Coma-II-14B Overview

Coma-II-14B is a 14.8 billion parameter model built on the Qwen 2.5 architecture, specifically engineered to significantly improve reasoning capabilities in 14B-parameter models. It is fine-tuned using a long chain-of-thought reasoning approach and specialized datasets, making it highly effective for general-purpose reasoning and complex question-answering.

Key Capabilities

• Enhanced General Knowledge: Provides broad and accurate knowledge across diverse domains.
• Improved Instruction Following: Excels at understanding and executing complex instructions, generating structured and coherent responses.
• Versatile Adaptability: Handles a wide array of topics and conversation styles, from open-ended to highly structured inquiries.
• Long-Context Support: Processes up to 128K input tokens and generates up to 8K output tokens, ideal for detailed and extended interactions.
• Multilingual Proficiency: Supports over 29 languages, including major global languages like English, Chinese, French, Spanish, German, and Japanese.

Good for

• General-Purpose Reasoning: Assisting with logical deduction, problem-solving, and diverse question-answering.
• Educational & Informational Assistance: Generating explanations, summaries, and research-based content.
• Conversational AI & Chatbots: Building intelligent agents requiring deep contextual understanding.
• Multilingual Applications: Facilitating global communication, translation, and content generation across languages.
• Long-Form Content Generation: Producing extended coherent outputs such as articles, reports, and guides.
• Structured Data Processing: Analyzing and generating structured outputs like tables and JSON.

Performance Highlights

Evaluations on the Open LLM Leaderboard show an Average score of 39.48%, with notable performance in areas like BBH (3-Shot) at 46.89% and MATH Lvl 5 (4-Shot) at 55.14%. Detailed results are available on the Open LLM Leaderboard.