Aaron Mueller's repositories
contextualized-topic-models
A python package to setup topic classification fine-tuning, run contextualized topic modeling, and run TCCTMs
emergent-syntax
Code for "How to Plant Trees in Language Models" (ACL 2023).
syntax-icl
Code and data for In-context Learning Generalizes, But Not Always Robustly: The Case of Syntax
aaronmueller.github.io
Aaron Mueller's personal website.
multilingual-lm-intervention
Multilingual causal mediation analysis
lm-evaluation-harness
Few-shot evaluation of language models. Fork for the BabyLM competition (CoNLL '23).
messing-with-fst
Trying out finite-state transducers.
dont-stop-pretraining
Adapting the Don't Stop Pretraining approach for multilingual applications. Modified by Aaron Mueller and Nathaniel Weir.
dotfiles
Config files for easy setup on new UNIX-based machines
earley-parser
Earley parser implementation.
inverse-scaling-eval-pipeline
Basic pipeline for running different sized GPT models and plotting the results
mBERT-docclass
Investigation of different methods of multilingual fine-tuning for document classification with mBERT.
minicons
Utility for analyzing Transformer based representations of language.
mt-decoders
Basic IBM-style machine translation models with various decoding methods.
neural-narrative-generation
Generating stories given prompts using GPT-2. We also try diverse decoding!
nshell
nshell: a basic shell environment written in C
parlai-hred
Implementation of Hierarchical Recurrent Encoder-Decoder (HRED) model for narrative generation in ParlAI.
pos-hmm
Hidden Markov Model tagger
smoothed-lm
Implementing smoothed n-gram language models.
sparse_coding
Using sparse coding to find distributed representations used by neural networks.
text-to-text-transfer-transformer
Code for the paper "Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer"
transductions
A PyTorch framework for creating, running, and reproducing experiments on seq2seq models.
wiktionary-derivations-parser
For foreign editions of Wiktionary, extract derivations on each page (if they exist).