• Working with large language models on supercomputers
• Measuring energy usage of GPU jobs

GitHub workspaces

Project partners in GreenNLP publish there software packages and tools on the following GitHub workspaces:

• Helsinki-NLP
• Turku-NLP
• CSC- IT Center for Science

Language modelling

• TurkuNLP Megatron-DeepSpeed fork
• LLM scaling on LUMI

Machine translation

• OpusTools for accessing parallel data in OPUS
• OpusFilter for parallel data curation
• OPUS-MT for deploying pre-trained MT models
• OPUS-MT-train for training NMT models
• OPUS-MT dashboard
• MAMMOTH for modular multilingual NMT

Tools for high-performance computing

• Machine learning scripts
• GPU energy usage counter for AMD/ROCm

Monolingual data

• The HPLT data sets

Parallel data

• OPUS - the open parallel corpus
• The Tatoeba translation challenge

Benchmarks

• MT testsets

Language models

• TurkuNLP models at huggingface
• Poro 34B from LumiOpen

Translation models

• OPUS-MT models (Marian-NMT-based)
• OPUS-MT models (converted to pytorch/trasformers)
• Marian-NMT models from the Tatoeba translation challenge