PsychoSafe
Pipeline for building and evaluating psychologically informed refusal behavior in large language models through data creation, prompting, fine-tuning, and evaluation.
Repositories
Open-source code maintained by OdenseNLP. Edit entries in `_data/repositories.yml`.
PropMe
Framework for evaluating memorization and propensity-aware memorization of training data in large language models.
DaLA
Source code for creating the Danish Corpus of Linguistic Acceptability, designed to evaluate Danish linguistic acceptability with real-world errors.
bitlinear
Production-ready implementation of 1.58-bit layers for quantization-aware training and efficient inference.
brainsurgery
Swiss-army-knife style toolkit for scripted tensor surgery on model checkpoints, with YAML plans, CLI workflows, and a Web UI.
DeToNATION
A communication framework that optimizes distributed AI training by reducing latency bottlenecks and handling heterogeneous clusters.
JustEval
A streamlined framework for evaluating language generation quality in two steps: generation and automatic evaluation.
mldataforge
Swiss-army-knife toolkit for transforming and processing machine learning datasets, including large-scale format conversion and splitting.
mlgroom
Utility for grooming and managing machine learning job queues.
projectives
AI-assisted analyser for projective techniques in qualitative health research.
SDU-Daisy
Benchmark and tooling for evaluating LLM understanding of Danish culture using closed question-answer pairs grounded in the Danish Culture Canon.