theprint/ReWiz-Qwen-2.5-14B
theprint/ReWiz-Qwen-2.5-14B is a 14.8 billion parameter Qwen2 model developed by theprint, fine-tuned from unsloth/qwen2.5-14b-bnb-4bit. This model was trained 2x faster using Unsloth and Huggingface's TRL library, offering an efficient implementation of the Qwen2 architecture. It is suitable for general language tasks, with specific benchmark scores available for reasoning and knowledge-based evaluations.
Loading preview...
Model Overview
theprint/ReWiz-Qwen-2.5-14B is a 14.8 billion parameter language model developed by theprint, based on the Qwen2 architecture. It was fine-tuned from unsloth/qwen2.5-14b-bnb-4bit and notably trained 2x faster by leveraging Unsloth and Huggingface's TRL library, highlighting an optimization in the training process.
Key Capabilities & Performance
This model's performance has been evaluated on the Open LLM Leaderboard, providing insights into its general language understanding and reasoning abilities. Key benchmark results include:
- Average Score: 29.64
- IFEval (0-Shot): 27.85
- BBH (3-Shot): 44.86
- MATH Lvl 5 (4-Shot): 26.89
- GPQA (0-shot): 17.34
- MuSR (0-shot): 15.44
- MMLU-PRO (5-shot): 45.47
These scores indicate its proficiency across various tasks, including complex reasoning, mathematical problem-solving, and general knowledge. The model's efficient training methodology makes it a notable option for developers seeking Qwen2-based models with optimized development cycles.
Good For
- Applications requiring a Qwen2-based model with 14.8 billion parameters.
- Use cases where efficient training and deployment are priorities.
- General language generation and understanding tasks, as indicated by its benchmark performance.