How gokumohandas/made-with-ml works
Made With ML is a course and reference implementation for building, evaluating, and serving production-oriented ML workloads. Its code workflow loads and preprocesses project data, trains a SciBERT-based classifier, evaluates checkpoints, and exposes prediction and evaluation endpoints. The README also describes tuning and scalable data/train/tune/serve workloads; tuning is represented as a documented capability, but its implementation was not sampled, so no implementation links are inferred for it.
Main entry points: train.py Distributed training · predict.py Checkpoint prediction · serve.py HTTP model deployment · data.py Load and preprocess
Components 12
Data preparation
- Load and preprocess
madewithml/data.py
Model development
- Distributed training
madewithml/train.py - SciBERT classifier
madewithml/models.py - Hyperparameter tuning
madewithml/tune.py
Experiment operations
- Runtime configuration
madewithml/config.py - Training utilities
madewithml/utils.py - Metrics and slices
madewithml/evaluate.py
Model serving
- Checkpoint prediction
madewithml/predict.py - HTTP model deployment
madewithml/serve.py - Deployment entrypoint
deploy/services/serve_model.py
Other components
- Developer / caller
- Project datasets
datasets
Connections 19
- Project datasets → Load and preprocess: reads CSV
madewithml/data.py - Distributed training → Load and preprocess: loads and splits
madewithml/train.py - Distributed training → SciBERT classifier: trains classifier
madewithml/train.py - Distributed training → Training utilities: records results
madewithml/train.py - Load and preprocess → Runtime configuration: reads settings
madewithml/data.py - Distributed training → Runtime configuration: imports settings
madewithml/train.py - Metrics and slices → Checkpoint prediction: loads checkpoint
madewithml/evaluate.py - Metrics and slices → Training utilities: saves metrics
madewithml/evaluate.py - Checkpoint prediction → Load and preprocess: preprocesses input
madewithml/predict.py - Checkpoint prediction → SciBERT classifier: loads model
madewithml/predict.py - Checkpoint prediction → Training utilities: collates batches
madewithml/predict.py - HTTP model deployment → Checkpoint prediction: predicts requests
madewithml/serve.py - HTTP model deployment → Metrics and slices: evaluates requests
madewithml/serve.py - HTTP model deployment → Runtime configuration: configures tracking
madewithml/serve.py - Deployment entrypoint → Runtime configuration: reads registry path
deploy/services/serve_model.py - Deployment entrypoint → HTTP model deployment: binds deployment
deploy/services/serve_model.py - Developer / caller → Distributed training: runs workload
README.md - Developer / caller → Hyperparameter tuning: runs workload
README.md - Developer / caller → HTTP model deployment: sends HTTP requests