This position is part of an ambitious initiative of the HiTZ center, funded, among others, by the Basque and Spanish Governments, to improve the performance of Large Language Models on languages with low-resource corpora, including Basque. The motivation is that the amount of digital text, instructions and preference data available for most of the languages is orders of magnitude smaller than the text available for the largest languages. Current techniques for building Large Language Models like GPT or Llama require massive amounts of text, and if fact, their performance for smaller languages is significantly worse.

The candidate will join a thriving team of other PhDs and researchers focused on the limitations of current Large Language Models like GPT, and how to overcome them. As a sample please check the recently accepted ACL paper at arxiv, as well as others in our website.

The candidate will do research on methods to reduce the carbon footprint of LLMs on deployment. The candidate will gain first-hand experience on building, fine-tuning and deploying LLMs from forefront experts in the center, both from academia and industry.