| name | ml-training-images |
| description | Use Azure ML / DSVM / NGC images for ready-to-run training stacks |
Ml Training Images
When to use this skill
- Distributed training, fast time-to-first-loss
Key concepts
- Data Science Virtual Machine (DSVM) — preinstalled frameworks, drivers, conda envs.
- NGC marketplace images — NVIDIA-curated TensorFlow/PyTorch containers.
- Azure ML compute clusters abstract VMSS for training jobs.
- Container approach (PyTorch + DeepSpeed image) often beats VM-level installs.
Common pitfalls
- Pinning DSVM image and never updating — drivers age out vs CUDA toolkit needs.
References
Validation-first: verify every Azure fact against the Microsoft Learn MCP server (cite the Learn URL); Learn overrides built-in knowledge. If no Learn MCP server is configured, flag the answer ⚠️ unverified and mark specs as indicative. Analysis only — verify against Microsoft documentation before applying.