Projects

Projects

3CPO: Comprehensive Cross-lingual Contextual Pretraining \& Optimization for Language Models

This project proposes to train language models with language diversity as a first priority. We adapt the data selection, tokenizer, and evaluation to reflect diversity in all aspects of model development.

Funded by: Nvidia Academic Grant Programme

LMLM: Linguistically Motivated Language Models

LMLM proposes to build Linguistically Motivated LLMs. This is done by adapting the training regime as well as the LLM architecture to incorporate different linguistic levels. This will lead to mitigation of shortcut learning and more reliable LLMs.

Funded by: Carlsberg Semper Ardens: Accelerate