Releases: haifengl/smile
Releases · haifengl/smile
4.0.0
- smile-deep package for deep learning
- Llama 3.1 model in Java
- Native Java implementation of tiktoken tokenizer
- EfficientNet model for image classification
- smile-shell has built-in training and inference functionalities, including streaming APIs.
- smile-serve is an LLM inference server with OpenAI-compatible APIs and fully functional frontend.
- Gradient boost is 10X faster on very large dataset
- Code refresh to leverage latest Java features.
- Various plain value classes are converted to records
- Smile shell for Java and Kotlin
- Java 21 required
3.1.1
3.1.0
3.0.3
3.0.2
3.0.1
3.0.0
- Switch to dual license model to meet open-source projects (GPL) and the development and distribution needs of commercial distributors (such as OEMs, ISVs and VARs).
- Java Module friendly with auto module name
- Redesigned feature engineering packages (missing value imputation, transform, selection, extraction, importance)
- One-class SVM
- Isolation forest
- Feature Hashing
- One-way ANOVA
- BigMatrix supporting more than 2 billion elements
- Latin hypercube sampling
- CLI supports training, batch prediction, endpoint, etc.
- Bug fixes
2.6.0
- Spark integration (thanks Pierre Nodet)
- t-SNE is 6X faster (thanks Brault Olivier-O)
- Fully redesigned Gaussian Process Regression with HPO
- L-BFGS-B
- Matern kernel and composed kernels
- Fully redesigned model validation facilities and metrics
- Various optimization and bug fixes