Lunzima/NQLSG-Qwen2.5-14B-OriginalFusion
Lunzima/NQLSG-Qwen2.5-14B-OriginalFusion is a 14.8 billion parameter language model created by Lunzima through a Model Stock merge of multiple Qwen2.5-14B variants and other 14B models, including specialized instruction-tuned and coder versions. Utilizing a 32768 token context length, this model is designed to leverage the strengths of its diverse constituent models, aiming for enhanced general instruction following and potentially improved coding capabilities due to the inclusion of Qwen2.5-Coder-14B-Instruct. Its primary use case is for applications requiring a robust, multi-faceted LLM that combines various optimizations from its merged components.
Loading preview...
Overview
Lunzima/NQLSG-Qwen2.5-14B-OriginalFusion is a 14.8 billion parameter language model developed by Lunzima. This model was created using the Model Stock merge method, building upon Lunzima/NQLSG-Qwen2.5-14B-MegaFusion-v8 as its base. The merge integrates a diverse set of 14B models, including several variants of Qwen2.5-14B-Instruct and Qwen2.5-Coder-14B-Instruct, alongside other specialized models like DeepSeek-R1-Distill-Qwen-14B and Lamarckvergence-14B.
Key Capabilities
- Enhanced Instruction Following: Benefits from the fusion of multiple instruction-tuned Qwen2.5-14B models.
- Potential for Improved Code Generation: Incorporates
Qwen/Qwen2.5-Coder-14B-Instruct, suggesting stronger performance in coding tasks. - Broadened General Knowledge: Combines knowledge from various base models, aiming for a more comprehensive understanding.
- Robustness: Leverages the Model Stock merge method, which is designed to effectively combine the strengths of multiple pre-trained models.
Good For
- Applications requiring a versatile instruction-following model.
- Development tasks that benefit from a model with strong coding capabilities.
- Scenarios where a fusion of diverse model strengths is advantageous for general-purpose AI tasks.