Developer Trains 113‑Million‑Parameter Earthquake LLM From Scratch

A developer announces training a 113‑million‑parameter language model for earthquake data. The model was built from absolute scratch without pre‑existing weights. The project leverages open‑source tools to handle seismic information. Training details include the scale of parameters and computational effort. The author shares a link to the code repository for public access. Early results suggest the model can generate plausible earthquake narratives. The work aims to advance AI applications in geoscience research. Future plans involve refining performance and expanding datasets.