quintinpope Goto Github PK

academic-budget-bert

Repository containing code for "How to Train BERT with an Academic Budget"

basin-connectivity-slt

Trying to connect basin connectivity between models with the tools of singular learning theory

closs

Code that implements the CLOSS method of adversarial attack on language-model based classifiers.

dioptra

Test Software for the Characterization of AI Technologies

epi-smt

fasta_perceiver

Code to prepare FASTA data and train a perceiver language model with masked language modeling

grokking

PyTorch implementation of "Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets"

knowledge-neurons

A library for finding knowledge neurons in pretrained transformer models.

learn2gradienthack

Uses learn2learn to train a fully connected network that implements the constant function and resists SGD

lm-data-curation

Code to investigate how LLMs can curate their own training data.

methylmercury-website

Code for a website showing a pharamacokinetic model of methylmercury elimination in a human body.

microbiome_transformers

perceiver-pytorch

Implementation of Perceiver, General Perception with Iterative Attention, in Pytorch

pz_dilemma

petting zoo game theory

reflectivity_benchmarks

Datasets to probe language model competence at various subtasks associated with reflectivity / metacognition / etc

rome

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

s2ml-art-generator

Two notebooks which allow the use of VQGAN+CLIP as well as CLIP-guided diffusion for generating images from text or image prompts. Includes ESRGAN implementation for upscaling and ffmpeg for video generation.

terrible_prng

A VERY bad pseudorandom number generator.

textattack

TextAttack 🐙 is a Python framework for adversarial attacks, data augmentation, and model training in NLP https://textattack.readthedocs.io/en/latest/

unsupervised_behavior_evaluation

Scalable and flexible characterization of language model behavior

val-eval

Systematically uncover differences in language model behaviors and "values"