DeepSPIN

entmax

The entmax mapping and its loss, a family of sparse softmax alternatives.

infinite-former

lp-sparsemap

LP-SparseMAP: Differentiable sparse structured prediction in coarse factor graphs

uncertainties_MT_eval

Code and data for the paper "Disentangling Uncertainty in Machine Translation Evaluation", accepted at EMNLP 2022.

sparse_text_generation

hallucinations-in-nmt

robust_MT_evaluation

Repository for "BLEU Meets COMET: Combining Lexical and Neural Metrics Towards Robust Machine Translation Evaluation", accepted at EAMT 2023.

sparse_continuous_distributions

This repository provides open-source code for sparse continuous distributions and corresponding Fenchel-Young losses.

lmt_hallucinations

tower-eval

explainable-qe-shared-task

IST-Unbabel 2021 Submission for the Quality Estimation Shared Task

spectra-rationalization

Repository for SPECTRA: Sparse Structured Text Rationalization, accepted at EMNLP 2021 main conference.

crest

Code for CREST: A Joint Framework for Rationalization and Counterfactual Text Generation, accepted at ACL 2023.

non-exchangeable-crc

chunk-based_knn-mt

unn

Code for the paper "Modeling Structure with Undirected Neural Networks"

ot-hallucination-detection

qe-corpus-builder

Builds a WMT18-like corpus for word-level QE with annotations in the source and target words.

translation_llm

SIGMORPHON2019

IT-IST's submission to SIGMORPHON 2019 Task 1

translation-hypothesis-ensembling

speech-continuous-attention

Speech Classification using Continuous Attention Mechanisms

SSHN

Sparse and Structured Hopfield Networks

stopes

A library for preparing data for machine translation research (monolingual preprocessing, bitext mining, etc.) built by the FAIR NLLB team.

axolotl

Go ahead and axolotl questions

fairseq-m2m100

Facebook AI Research Sequence-to-Sequence Toolkit written in Python.

Megatron-DeepSpeed

Ongoing research training transformer language models at scale, including: BERT & GPT-2

megatron-LM-moe

robust-mt

towerllm-alignment

Code for alignment for the towerllm project.