Is an approach in training Embedding models for Matryoshka Embeddings.

The structure is as such:
Generate an Embedding Model
Generate a loss function to determine the value of the embedding
During training, the optimizer will shift the weights so loss function is minified. In matryoshka embedding models, loss will be calculated at different truncate points, summing up this value will generate a otal loss value, and that's the number we're trying to minimize

2024 12 30 03 42 50 - 🪆 Introduction to Matryoshka Embedding Models
SentenceTransformers Documentation — Sentence Transformers documentation
Paper page - Matryoshka Representation Learning