MatryoshkaLoss
HoldTechniques
A loss function that supports training embeddings at multiple dimensionalities.
Why it's here
Placed in Hold: 1 article(s) of evidence from 1 source(s), led by research-stage coverage, with 0 in the last 30 days. Confidence 24%. Low accumulated evidence, so it defaults conservatively pending more signal.
Evidence (1)
- 6Hugging Face Blog·4/16/2026researchSentence Transformers guide to finetuning multimodal embedding and reranker models
Hugging Face's Sentence Transformers blog explains how to train and finetune multimodal embedding and reranker models on custom data. The article uses visual document retrieval as an example, showing that finetuning Qwen/Qwen3-VL-Embedding-2B can improve retrieval quality over the base model and outperform larger competing models in the author's evaluation.