In this project, we will extend our recent work using transformer models for abstract symbolic mathematics in particle physics (https://arxiv.org/pdf/2501.09729). Our previous NAISS allocation allowed us to produce a paper demonstrating, for the first time, that transformers can accurately predict Lagrangians from particle lists. This work was presented at several conferences and has initiated multiple new collaborations. Building upon those results, we aim to refine our transformer models to handle realistic scenarios relevant for state-of-the-art phenomenology studies. Furthermore, we will broaden our approach beyond particle physics Lagrangians toward a wider class of symbolic mathematics tasks, leveraging recent advances in structured mathematical reasoning and equation manipulation. During the previous NAISS project, our models worked so well that we shifted to larger-scale training sooner than expected, and additional Google Cloud research credits further reduced our resource needs. As a result, NAISS usage was lower than initially projected. Now, having confirmed that NAISS is the best platform for our needs, and with multiple new collaborations and more ambitious projects, we anticipate significantly higher usage in this cycle.