Description
GPUs spread across cloud vendors, Kubernetes clusters and on-prem each run training and inference differently. dstack is a unified orchestration layer for heterogeneous AI compute, running jobs the same way on GPU clouds, K8s, VMs or bare metal.
It supports NVIDIA and AMD GPUs for dev environments, training, fine-tuning and serving, a lighter alternative to Kubernetes and Slurm.
Unified:Any compute.
GPUs:NVIDIA and AMD.
Workloads:Train and serve.
Lightweight:K8s and Slurm alternative.
It supports NVIDIA and AMD GPUs for dev environments, training, fine-tuning and serving, a lighter alternative to Kubernetes and Slurm.
Features
Unified:Any compute.
GPUs:NVIDIA and AMD.
Workloads:Train and serve.
Lightweight:K8s and Slurm alternative.
