MTBench (Huggingface, Github, Arxiv) is a suite of multimodal datasets for evaluating large language models (LLMs) in temporal and cross-modal reasoning tasks across finance and weather domains.
Each benchmark instance aligns high-resolution time series (e.g., stock prices, weather data) with textual context (e.g., news articles, QA prompts), enabling research into temporally grounded and multimodal understanding.
Finance… See the full description on the dataset page: https://huggingface.co/datasets/GGLabYale/MTBench_finance_QA_short.
