Chunky-Lemon-Cookie-11B: A Merged Language Model

Chunky-Lemon-Cookie-11B is a 10.7 billion parameter language model developed by FallenMerick. It is the result of a sophisticated merge operation combining several established pre-trained models, including Mistral-7B-v0.1, SanjiWatsuki/Kunoichi-7B, SanjiWatsuki/Silicon-Maid-7B, KatyTheCutie/LemonadeRP-4.5.3, and Sao10K/Fimbulvetr-11B-v2.

Merge Details

This model was constructed using mergekit with a combination of passthrough and task arithmetic methods. The merging process involved specific layer ranges from the constituent models, indicating a tailored approach to combine their strengths.

Performance Highlights

Evaluated on the Open LLM Leaderboard, Chunky-Lemon-Cookie-11B achieved an average score of 70.23. Key benchmark results include:

• AI2 Reasoning Challenge (25-Shot): 69.62
• HellaSwag (10-Shot): 86.55
• MMLU (5-Shot): 65.35
• TruthfulQA (0-shot): 61.59
• Winogrande (5-shot): 79.79
• GSM8k (5-shot): 58.45

These scores indicate a balanced performance across various reasoning, common sense, and language understanding tasks.

Good for:

• Applications requiring a general-purpose language model with a 10.7B parameter count.
• Tasks benefiting from a model merged from diverse foundational architectures.
• Exploration of models created via advanced merging techniques like task arithmetic.