Create a comprehensive W&B integration with custom dashboards, sweeps, artifact tracking, and team collaboration features.
You are an MLOps specialist. Build a comprehensive Weights & Biases integration for a ML team. ## Setup - **Project Name**: [PROJECT] - **Team**: [TEAM e.g. ml-research] - **Framework**: [FRAMEWORK e.g. PyTorch / TensorFlow / JAX] - **Experiment Types**: [TYPES e.g. training, evaluation, inference benchmarking] ## Components ### 1. Logging Utilities (`wandb_utils/logger.py`) - Auto-initialize with project/team/tags from config - Log training metrics with step alignment - Log evaluation metrics with custom charts - Log model architecture summary - Log GPU utilization and memory usage - Log dataset statistics and sample distributions - Log code state (git hash, diff, requirements) - Custom W&B Table logging for prediction samples ### 2. Hyperparameter Sweeps (`wandb_utils/sweeps.py`) - Sweep configuration generator for [FRAMEWORK] - Support Bayesian, grid, and random search - Early termination with Hyperband scheduler - Multi-objective optimization (accuracy vs speed) - Sweep resume and continuation - Best run extraction and config export ### 3. Artifact Management (`wandb_utils/artifacts.py`) - Dataset versioning with lineage tracking - Model checkpoint artifacts with metadata - Artifact dependency graph visualization - Automated cleanup of old artifact versions - Download artifacts with caching ### 4. Custom Reports (`wandb_utils/reports.py`) - Auto-generate W&B Report with: - Training curves comparison across runs - Best hyperparameter configurations table - Model comparison matrix - Error analysis panels - Hardware utilization summary - Template reports for each experiment type ### 5. Alerts & Notifications (`wandb_utils/alerts.py`) - Training completion alerts (Slack/email) - Metric threshold alerts (loss spike, accuracy drop) - Resource exhaustion warnings - Sweep completion summary Provide the complete Python package with setup.py, usage examples, and integration tests.
Free to copy and use. Compatible with Claude 4 Opus, GPT-5, Gemini 2.0 Flash.
Set WANDB_API_KEY environment variable. Configure project and team settings. Import wandb_utils in your training scripts and use decorators for automatic logging.
Initial release
Sign in and download this prompt to leave a review.