Adam Hibble

Bayesian-Flow-Networks. A simple implimentation of Bayesian Flow Networks (BFN)

Full-Stack-Scala-Starter. Play 2.5, ScalaJS, Binding.scala starter project.

Minimal-Drifting-Models. A minimal implementation of Drifting Models for 2D toy data. Unlike diffusion/flow models that iterate at inference, drifting models evolve the pushforward distribution during training and generate in a single forward pass (1-NFE). The drifting field V governs sample movement: V -> 0 as generated matches data.

VCReg. Minimal Implimentation of VCRec (2024) for collapse provention.

Code-Network-Starter. Seed Projects for Hackathons, Resources for rapid prototyping, tutorials

The-Daily-Train. Training Models Daily

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

flow_jepa. Initial explorations for lossless representation learning.

Positional-Token-Linear. Neat.

Rank-Collapse-Adversary. A novel (?) generative model abusing dimensionality collapse to play an adverserial game.

Adverserial-Flow-Matching-with-Learned-Prior. Extension of adversarial flow matching where the noise distribution (prior) is learned rather than fixed to N(0, I). This allows the model to potentially find a better matching between the prior and data manifold.

BandPassMask. A simple masking strategy that out performs random and block masking, and some semantic masking for training masking based self supervised models.

Mean-Bayesian-Flow-Network. A mean flow like, bayesian flow like model for single function evaluation sampling of catagorical/discrete data.

Stablizing-Transformers. Stablizing transformer training by preventing attention entropy collapse

RankUp. RankUp is a PyTorch extension that nudges gradient updates away from directions that would lower the rank of your model’s parameters or features. It can be useful if you need to preserve full-rank behavior, for instance in certain regularization or constraint settings.

Hyperspherical-Discrete-Bayesian-Flow-Like-Vibe-But-also-kinda-tangental-discrete-diffusion. It's a hyperspherical thingy for discrete diffusionishness.

Rank-Regularisation. Rank Regularisation for Rank Collapse Prevention in Self Distillation Methods

Self-Referential-Weight-Update. Off the dome quick jot, wrote it while waiting for an eval plot. Unverified and slow, probably buggy. Don't use.

latent_trajectory_transformer. I am suprised this works. It's pretty cool.

Research-In-Public. Code to go along with the exploration in this x thread https://x.com/Algomancer/status/2019949738037272631

pscan. Python

awesome-projects-boilerplates. Boilerplates for mobile and web apps

Gaussianity-Grad. leveraging the analytic gradients of characteristic function statistics to structure embedding spaces. inverts the epps-pulley normality test to act as gradient modifier. strictly enforces gaussianity

Minimal-Free-Form-flows. A minimal implementation of Free-Form Flows (FFF) for 2D toy data. FFF learns an encoder-decoder pair where latent codes z = enc(x) follow a simple prior p_Z = N(0,I)