Choose a page by goal. Read the tutorial in order; use concepts, guides, and reference pages for design context or exact answers.
- Tutorial index: environment through a complete small CPU run.
- Kaggle training: submit a T4 run and download artifacts.
- Evaluation and inference: split outputs, errors, and text prediction.
- Troubleshooting: data, devices, sequence length, checkpoints, and outputs.
- Classification flow: how text becomes auditable metrics.
- Code tour: package boundaries and reading order.
- Configuration flow: defaults, YAML,
--set, and validation. - Model tutorial: tensor contracts for EmbeddingBag, TextCNN, and BiLSTM.
- Architecture decision: reproducible experiment contracts.
- Generic CSV decision: adapters, label order, and audit boundaries.
| Question | Page |
|---|---|
| What does a config field mean? | Config reference |
| What are manifests, labels, and hashes? | Dataset format |
| How do I read accuracy, macro-F1, and confusion? | Metrics |
| What is stored in a checkpoint? | Checkpoint schema |
| Which model should I use? | Model catalog |
| What does each CLI command write? | CLI and outputs |
| How should experiments be compared? | Experiments |
| How do I add a model? | Adding models |
| Can AG News data be redistributed? | Dataset note |
Directory-level guides are available for configs, examples, scripts, and tests.