Dataset · Xnhyacinth
LongBench Long-Context Benchmark in Parquet Format
Xnhyacinth/LongBench
A parquet-repackaged set of English LongBench test configurations (including the _e version) for evaluating LLM performance on long texts.
What's inside
Contains text and tabular data for evaluating models on long-context tasks, including subsets with the _e suffix. The volume ranges from 1k to 10k English examples available as test splits. The configurations are converted to parquet format based on the original MIT-licensed LongBench repository.
What you can do with it
- Evaluating LLM quality when processing long English documents
- Comparing models on specialized subtasks from the LongBench-e configuration
- Testing context windows and information retrieval in large texts
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 21.6K
- Likes
- 7
How to load it
from datasets import load_dataset
ds = load_dataset("Xnhyacinth/LongBench")