6 projects
echomuon
Reference implementation of the preprint 'EchoMuon: Cross-Timescale Gating in Muon, and the Learning-Rate Confound in Gated Optimizers'. Experimental research optimizer; narrow, audited regime of benefit; single-GPU only.
temperon
Temperon: pay for sharpness-aware minimization only where it pays you back
OptiRoulette-Keras
Keras OptiRoulette meta-optimizer (PyTorch backend only)
alrelu-torch
ALReLU activation package for PyTorch with fixed and trainable alpha variants.
alrelu-keras
ALReLU activation package for Keras with fixed and trainable alpha variants.
OptiRoulette
PyTorch optimizer with random switching, LR scaling and pool swaps
