Model Development Index

Controlled offline model development and evaluation.

Notes

• Experiment Tracking — MLflow runs/registry/aliases, W&B sweeps, and DVC data versioning for reproducible experiments
• Distributed Training — DDP, model parallelism, Horovod/Accelerate, gradient checkpointing, and multi-node setups
• Offline Evaluation — hold-out strategy, cross-validation, calibration, bias audits, and eval infrastructure

Links

• ↑ ML Engineering
• Prev: ← 04 — Feature Engineering
• Next: 06 — Deployment & Serving →