Internship: Corpus Mixes for Data-Efficient LLM Pretraining

TNO; Den Haag; On-site; EUR 6,200 - 7,500; Full time

Benefits offered by this job

Professional internship environment

Dedicated supervision

Room to develop

Laptop provided

615€ monthly internship allowance

Leave entitlement

Travel reimbursement

Jong TNO membership

Job summary

TNO and SURF’s GPT-NL team invite a master’s student in AI, CS, or Data Science to join an internship starting between November 2026 and January 2027. You will define target tasks, domains, languages and compute budgets, then build controlled corpus mixtures for language-model experiments.

You will train and evaluate small language models, develop a method to predict performance from corpus composition, and validate findings in follow-up experiments.

Qualifications

• Master's student in Artificial Intelligence, Computer Science, Data Science, or a related field.
• Experience with Python and machine learning.
• Experience training language models is useful but not required if motivated to learn.
• Strong experimental design and attention to reproducibility.

Responsibilities

• Define target tasks, domains, languages and compute budget with supervisors.
• Create controlled corpus mixtures from Propella annotations and train/evaluate small language models.
• Develop methods (e.g. genetic algorithm) to predict performance from corpus composition.
• Validate predictions in new experiments and transfer conclusions to larger models or different tasks.
• Translate findings into a master’s thesis and supporting corpus-building software.

Skills

Python

ML experience

Experiment design

Education

Master's student in AI/CS/DS or related field

Tools

Git

Jupyter

Job description

TNO and SURF’s GPT-NL team invite a master’s student in AI, CS, or Data Science to join an internship starting between November 2026 and January 2027. You will define target tasks, domains, languages and compute budgets, then build controlled corpus mixtures for language-model experiments.

You will train and evaluate small language models, develop a method to predict performance from corpus composition, and validate findings in follow-up experiments.