Gradual rollout of a model on Azure Machine Learning
Procedure from Azure Machine Learning studio: two versions of a model on the same endpoint, traffic mirrored to the new one, traffic split, rollback to the previous version, full switch, with a .NET console that counts the responses of each version.
Fine-tune a model on Microsoft Foundry and call it from .NET
Seventeen-step procedure: sample data in JSONL, supervised fine-tuning of gpt-4.1-nano on Microsoft Foundry, metrics and checkpoints, a Developer deployment, a comparison with the base model, a call from .NET with Entra ID, then delete the resources.
Train a neural network on Azure Machine Learning and call it from .NET
Procedure from the portal and Azure Machine Learning studio: two training runs of a small PyTorch network compared, the model registered in two versions, deployed to an endpoint with no scoring script and called from a .NET console, then delete the resources.