Multibody Models Generated from Natural Language-Based Text

EasyChair Preprint 13428

Abstract

Large Language Models (LLMs) are currently experiencing a high level of attention due to their extraordinary skills. They are experiencing strong growth and new applications with high benefits for industry and business are added every day. In this work, we investigate the current capabilities of LLMs related to the generation of multibody dynamics models from natural language inputs. In particular, we investigate LLMs that have been trained on our specialized multibody code, Exudyn, and which are able to capture and translate the complexities of kinematics and dynamics into functional programming interfaces. Additionally, we investigate the fine tuning of open-source LLMs from HuggingFace and compare the performance with closed-source models.

Keyphrases: Generative Pre-trained Transformers, Large Language Model, simulation model

@booklet{EasyChair:13428,
  author    = {Johannes Gerstmayr and Peter Manzl and Michael Pieber},
  title     = {Multibody Models Generated from Natural Language-Based Text},
  howpublished = {EasyChair Preprint 13428},
  year      = {EasyChair, 2024}}