Oskar Hallström

WORK HISTORY

EDUCATION

ABOUT OSKAR HALLSTRÖM

Hello! I am working at the startup LightOn, the first pure GenAI company publicly listed in Europe, delivering secure GenAI solutions on premise for large enterprises. At LightOn, I head the Training and Inference team, where we train and adapt Vision and Large Language Models to perform well on enterprise use cases and retrieval, as well as to run efficiently given the GPU setups of our customers. In addition to post-training, I\'ve also worked on large scale pre-training of smaller models, such as ModernBERT which got 20M+ downloads during the first 5 months after it was released. In my earlier days at LightOn, I deep-dived into Reward Modeling and worked on Alfred-40B, one of the first ever open-sourced LLMs trained with PPO. Prior to joining LightOn in the beginning of 2023, I have worked at the Reasoning and Learning Lab at Linköping University, as well as done internships within AI R&D at Telenor (Telecommunications company, Norway) and Trafikverket Labs (AI Lab of the Swedish Transport Administration ). These experiences include working with Transformers and Large Language Models, but also other architectures such as GANs. I have a Master\'s degree in Computer Science specialized in Machine Learning from Linköping University (Sweden), which includes taking 65 credits in Machine Learning/Computer Science courses at EPFL (Switzerland). Are you interested in collaborating with my team at LightOn, discuss something, or even come and work for us? Please don\'t hesitate to shoot me a message!