The Challenges and Opportunities in Delivering Large Language Models to Australian Researchers

Large language models (LLMs) such as chatGPT have the potential to massively augment the way in which research is conducted. The Melbourne eResearch Group at The University of Melbourne (www.eresearch.unimelb.edu.au) have been exploring LLMs and the capabilities they can bring for over a year. This presentation will introduce LLMs and the practical challenges that are faced in hosting and operating production level LLM capabilities for the broader research community. Specifically the talk will cover: the benefits of direct deployment and hosting of LLMs, i.e. rather than direct use of the openAI API; the opportunities of LLMs with example case studies in their use; the challenges of hosting LLMs on national Cloud platforms such as NeCTAR and the demands this gives rise to on the underpinning technologies and hardware; supporting current and future LLM deployments and the technological requirements this gives rise to. The talk will focus especially on the NeCTAR Cloud-based deployment of LLAMA3 and Mistral (https://kalliope.pages.gitlab.unimelb.edu.au/chat/) two open sources LLMs.