Andrey Cheptsov
@andrey_cheptsov
@dstackai. AI infra. Previously @JetBrains
In case you missed it: @nebiusai AI Cloud now supports dstack — a way friendlier alternative to Slurm and Kubernetes for ML teams. Dev environments, cluster training, inference — all covered. Docs 👉 docs.nebius.com/3p-integration…
We joke that @nebiusai is more of a real estate business than a tech company—our most important meetings are always about capacity planning. The toughest job? Capacity manager. Every seller is after you, begging for approvals. And me? No good days. It’s always one of two…
Is it just me, or is the future starting to feel more unsettling with each of these announcements?
Now that we have an API, we're officially on the @dstackai roadmap for direct support in provisioning a 1xMI300x vm. Thanks @andrey_cheptsov! github.com/dstackai/dstac…
In our continued commitment to open-science, we are releasing the Voxtral Technical Report: arxiv.org/abs/2507.13264 The report covers details on pre-training, post-training, alignment and evaluations. We also present analysis on selecting the optimal model architecture, which…
>>> Qwen3-Coder is here! ✅ We’re releasing Qwen3-Coder-480B-A35B-Instruct, our most powerful open agentic code model to date. This 480B-parameter Mixture-of-Experts model (35B active) natively supports 256K context and scales to 1M context with extrapolation. It achieves…
every time you read a tech report from a SOTA open-source model that basically puts to shame 99% of western labs' models (besides the absolute top 2 ones), it always reads like: "yeah we basically curated very high quality data, carefully generated a lot of data [due to not being…
If you want to be an AI developer and target those high paying gigs that are all the rage right now, you need access to enterprise compute. Saying that you trained a SOTA model on your home gaming GPU rig doesn’t sound nearly as impressive as saying that you setup @dstackai to…
great work! for everyone who is too lazy to read, bottom line is VMs perform about the same as bare metal (only tested on single nodes)
We asked @dstack and @andrey_cheptsov to do an unbiased investigation into whether or not our 1xMI300x virtual machines have any performance issues and this is what they discovered...
How do @AMD GPUs perform in VMs vs bare-metal environments? We benchmarked containerized workloads on both — for training and inference using the MI300X. Read the results: dstack.ai/blog/benchmark…
How do @AMD GPUs perform in VMs vs bare-metal environments? We benchmarked containerized workloads on both — for training and inference using the MI300X. Read the results: dstack.ai/blog/benchmark…
Kimi K2 tech report just dropped! Quick hits: - MuonClip optimizer: stable + token-efficient pretraining at trillion-parameter scale - 20K+ tools, real & simulated: unlocking scalable agentic data - Joint RL with verifiable + self-critique rubric rewards: alignment that adapts -…
Day 5: nobody responds except the standard comment on X „we will create a ticket“ @AIatAMD @AnushElangovan you can do better
.@AMD generously supported the @dstackai team with cloud credits to try MI300X GPUs (and explore AMD Developer Cloud integration). Apparently, the credits can only be used via @digitalocean — and that part isn’t working. Credits require a credit card. Credit card can’t be added…
Speculation: Within a year a <100B open weights model will also solve 5/6 IMO problems.
dstack 0.19.20 is out! The last release revamped log handling — this one fixes the UI issues that came with it. Log handling is now smoother, faster, and a lot more usable. 👇 github.com/dstackai/dstac…
Oh, brilliant @sama! Next, turn OpenAI into the McDonald’s of AI—buy data centers, rent GPUs to other clouds, and call it "AI Real Estate Trust
we will cross well over 1 million GPUs brought online by the end of this year! very proud of the team but now they better get to work figuring out how to 100x that lol
🎥 In this Tech Talk, learn how to deploy large language models efficiently using @dstackai on Vultr. Discover how dstack simplifies AI workflows while leveraging Vultr’s high-performance cloud. youtu.be/MoyMTnZ43lU #AI #MachineLearning #DevOps #CloudComputing #TechTalk #LLM
if you haven’t realized it yet, the game is already on
nvidia could do the most viral ai competition in history: start with 10,000 researchers and give each a free gpu to work on a public leaderboard but do rounds of elimination where the winners take the remaining hardware. the final winner gets all the gpus for a year.