Models folder

This folder structure organizes the models used by the API by programming language and model family. The top-level models directory contains subfolders for each language (java, pharo, python) and a model_cards directory that stores documentation for every trained model variant.

Under model_cards, each language has its own folder (java, pharo, python), and within those folders models are split by approach, such as setfit and transformer, containing the respective model cards. For Java, for exemaple, the setfit model card describes a SetFit classifier built on top of sentence-transformers/paraphrase-MiniLM-L6-v2, trained on the Java portion of the NLBSE Code Comment Classification Dataset to predict seven documentation-related labels (summary, ownership, usage, etc.) with accuracy in the mid‑0.7 range on the test set. The corresponding Java transformer model card documents a CodeBERT‑based multi‑label classifier (microsoft/codebert-base) fine‑tuned on the same dataset, reporting detailed micro/macro F1, precision, recall, and subset accuracy, and explaining that it operates on the concatenated comment+context field.

For further technical and experimental details (e.g., hyperparameters, runtime benchmarks, and citation information), each individual model card file in the model_cards/<language>/<model_type>/README.md directories can be consulted directly.