Vikhrmodels/QVikhr-2.5-1.5B-Instruct-r

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:1.5BQuant:BF16Ctx Length:32kPublished:Feb 11, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Warm

QVikhr-2.5-1.5B-Instruct-r is a 1.5 billion parameter instruction-tuned causal language model developed by Vikhrmodels, specialized for Russian language tasks. It is based on the QVikhr-2.5-1.5B-Instruct-r architecture and has been specifically trained using the RuMath dataset. This model supports bilingual (Russian/English) interactions and is optimized for mathematical reasoning in Russian.

Loading preview...

QVikhr-2.5-1.5B-Instruct-r Overview

QVikhr-2.5-1.5B-Instruct-r is a 1.5 billion parameter instruction-tuned language model developed by the Vikhr Team. It is built upon the QVikhr-2.5-1.5B-Instruct-r base model and has undergone specialized training using the RuMath dataset, focusing on mathematical reasoning in Russian.

Key Capabilities

  • Specialized for Russian: Primarily optimized for Russian language understanding and generation.
  • Bilingual Support: Capable of handling both Russian and English inputs and outputs.
  • Mathematical Reasoning: Enhanced for tasks involving mathematical problem-solving due to its training on the RuMath dataset.
  • Instruction Following: Designed to follow instructions effectively, making it suitable for various interactive applications.

Training Details

The model's alignment phase utilized GRPO and was trained on the Vikhrmodels/russian_math dataset, alongside GSM8k. This targeted training contributes to its proficiency in mathematical contexts.

Use Cases

This model is particularly well-suited for applications requiring:

  • Russian-centric AI assistants: Where accurate and contextually relevant responses in Russian are crucial.
  • Educational tools: Especially for mathematics-related content in Russian.
  • Bilingual conversational agents: That need to operate in both Russian and English with a focus on instructional tasks.

For optimal generation quality, a recommended temperature of 0.4 is suggested.