Model Overview

allknowingroger/QwenSlerp5-14B is a 14.8 billion parameter language model developed by allknowingroger. It was created using the SLERP (Spherical Linear Interpolation) merge method via mergekit, combining two base models: CultriX/Qwestion-14B and CultriX/SeQwence-14Bv1.

Merge Configuration

The merge utilized a specific t parameter configuration [0, 0.5, 1, 0.5, 0], which implies a V-shaped curve for layer contributions. This setup was designed to integrate specific characteristics from the base models, such as "Hermes for input & output" and "WizardMath in the middle layers," suggesting an optimization for both general understanding and potentially mathematical reasoning.

Performance Metrics

Evaluations on the Open LLM Leaderboard provide insights into its capabilities. Key scores include:

• Avg.: 38.94
• IFEval (0-Shot): 71.19
• BBH (3-Shot): 47.39
• MATH Lvl 5 (4-Shot): 33.16
• MMLU-PRO (5-shot): 48.78

These results indicate its performance across various reasoning, instruction following, and knowledge-based tasks. Detailed evaluation results are available here.