GenueAI/geode-onyx
TEXT GENERATIONConcurrency Cost:1Model Size:3.1BQuant:BF16Ctx Length:32kPublished:Apr 11, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

GenueAI/geode-onyx is a 3-billion parameter conversational AI model developed by Genue AI, fine-tuned from Qwen 2.5 3B Instruct. This model is part of the Geode family, optimized for a balance of logical reasoning and personality. It excels in tasks requiring identity and self-awareness, mathematical reasoning, general knowledge, and HTML/CSS/JS code generation.

Loading preview...

Genue Geode Onyx: A Balanced Conversational AI

GenueAI/geode-onyx is a 3-billion parameter conversational AI model developed by Genue AI, fine-tuned from the Qwen 2.5 3B Instruct base model. It is a key member of the Geode family, designed to offer a balanced blend of logical reasoning and personality, making it suitable for a variety of interactive applications.

Key Capabilities

  • Conversational AI: Optimized for engaging in natural dialogue.
  • Mathematical Reasoning: Handles arithmetic, algebra, and word problems effectively.
  • General Knowledge: Possesses a broad understanding of various topics.
  • Code Generation: Capable of generating HTML, CSS, and JavaScript code snippets.
  • Identity & Self-awareness: Fine-tuned on datasets that include identity-related conversational examples.

Training and Technical Details

Onyx was fine-tuned using LoRA (r=32, alpha=32) and achieved a training loss of 0.68. It operates with FP16 precision and is released under the Apache 2.0 License. Its prompt format is a simple "Direct Mode" (e.g., User: What is your name?\nBot:), making it straightforward to integrate with frameworks like Hugging Face Transformers.