ZeroXClem/Qwen3-8B-HoneyBadger-EXP
ZeroXClem/Qwen3-8B-HoneyBadger-EXP is an 8 billion parameter experimental model fusion based on the Qwen3 architecture, created by ZeroXClem using the Model Stock merge method. It combines instruction-following, deep reasoning, creative roleplay, and code capabilities, leveraging several Qwen3-8B fine-tunes. This model is designed for versatile performance across symbolic reasoning, narrative generation, and technical comprehension, with a context length of 32768 tokens.
Loading preview...
ZeroXClem/Qwen3-8B-HoneyBadger-EXP: A Versatile Merged Model
ZeroXClem/Qwen3-8B-HoneyBadger-EXP is an 8 billion parameter experimental model developed by ZeroXClem. It is a fusion created using the Model Stock merge method from MergeKit, combining several Qwen3-8B-based fine-tunes to achieve a broad range of capabilities. The base model for this merge is AXCXEPT/Qwen3-EZO-8B-beta.
Key Capabilities
This model is engineered to excel in multiple domains:
- Deep Symbolic Reasoning: Enhanced through techniques from models like Shadow-FT.
- Immersive Roleplay & Storytelling: Benefits from merges like Bald-Beaver and CavesOfQwen, which loosen instruct bias for more natural narrative generation.
- Code Understanding & Generation: Supports Python, C++, and JavaScript.
- Structured Outputs: Capable of generating Markdown, JSON, and LaTeX.
- Conversational Alignment: Optimized for rich conversational interactions, drawing from Della-style merges.
- Instruction Following: Tuned for precise instruction adherence.
Optimal Usage
For best performance, users are advised to use the provided Ollama modelfile or a customized prompt with the default Qwen3 chat template. The model is also ChatML friendly.
Experimental Status
It is important to note that this is an experimental prototype and is not intended for production environments. It is best suited for research, prompt testing, and further fine-tuning workflows due to its early-stage development and potential for unaligned behaviors.