ml-ops

What problem does it solve?

This Skill addresses the complexities of managing the entire machine learning model lifecycle, from initial training to ongoing monitoring in production. It aims to reduce the manual effort and potential for error in MLOps workflows.

Core Features & Use Cases

• Model Training: Provides guidance on setting up basic training loops and fine-tuning pre-trained models.
• Experiment Tracking: Emphasizes the importance of logging hyperparameters, metrics, and artifacts for reproducibility.
• Model Versioning & Deployment: Covers saving model checkpoints and outlines various serving options like FastAPI and TorchServe.
• Monitoring: Highlights the need to track data drift, model drift, latency, and error rates.
• Use Case: When developing a new image classification model, use this Skill to set up experiment tracking with MLflow, fine-tune a BERT model using HuggingFace's Trainer, and deploy it as a FastAPI service.

Quick Start

Use the ml-ops skill to train a PyTorch model with experiment tracking using MLflow.