Code Release for "Broken Neural Scaling Laws" (BNSL) paper

A library for mechanistic interpretability of GPT-style language models

Fast and memory-efficient exact attention

Effective Attention Sheds Light On Interpretability - Findings of ACL2021

Uniform Manifold Approximation and Projection

Code for the numerical experiments of the paper "A singular Riemannian geometry approach to Deep Neural Networks II. Reconstruction of 1-D equivalence classes."

Algorithms for computations on random manifolds made easier

Uniform Manifold Approximation with Two-phase Optimization (IEEE VIS 2022 short)

TensorFlow implementation of On the Sentence Embeddings from Pre-trained Language Models (EMNLP 2020)

Code release for "Git Re-Basin: Merging Models modulo Permutation Symmetries"

Utilities for the HuggingFace transformers library

Locating and editing factual associations in GPT (NeurIPS 2022)

Interactive Composition Explorer: a debugger for compositional language model programs

Repository for the code of the "PPL-MCTS: Constrained Textual Generation Through Discriminator-Guided Decoding" paper, NAACL'22

Implementation of RETRO, Deepmind's Retrieval based Attention net, in Pytorch

Render probabilistic graphical models using matplotlib

Rank-One Model Editing for Locating and Editing Factual Knowledge in GPT

functorch is JAX-like composable function transforms for PyTorch.

Hessian spectral density estimation in TF and Jax

ADAHESSIAN: An Adaptive Second Order Optimizer for Machine Learning

PyHessian is a Pytorch library for second-order based analysis and training of Neural Networks

Checking the divisibility of neural networks, and investigating the nature of the pieces networks can be divided into.