LongBench is a bilingual, multitask benchmark for evaluating long-context understanding in large language models. It covers scenarios including single-document QA, multi-document QA, summarization, few-shot learning, synthetic tasks, and code completion. The dataset was authored by GinkgoQ and repackaged for Hugging Face.
Use Cases
- Benchmarking model performance on single-document question answering tasks.
- Evaluating model capability on multi-document question answering tasks.
- Testing summarization quality for long documents.
- Assessing few-shot learning performance in long-context scenarios.
- Evaluating code completion for long code snippets.
Strengths
- Benchmark covers multiple long-text application scenarios.
- Dataset is bilingual, supporting evaluation across languages.
- Last updated on 2026-05-24, indicating recent maintenance.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- GinkgoQ
- Freshness
- Last updated 2026-05-24 08:44:33.