# Notebooks

Exploratory analysis and Colab training notebooks. Run them in order. The
first is best run locally; the training ones expect a Colab (T4+/A100) or
CUDA GPU environment.

| Notebook | Purpose |
|----------|---------|
| `01_data_exploration.ipynb`     | Dataset overview, language distribution, label stats |
| `02_train_colab.ipynb`          | End-to-end training (aspect extraction + sentiment) on Colab |
| `03_model_comparison.ipynb`     | Compare trained runs from the MLflow tracking server |
| `04_qlora_colab.ipynb`          | QLoRA 4-bit fine-tuning of the sequence classifier |
| `05_final_evaluation.ipynb`     | Final cross-lingual evaluation and latency benchmarks |

These notebooks are generated/updated from `scripts/generate_notebooks.py`
where applicable.