PhD: Pioneering Foundation Models & Efficient LLM Training

University of St.Gallen; Schweiz; Vor Ort; CHF 60.000 - 90.000; Vollzeit

Benefits dieser Stelle

Fully funded PhD position

Modern compute infrastructure

Support for publishing at top-tier-con

Zusammenfassung

The University of St.Gallen (HSG) in Switzerland invites applications for a fully funded PhD position focused on foundation model training and efficient learning methods. You will pursue a doctoral thesis while researching training dynamics of large language models and incorporating linguistic insights into token weighting strategies.

The role includes teaching (~2 SWS per semester), supervision of Bachelor and Master theses, and supporting the chair in organizational and research tasks.

Qualifikationen

• Master's degree in computer science, ML, computational linguistics or related field.
• Desirable: prior research experience or publications in NLP, ML, or related areas.
• Strong programming skills in Python.
• Desirable: experience with PyTorch or TensorFlow/Keras.
• Experience or strong interest in training large language models (transformers).
• Experience with efficient training techniques (e.g., LoRA, quantization, sparsity, distillation).
• Knowledge of scaling laws, training dynamics, or evaluation of generative models.
• Experience with dataset construction and large-scale data processing.
• Desirable: background or interest in linguistics or computational linguistics, particularly tokenization and token weighting strategies.
• Ability to work independently, structure research, and manage time.
• Motivation to contribute to teaching and collaborative research.

Aufgaben

• You will work on your doctoral thesis in the area of foundation model training and efficient learning methods.
• You design and conduct research on training dynamics of large language models, including scaling behavior, optimization, and generalization.
• You incorporate linguistic insights into the development and analysis of language models, with a particular focus on token weighting strategies and their effects on training dynamics.
• You engage in teaching activities (approx. 2 SWS per semester), including tutorials, exercises, and supervision of Bachelor's and Master's theses and projects.
• You support the chair in organizational and research-related tasks.

Kenntnisse

Strong research mindset

English writing

Ausbildung

Master's degree in computer science / ML / computational linguistics or related field

Tools

Python

PyTorch

TensorFlow/Keras

Jobbeschreibung

The University of St.Gallen (HSG) in Switzerland invites applications for a fully funded PhD position focused on foundation model training and efficient learning methods. You will pursue a doctoral thesis while researching training dynamics of large language models and incorporating linguistic insights into token weighting strategies.

The role includes teaching (~2 SWS per semester), supervision of Bachelor and Master theses, and supporting the chair in organizational and research tasks.