602 manually constructed samples in multiple-choice format assess large language models' knowledge of French idiomatic expressions in context. The EIFFEL benchmark was created by native speakers and linguists from OpenLLM-France. The dataset was last updated on June 8, 2026.
Use Cases
- Benchmarking LLM performance on French idiomatic language based on the multiple-choice format
- Fine-tuning LLMs for improved natural language understanding in French based on contextual expression completion
- Linguistic research on idiom usage and comprehension based on the annotated samples
Strengths
- 602 manually constructed and annotated samples provide a focused evaluation set
- Samples were created and annotated by native speakers and linguists, suggesting quality control
Limitations
- Column-level documentation is absent; field semantics must be inferred after download
- Row count is unknown, which may limit suitability assessment
Provenance
- Source
- OpenLLM-France
- Collection Method
- Samples were manually constructed and annotated by native speakers and linguists.
- Freshness
- Last updated 2026-06-08 13:32:42; freshness should be verified