Cloud

Training pipelines

Multi-node distributed training with sovereign checkpointing.

Orchestration

Xobriq Cloud provides both Slurm (for research teams) and Kubernetes (for production MLOps). Both are fully managed — you just submit jobs.

Submit a training job

bash
`xobriq train submit \ --nodes 4 \ --gpus-per-node 8 \ --image xobriq/pytorch-2.4 \ --script train.py \ --checkpoint-bucket sov://checkpoints/llama-13b`

Features

  • NVLink Fabric across nodes for near-linear scaling.
  • Automatic checkpoint resumption on preemption.
  • Sovereign object storage (S3-compatible, in-country).
  • Integrated Weights & Biases and MLflow.
  • TensorBoard, Grafana, and cost dashboards out of the box.