Model reference · open weights

dragon-plus-query-encoder

dragon-plus-query-encoder is an open-weight embedding model from Meta. dragon-plus-query-encoder (BF16) weighs 438 MB; the smallest configuration that runs it is RTX 3060 12 GB.

• dragon-plus-query-encoder is a BERT-base sized dense retriever developed by Meta for feature extraction tasks.
• It is part of the DRAGON+ model, which uses an asymmetric dual encoder architecture initialized from RetroMAE and trained on augmented MS MARCO data.
• The model supports a context length of 512 tokens.

Summary of the facebook/dragon-plus-query-encoder model card, 2026-10-01

Hugging Face ↗GitHub ↗Paper ↗