carbono

changelog

• v5: new metadata fields added; removed unnecesary "layers" metadata
• v4: optimized save/load handling for new json-based extension name ".uai"
• v3: adam optmizer, a complete code rewrite for maintainance purposes (official release)
• v2: softmax, cross-entropy and labels for classifications
• v1: first release

a micro-library with a simple api to build and train feed-forward neural networks, all within ~500 lines of code. Take a look at the playground and try training your first model. Also check some examples to get familiar with it. Its straightforward api lets anyone to prototype with neural networks like never before.

Quick Usage

<script src="carbono.js"></script>

new carbono().layer(2,1).train([{ input:[1,1], output:[0] }])

if you need help, you can reach to me as @apppvoidofficial on X, you can also find me on tinyllama discord server sometimes.