GoLLIE
A large language model fine-tuned to follow annotation guidelines, improving zero-shot Information Extraction on schemas it has never seen. Built with HiTZ.
Oscar Sainz
I am a postdoctoral researcher at the University of the Basque Country in the HiTZ center and IXA group. My research interests are focused on low-resource scenarios and particularly on low-resource Information Extraction.
I obtained my PhD at the University of the Basque Country in 2024, where I worked on zero and few-shot Information Extraction using entailment models and making LLMs learn to follow annotation guidelines.
For more, visit my GitHub or Hugging Face page.
A large language model fine-tuned to follow annotation guidelines, improving zero-shot Information Extraction on schemas it has never seen. Built with HiTZ.
An open family of language models for Basque (7B-70B), with its pretraining corpus and evaluation suite. Best Resource Paper, ACL 2024.
A framework for zero-shot text classification built on textual entailment models - the basis for much of my early work on zero-shot Information Extraction.
Assessing Tool-Usage Capabilities of Basque Agentic RAG Systems
Synthetic Data Generation for Sentence-level and Paragraph-level Closed Information Extraction
Modelando las preferencias humanas en euskera, entrenamiento y evaluación
Evaluación de LLMs para lenguas de pocos recursos: el caso del occitano
Hizkuntzen arteko transferentzia analisia Euskarazko Informazio-Erauzketan
For more, visit my Google Scholar or Semantic Scholar page.