R

Rubert Mini Uncased

Developed by sergeyzh
This model is used to compute embedding vectors for Russian and English sentences, obtained by distilling the embedding vectors from ai-forever/FRIDA. The model is of the uncased type, meaning it does not distinguish between uppercase and lowercase letters in the text.
Downloads 724
Release Time : 3/25/2025

Model Overview

This model is used to compute embedding vectors for Russian and English sentences, obtained by distilling the embedding vectors from FRIDA. The embedding vector size is 384, with 7 layers and a context size of 512 tokens. The model supports various prefix functionalities to enhance performance across different tasks.

Model Features

Multilingual support
Supports embedding vector computation for Russian and English sentences.
Prefix functionality
Inherits multi-task prefix functionality from FRIDA, allowing performance optimization for different tasks.
Mini model
Lightweight design with an embedding vector size of 384 and 7 layers, suitable for resource-constrained environments.
Case-insensitive
Uncased type, meaning it does not distinguish between uppercase and lowercase letters in the text.

Model Capabilities

Compute sentence embedding vectors
Semantic text similarity calculation
Paraphrase identification
Natural language inference
Sentiment analysis
Toxicity identification

Use Cases

Text similarity
Search query matching
Optimize the matching of search queries with documents using the search_query prefix.
Achieved an NDCG@10 score of 0.791 in the ruMTEB benchmark.
Paraphrase identification
Identify semantically similar sentences using the paraphrase prefix.
Scored 0.760 in paraphrase identification tasks.
Text classification
Sentiment analysis
Perform sentiment classification using the categorize_sentiment prefix.
Scored 0.798 in sentiment analysis tasks.
Topic classification
Perform topic classification using the categorize_topic prefix.
Achieved an accuracy of 0.884 in headline classification tasks.
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase