README.md

metadata

dataset_info:
  features:
    - name: DATE
      sequence: string
    - name: temperature
      sequence: float32
    - name: dew_point_temperature
      sequence: float32
    - name: relative_humidity
      sequence: float32
    - name: station_level_pressure
      sequence: float32
    - name: sea_level_pressure
      sequence: float32
    - name: wind_speed
      sequence: float32
    - name: wind_direction
      sequence: float32
    - name: visibility
      sequence: float32
    - name: altimeter
      sequence: float32
    - name: precipitation_3_hour
      sequence: float32
    - name: precipitation_6_hour
      sequence: float32
    - name: precipitation_9_hour
      sequence: float32
    - name: precipitation_12_hour
      sequence: float32
    - name: precipitation_15_hour
      sequence: float32
    - name: precipitation_18_hour
      sequence: float32
    - name: precipitation_21_hour
      sequence: float32
    - name: precipitation_24_hour
      sequence: float32
  splits:
    - name: train
      num_bytes: 1173537799
      num_examples: 49
  download_size: 168154556
  dataset_size: 1173537799
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

MTBench: A Multimodal Time Series Benchmark

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 Datasets

🔹 Finance Domain

• MTBench_finance_news
  20,000 articles with URL, timestamp, context, and labels

• MTBench_finance_stock
  Time series of 2,993 stocks (2013–2023)

• MTBench_finance_aligned_pairs_short
  2,000 news–series pairs

  • Input: 7 days @ 5-min
  • Output: 1 day @ 5-min

• MTBench_finance_aligned_pairs_long
  2,000 news–series pairs

  • Input: 30 days @ 1-hour
  • Output: 7 days @ 1-hour

• MTBench_finance_QA_short
  490 multiple-choice QA pairs

  • Input: 7 days @ 5-min
  • Output: 1 day @ 5-min

• MTBench_finance_QA_long
  490 multiple-choice QA pairs

  • Input: 30 days @ 1-hour
  • Output: 7 days @ 1-hour

🔹 Weather Domain

• MTBench_weather_news
  Regional weather event descriptions

• MTBench_weather_temperature
  Meteorological time series from 50 U.S. stations

• MTBench_weather_aligned_pairs_short
  Short-range aligned weather text–series pairs

• MTBench_weather_aligned_pairs_long
  Long-range aligned weather text–series pairs

• MTBench_weather_QA_short
  Short-horizon QA with aligned weather data

• MTBench_weather_QA_long
  Long-horizon QA for temporal and contextual reasoning

🧠 Supported Tasks

MTBench supports a wide range of multimodal and temporal reasoning tasks, including:

• 📈 News-aware time series forecasting
• 📊 Event-driven trend analysis
• ❓ Multimodal question answering (QA)
• 🔄 Text-to-series correlation analysis
• 🧩 Causal inference in financial and meteorological systems

📄 Citation

If you use MTBench in your work, please cite:

@article{chen2025mtbench,
  title={MTBench: A Multimodal Time Series Benchmark for Temporal Reasoning and Question Answering},
  author={Chen, Jialin and Feng, Aosong and Zhao, Ziyu and Garza, Juan and Nurbek, Gaukhar and Qin, Cheng and Maatouk, Ali and Tassiulas, Leandros and Gao, Yifeng and Ying, Rex},
  journal={arXiv preprint arXiv:2503.16858},
  year={2025}
}