Algomancer/Hyperspherical-Discrete-Bayesian-Flow-Like-Vibe-But-also-kinda-tangental-discrete-diffusion

Stars
2
Language
Python
Created 28 days ago
Updated 27 days ago

Algomancer/Hyperspherical-Discrete-Bayesian-Flow-Like-Vibe-But-also-kinda-tangental-discrete-diffusion

Algomancer

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

It's a hyperspherical thingy for discrete diffusionishness.

Bayesian-Flow-Networks

A simple implimentation of Bayesian Flow Networks (BFN)

Jupyter Notebook

Full-Stack-Scala-Starter

Play 2.5, ScalaJS, Binding.scala starter project.

The-Daily-Train

Training Models Daily

Code-Network-Starter

Seed Projects for Hackathons, Resources for rapid prototyping, tutorials

VCReg

Minimal Implimentation of VCRec (2024) for collapse provention.

Low-Rank-Extension-Token-Linear

This linear layer allows dynamic reallocation of parameters post training (tokenformer). Additionally, it creates a low rank transformation effectively doubling the amount of parameter tokens you can attend to.

Stablizing-Transformers

Stablizing transformer training by preventing attention entropy collapse

Energy-Sliced-Wasserstein

Computes the Energy Sliced Wasserstein Loss between two distributions. An optimal-transporty-energyish vibe distribution matching loss/regulariser.