Featuring time-series evaluation samples across multiple skill categories formatted for Large Language Model benchmarking. Each record provides structured temporal sequences including history_values, future_values, and associated metadata such as frequency and timestamps.
Use Cases
- Benchmark time-series forecasting accuracy by prompting models with history_values and validating against future_values.
- Assess the impact of data granularity on model performance using the frequency field.
- Categorize model failures across different temporal reasoning types using the skill labels.
Strengths
- Includes history_start and history_end timestamps to provide precise temporal boundaries for each sample.
- Features a skill column that categorizes the types of context-based reasoning required for the evaluation.
- Provides history_values and future_values as string-formatted sequences compatible with text-based model inputs.
- Stored in Parquet format for high compatibility with modern data processing and LLM training pipelines.