Text embedding model family trained on weakly supervised pairs.
| Critère | E5 | GTE |
|---|---|---|
| Dimensions | 1024 | 1024 |
| License | MIT | Apache-2.0 |
| Max tokens | 512 | 8192 |
| Open weights | Yes | Yes |
| Provider | Microsoft | Alibaba |
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.
model · Embedding Models
E5 is an embedding model family trained on large collections of weakly supervised text pairs, with the task described in the input so one model serves queries, passages and classification. It solves the problem that an embedding model trained on one kind of pair transfers poorly to another. It exists because describing the task in the text is a cheap way to make one model cover several retrieval settings.
model · Embedding Models
GTE is a family of text embedding models trained for retrieval, with variants covering English and multilingual use and long inputs. It solves the same problem as the other retrieval-focused embedding families: a general-purpose encoder is not optimised for the asymmetric comparison of a short query against a long document. It exists because retrieval quality is decided at the embedding step, and a family that scores well on public retrieval benchmarks is the cheapest way to improve everything downstream.
