SAELens
by jbloomAus
Training Sparse Autoencoders on Language Models
AI summary
Autoencoder trainer
A tool for training and analyzing sparse autoencoders to improve the understanding of neural networks and create safer AI systems.
- stars
- 526
- forks
- 129
- watching
- 7
Similar projects
Found by comparing what the projects do, not just their names.
Autoencoder library
An implementation of Adversarially Regularized Autoencoders for language generation and discrete structure modeling.
Autoencoder library
A software framework for semi-supervised generative Autoencoder models applied to single-cell data analysis.
Distributed Training System
Automates large-scale deep learning training on distributed clusters, providing fault tolerance and fast recovery from failures.
Autoencoder
An implementation of Adversarial Autoencoders using PyTorch for training neural networks on structured data.
Sentence encoder
Trains an autoencoder to learn generic sentence representations using convolutional neural networks
Batch trainer
A tool for training neural networks using large batch sizes and analyzing the trade-offs between longer training periods and better generalization performance.
Algorithm Trainer
A framework to learn simple algorithms from examples by generating and visualizing intermediate solutions
Autoencoder
A Tiny AutoEncoder implementation for Stable Diffusion's image generation model
Neural network optimizer
Trains artificial neural networks using the genetic algorithm
Language model trainer
Develops a method for pre-training language understanding models by combining masked and permuted techniques, and provides code for implementation and fine-tuning.
Autoencoder library
A library of autoencoders for sequential data
Neural network trainer
A library for training and evaluating neural networks with a focus on adversarial robustness.
Speech trainer
An Android app for training speech recognition in low-resource languages, utilizing FieldDB corpora and user-defined vocabularies.
Keras Trainer
Trains Keras models using the Genetic Algorithm
CNN Trainer
Trains CNNs using a genetic algorithm for classification problems where one class per sample is allowed.
