i-machine-think

diagNNose

diagNNose is a Python library that facilitates a broad set of tools for analysing hidden activations of neural models.

am-i-compositional

awesome-compositionality

A list of resources dedicated to compositionality

machine

Codebase for the i-machine-think project, a modular and fully tested Pytorch implementation for seq2seq models.

machine-tasks

Datasets for compositional learning

emergent_grammar_induction

A toolkit for finding and analysing the grammars of emergent languages.

compositionality_paradox_mt

Codebase for analysing compositional generalisation in NMT models, which allows you to run systematicity, productivity, substitutivity and idiom processing analyses.

signal

A library built on top of pytorch that provides tools to simplify the development of multi-agent referential games models that aims to research language emergence.

attentive_guidance

Repository implementing attentive guidance

grasp

[IJCAI 2024] GRASP: A novel benchmark for evaluating language GRounding And Situated Physics understanding in multimodal language models

incremental_encoding

morphology_and_generalisation

Repository with code for interpretability analyses of a recurrent neural network predicting German plural nouns.

seq2attn

specialised_attention

Specialised attention implementation : adding position attention and improving content attention.

gcd4lm

Scripts for Generalised Contextual Decomposition

i-machine-think.github.io

attention-cd

learn

machine-analysis

library to do analysis on machine

matrics

Machine Metrics