Sendoa Moronta's escape route from the final_v2_last.ipynb trap: treat data as fundamentally separate from code, version it with DVC, and keep notebooks as consumers — not owners — of your logic. 💫
👉🏽 hubs.la/Q04nZC2Q0
🦉DVC
@DVCorg
Open source tool for data, models, & experiment versioning for ML projects. Join our stellar community dvc.org/support for help, support and insights.
- For AI touching power grids and pipelines, the baseline here is non-negotiable: DVC and Git tracing every dataset behind a recommendation, plus physics-based constraint checks before anything reaches a control system. 👉🏽 hubs.la/Q04nYtMV0
- @truefoundry published a hands-on guide (auf Deutsch) to deploying DVC-versioned models straight to @kubernetes — dvc pull the model from S3, wrap it in @FastAPI, and ship it. 👉🏽 hubs.la/Q04nZ4Lg0
- The Swiss AI Center (@hes_so) turned a full MLOps course into a space mission: DVC versions your data at liftoff, CML reviews model diffs in orbit, and drift alerts trigger reentry back into Label Studio for retraining. 👉🏽 hubs.la/Q04nYNL00
- Muhammad Ishfaq trained a great emotion-detection model, then couldn't reproduce it. His fix: MLflow's registry for model versions, DVC for the data behind them, and treating experiment logging as non-negotiable from day one. 👉🏽 hubs.la/Q04nYX3X0
