R

Ruadaptqwen2.5 7B Lite Beta

Developed by RefalMachine
This project adapts the T-lite-it-1.0 model to Russian. By replacing the tokenizer, continuing pre-training on the Russian corpus, and applying LEP technology, the generation speed of Russian text is improved.
Downloads 1,603
Release Time : 1/27/2025

Model Overview

A text generation model adapted to Russian based on the Qwen/Qwen2.5 - 7B and t-tech/T-lite-it-1.0 models, supporting Russian text generation tasks.

Model Features

Optimization of the Russian tokenizer
Replace the tokenizer with an extended tiktoken cl100k (extended by a 48k unigram tokenizer), significantly improving the generation speed of Russian text.
Continued pre-training on the Russian corpus
Continue pre-training on Russian datasets such as IlyaGusev/saiga_scored, IlyaGusev/saiga_preferences, and dichspace/darulm.
Application of LEP technology
Apply LEP (Learned Embedding Propagation) technology to optimize the model performance.
Improvement in generation speed
Due to the adoption of a new tokenizer, the generation speed of Russian text is 60% faster than that of the original T-lite-it-1.0 model.

Model Capabilities

Russian text generation
Natural language processing

Use Cases

Text generation
Russian text generation
Generate high-quality Russian text content
The generation speed is improved by 60%
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
Š 2025AIbase