI teams low-cost on-demand & spot GPUs from the global marketplace — or from their own cloud accounts.
Fully compatible with open-source dstack.
1
4
4
476
Stop paying 3-7× more for GPU workloads.
Our guest post on @runpod's blog shows how dstack's container-native control plane cuts costs while orchestrating GPUs at scale 🚀
runpod.io/blog/orchestrating…
1
5
8
1,408
🔥 New orchestration capabilities!
dstack now natively integrates with @digitalocean and @AMD Developer Cloud.
⚡Provision & manage AMD Instinct GPUs cost-effectively
⚡Run dev environments, training, inference & benchmarks
⚡Lightweight, GPU-native orchestration - built for ML teams
More on our blog: dstack.ai/blog/digitalocean-…
6
5
19
1,252
Heading to the @PyTorch conference in SF this October?
Join us the day before for our 3rd AI Infra & Open-Source Models meetup, co-hosted with @LambdaAPI! 🔥
We're bringing together GPU hackers, infra engineers & OSS model builders.
Speakers from @axolotl_ai, @mako_dev_ai, @EA, & more TBA.
RSVP now: lu.ma/tmpnldk5
1
5
9
2,104
On Beyond CUDA with @tensorwave, @andrey_cheptsov, CEO of dstack, unpacks the gaps Kubernetes & Slurm leave in AI infrastructure - and why we need AI-first orchestration⚡️
3
4
11
814
Check out the new dstack example!
Generate stunning AI videos with @Alibaba_Wan 2.2, the new SOTA open-source video-generation model. 🚀🎥
Run it as a task or service on any GPU cloud or on-prem cluster – with a single command.
👇
dstack.ai/examples/models/wa…
4
8
743
🚀 dstack 0.19.26 is out!
You can now mount Git repos in dev environments, tasks, and services - configured directly in the run configuration.
More in the release notes 👉🏼 github.com/dstackai/dstack/r…
1
3
6
349