metadata

license: mit
base_model:
  - Qwen/Qwen2.5-7B-Instruct
tags:
  - merge

Qwen2.5-7B-CABS-v2: A Merged Model Using a Novel Technique

This model is a merged version of Qwen/Qwen2.5-7B-Instruct utilizing a novel model merging technique to optimize performance while maintaining robustness across tasks.

Official Performance

This model has been officially evaluated on open_llm_leaderboard, achieving the following results:

IFEVAL	BBH	MATH	GPQA	MUSR	MMLU-PRO	AVG
74.18	36.28	49.02	7.61	14.86	37.75	36.61

As of February 19, 2025, this model ranks 3rd among all 7B and smaller models on open_llm_leaderboard.

For the latest and complete leaderboard rankings, please refer to the open_llm_leaderboard.

Recipe Coming Soon

We will release details on the merging technique and methodology soon, Stay tuned for updates! 🚀