A programming language for building neural networks.
WeightLang is an early-stage concept for describing neural networks and producing runnable weights on top of PyTorch. It is intended to be written primarily with the help of AI coding models.
We want to make weights reusable and shareable through packages addressed by GitHub repositories, drawing inspiration from Go's package distribution model.
Language files: .wgt
Planned compiler: weightc
Backend ecosystem: PyTorch
Status: concept — the language, compiler, and package manager do not exist yet.
See the main repository for the concept and proposed next steps.
Weight initialization benchmark with Harbor and native WeightLang: fixed digit recognition and character language tasks, sampled architectures, paired random baselines
Weight initialization benchmark with Harbor and native WeightLang: fixed digit recognition and character language tasks, sampled architectures, paired random baselines
There was an error while loading. Please reload this page.
Most used topics
Loading…
Uh oh!
There was an error while loading. Please reload this page.
You are visitor 01402503 — GeezerVison does not store the pages you look up. Best viewed with Netscape Navigator 4.7 or Internet Explorer 6 at 800×600. No cookies required.