Overview

FelixChao/NinjaDolphin-7B is a 7 billion parameter language model developed by FelixChao, created through a merge of several specialized models. It combines beowolx/CodeNinja-1.0-OpenChat-7B and beowolx/MistralHermes-CodePro-7B-v1 with FelixChao/WizardDolphin-7B as its base model, utilizing the dare_ties merge method.

Key Capabilities

• Enhanced Coding Performance: The model is specifically engineered to improve coding abilities, building upon the foundation of WizardDolphin-7B.
• HumanEval Benchmark: Achieves a score of 52.439% on the HumanEval Python benchmark (uninstructed and without post-processing), indicating strong code generation capabilities.
• General Reasoning: Demonstrates solid performance across various benchmarks, including an average score of 69.74 on the Open LLM Leaderboard, with notable results in AI2 Reasoning Challenge (65.61) and HellaSwag (85.35).

Good For

• Code Generation: Ideal for tasks requiring the generation of Python code.
• Programming Assistance: Can be used in applications where improved coding ability is a primary requirement.
• Research and Development: Suitable for researchers exploring model merging techniques and their impact on specific capabilities like coding.