Embedding model family that ranks well on retrieval benchmarks.
| Critère | BGE | Instructor |
|---|---|---|
| Dimensions | 1024 | 768 |
| License | MIT | Apache-2.0 |
| Max tokens | 512 | 512 |
| Open weights | Yes | Yes |
| Provider | BAAI | HKU NLP |
Les lignes grisées sont identiques pour les deux ; ce sont les lignes qui diffèrent qui motivent la recommandation ci-dessous.
Généré à partir des différences ci-dessus par des règles structurées, et non par un modèle.
Choose BGE if…
Aucun des attributs enregistrés ne le distingue à son avantage.
Choose Instructor if…
Aucun des attributs enregistrés ne le distingue à son avantage.
model · Embedding Models
BGE is a family of embedding models trained specifically for retrieval, available in sizes small enough to run locally. It solves the problem that a general-purpose embedding model is not optimised for the asymmetric comparison a search query and a document require. It exists because retrieval quality is set at the embedding step, so a model trained for it improves everything downstream without changing anything else.
model · Embedding Models
Instructor embeds text together with an instruction describing what the embedding is for, so one model serves retrieval, clustering and classification without fine-tuning for each. It solves the problem that a single fixed embedding cannot be optimal for several different tasks, and fine-tuning a copy per task is expensive and easy to get wrong. It exists because telling the model what kind of similarity you want is a cheap substitute for training it to want that.
