Sketch of the pretraining and fine-tuning workflow. Pretraining uses normalized spectra or foundation-model spectral embeddings to predict labels: [Fe/H] from APOGEE (>−2.0), supplemented at lower metallicities by PASTEL, SAGA, and other VMP and UMP datasets; and [α/Fe] from APOGEE. The foundation model is trained on large unlabeled datasets. Fine-tuning then adapts the network using fewer labeled spectra, either from spectra or spectral embeddings, with different fine-tuning modules applied (see Section 3).