tfd

vectorflow

Vectorflow is a lightweight deep learning framework developed by Netflix, designed specifically for efficient neural network training on sparse datasets in single-machine settings. It provides a streamlined alternative to larger frameworks when working with recommendation systems, natural language processing, and other sparse data applications. ## Key Features - **Sparse Data Optimization** — Efficiently handles sparse input data common in recommendation systems and NLP tasks. - **Minimalist Design** — Lightweight library with no external dependencies beyond a D compiler. - **Single-Machine Focus** — Optimized for training on individual servers rather than distributed clusters. - **D Language Implementation** — Built using the D programming language for performance and simplicity. - **Easy Integration** — Distributed as a dub package for straightforward dependency management. ## Philosophy Vectorflow prioritizes simplicity and efficiency, offering a focused toolset for sparse data problems without the overhead of larger, more complex deep learning frameworks.

bindbc-onnxruntime

bindbc-onnxruntime provides dynamic D language bindings to the C API of Microsoft's ONNX Runtime, a cross-platform, high-performance machine learning inferencing and training accelerator. These bindings enable D developers to leverage ONNX Runtime's capabilities for ML model deployment and execution within D applications. ## Key Features - **Dynamic Bindings** — No link-time dependency on the ONNX Runtime library, requiring manual runtime loading of the shared library. - **@nogc and nothrow Compatibility** — Designed to work in D code that prohibits garbage collection and exceptions. - **-betterC Support** — Can be compiled for compatibility with D's -betterC mode, allowing integration into minimal environments. - **Version Configuration** — Support for specific ONNX Runtime versions (e.g., 1.2) configurable at compile time via D version identifiers. ## Philosophy The project follows the bindbc pattern, focusing on providing flexible, low-overhead bindings that maintain compatibility with D's advanced language features and constraints.

grain2

Autograd and GPGPU library for dynamic neural networks in D