smile

Why we included this project

Teams building on the JVM often end up stitching together several libraries to cover classic statistics, clustering, and deep learning. SMILE tries to close that gap with one framework: the usual workhorses (random forests, SVMs, gradient boosting, k-means, DBSCAN, t-SNE) sit behind a single consistent API, so going from a data frame to a fitted model and its metrics takes a few lines of code. Beyond that core, it adds a GPU-accelerated deep learning module on LibTorch with pretrained EfficientNet vision models, plus native LLaMA-3 inference and an OpenAI-compatible server, all without a Python bridge. Optional native BLAS/LAPACK acceleration helps heavier workloads scale, and because the core is self-contained, it slots cleanly into ordinary JVM services and build pipelines. Java, Scala, and Kotlin each get idiomatic APIs, which matters for teams that want the whole modeling surface in one place.