ehristoforu/RQwen-v0.1
ehristoforu/RQwen-v0.1 is a 14.8 billion parameter Qwen2 Instruct model developed by ehristoforu, based on Qwen/Qwen2.5-14B-Instruct. It features reflection tuning and is optimized for logic and deep context work, supporting both English and Russian languages. This model is fine-tuned using Unsloth (Transformers SFT) and aims to provide enhanced reasoning capabilities.
Loading preview...
RQwen-v0.1 Overview
RQwen-v0.1 is a 14.8 billion parameter instruction-tuned language model developed by ehristoforu. It is built upon the Qwen/Qwen2.5-14B-Instruct base model, utilizing the Qwen2 Instruct (ChatML) architecture. A key differentiator for RQwen-v0.1 is its focus on reflection tuning and enhanced capabilities for logic and deep work with context, making it suitable for tasks requiring nuanced understanding and reasoning.
Key Capabilities & Features
- Multilingual Support: Optimized for both English and Russian languages.
- Contextual Understanding: Designed for deep engagement with context, improving coherence and relevance in responses.
- Reflection Tuning: Incorporates specific tuning for reflective processes, potentially leading to more thoughtful and reasoned outputs.
- Training Methodology: Trained using Unsloth (Transformers SFT) for efficient fine-tuning.
Performance Insights
Evaluations on the Open LLM Leaderboard show an average score of 32.48. Notable scores include 76.25 on IFEval (0-Shot) and 48.49 on BBH (3-Shot), indicating its ability in instruction following and general reasoning. While its MATH Lvl 5 (4-Shot) score is 2.95, its strengths lie more in general language understanding and logical processing.
Good For
- Applications requiring strong contextual understanding and logical reasoning.
- Use cases involving both English and Russian language processing.
- Scenarios where reflective capabilities in an LLM are beneficial.