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.
📦 MTBench… See the full description on the dataset page: https://huggingface.co/datasets/GGLabYale/MTBench_weather_aligned_pairs_long_multi.
