Overview

This model, djuna/Q2.5-Veltha-14B-0.5, is a 14.8 billion parameter language model developed by djuna. It was created using the della_linear merge method via mergekit, combining multiple pre-trained Qwen2.5-14B based models with arcee-ai/SuperNova-Medius as its base.

Merge Details

The merge process integrated the following models:

• huihui-ai/Qwen2.5-14B-Instruct-abliterated-v2
• allura-org/TQ2.5-14B-Aletheia-v1
• EVA-UNIT-01/EVA-Qwen2.5-14B-v0.2
• v000000/Qwen2.5-Lumen-14B

Performance

Evaluations on the Open LLM Leaderboard show an average score of 39.96. Specific metrics include:

• IFEval (0-Shot): 77.96
• BBH (3-Shot): 50.32
• MATH Lvl 5 (4-Shot): 33.84
• MMLU-PRO (5-shot): 47.72

Use Cases

Given its merged architecture and general performance, this model is well-suited for a variety of natural language processing tasks, particularly those benefiting from the combined capabilities of its constituent Qwen2.5-14B base models. Its 131072 token context length supports handling extensive inputs.