# Training and Finetuning Multimodal Embedding & Reranker Models with Sentence Transformers

Canonical URL: https://www.traeai.com/articles/09858152-e8b5-4faa-8e78-c997ec12c196
Original source: https://huggingface.co/blog/train-multimodal-sentence-transformers
Source name: Hugging Face Blog
Content type: article
Language: 英文
Score: 8.7
Reading time: 17 分钟
Published: 2026-04-16T00:00:00+00:00
Tags: Sentence Transformers, 多模态学习, 模型微调, 嵌入模型, Hugging Face

## Summary

文章详解如何使用 Sentence Transformers 微调多模态嵌入与重排序模型，并以视觉文档检索任务为例展示显著性能提升。

## Key Takeaways

- 微调多模态嵌入模型可显著提升特定任务（如视觉文档检索）的检索效果
- 使用 CachedMultipleNegativesRankingLoss 和 MatryoshkaLoss 可高效训练多模态模型
- 2B 参数的微调模型在 VDR 任务上超越 4 倍规模的基线模型

## Citation Guidance

When citing this item, prefer the canonical traeai article URL for the AI-readable summary and include the original source URL when discussing the underlying source material.