dstackai/dstack is an orchestration layer that organizes different AI compute resources in a single layer so training and inference can run on GPU clouds, Kubernetes, virtual machines, and bare-metal clusters. The latest release of this repository is 0.22.4.
The project is a unified control plane responsible for GPU provisioning and orchestration. GPU provisioning means preparing and allocating GPU resources in a cloud or cluster. It supports development, training, and inference, and supports NVIDIA, AMD, Google TPU, and Tenstorrent accelerators by default. The Fleets, Dev environments, Tasks, Services, and Volumes configurations cover cluster management and job execution, inference endpoint deployment, and data persistence. Presets are still marked as experimental. The implementation language is Python, the license is MPL-2.0, and the repository has 2284 stars.
However, the given record contains no changelog for 0.22.4, so what changed in this version cannot be known. Among the evaluation scores, the novelty, technology, and quality items have no values, and only the momentum score remains at 0.0. Performance benchmark numbers are not included in this record either. The feature list in the README alone does not show how well this tool has been validated at scale. The only way to confirm the actual changes in 0.22.4 is to read the repository's release notes directly.