SKaiNET
SKaiNET is a Kotlin Multiplatform tensor / compile / graph engine. It provides a tensor DSL, execution contexts, a graph IR, model loaders (GGUF, SafeTensors, ONNX), quantization primitives (Q4_K, Q8_0, ternary, TurboQuant), a StableHLO emitter for cross- platform compile targets, Minerva secure MCU export for supported static MLPs, and a pluggable backend API that CPU, GPU, and NPU backends can implement independently.
This site is split by what you’re here to do:
-
You’re building an app with SKaiNET — pulling artifacts from a BOM, constructing tensors, running a forward pass, loading a GGUF model, training a small network. Inside, content follows the Diátaxis framework: Tutorials (learning-oriented), How-to guides (task-oriented), Reference (lookup), Explanation (background). Java consumers — see the "From Java" callout on that page.
-
You need to package a supported static MLP for libminerva: start with the tiny dry-run sample, then run the real-runtime profile, inspect
manifest.json, and hand off the generated host and firmware bundle. -
You’re modifying SKaiNET itself — building from source, adding kernels, running the benchmark suite, maintaining CI, operating the self-hosted runner. Audience admonition at the top of each page.
|
LLM-specific runtimes (Llama, Gemma, Qwen, BERT) live in the sibling SKaiNET-transformers repository and its own documentation site. Mainline SKaiNET covers the engine layer only. |