BEIR
Heterogeneous retrieval benchmark for zero-shot evaluation.
Überblick
BEIR evaluates retrieval models across eighteen datasets from different domains, without training on them. It solves the problem that a model tuned for one collection looks excellent on it and generalises poorly, which is invisible when you only test where you trained. It exists as the benchmark that made general-purpose retrieval a measurable claim rather than an assumed one.
Technologiegraph
Wie BEIR mit dem übrigen Stack zusammenhängt.
Nutzt und hängt ab von
- MTEB benchmark Benchmarks MTEB based on BEIR High ✓ source
Angewendet auf
- AI Search application Applications BEIR used for AI Search High ✓ source
- Embedding concept Techniques BEIR used for Embedding High ✓ source