risenyard/egms-qa-dataset
Viewer • Updated • 690k • 1.18k • 1
EGMS-QA dataset, frozen spatio-temporal encoder, and four host-LLM translators for natural-language ground-motion monitoring questions.
Note QA splits, 10,000 model-ready 294-step EGMS tiles, labels, encoder tokens, reference tables, and integrity metadata.
Note Frozen spatio-temporal encoder mapping each 7 km EGMS tile to 65 × 256 tokens.
Note Projectors and LoRA adapters for EGMS tokens with Qwen, Gemma, Llama, and Mistral.