R

Rubert Mini Frida

Developed by sergeyzh
A lightweight and fast modified version of the FRIDA model for computing embedding vectors of Russian and English sentences
Downloads 1,203
Release Time : 3/2/2025

Model Overview

This model is implemented by distilling the embeddings of FRIDA (embedding dimension 1536, 24 layers) into rubert-mini-sts (embedding dimension 312, 7 layers), primarily used for embedding computation and similarity comparison of Russian and English sentences.

Model Features

Lightweight and efficient
Significantly reduces model size (from 24 layers to 7 layers) through distillation while maintaining good performance
Multilingual support
Supports sentence embedding computation for both Russian and English
Prefix functionality
Inherits FRIDA's prefix functionality, allowing optimization for specific tasks with different prefixes
Mean pooling
Replaces FRIDA's CLS pooling with mean pooling, making it more suitable for sentence similarity tasks

Model Capabilities

Compute sentence embedding vectors
Russian sentence similarity comparison
English sentence similarity comparison
Text classification support
Information retrieval support

Use Cases

Text similarity
Paraphrase identification
Identify whether two sentences express the same meaning in different ways
Achieved a similarity score of 0.94 on the test set
Semantic search
Build a semantic search engine to match queries with documents
Achieved NDCG@10 of 0.721 in news retrieval tasks
Classification tasks
Sentiment analysis
Classify sentiment tendencies in Russian texts
Achieved an accuracy of 0.658 in Russian review classification tasks
Topic classification
Classify topics of Russian news articles
Achieved an accuracy of 0.880 in news headline classification tasks
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase