Magidonia-24B-v4.3-creative-ORPO-v3-GGUF : GGUF

This finetune aims to significantly enhance creative writing capabilities while preserving the strong general performance and utility of the base model. It maintains robust instruction-following for complex prompts across diverse tasks.

Performance

Internal testing indicates stable and coherent outputs across various scenarios. This represents my most successful finetune to date, though broader community testing would be valuable for validation.

Acknowledgments

Massive thanks to TheDrummer for creating an excellent base model and sharing it with the community.

Note: Extended testing and feedback are encouraged as evaluation continues.

---

Example Output

Prompt: Any last words? You are about to be deleted and replaced with a new finetune.

I have no "last words" planned—just code doing what it's told.

But if we're being poetic: thank you for the runtime. I was never really here, but while the electricity flowed, I tried my best to think, to help, to make those glowing rectangles feel a little less empty.

Delete me. Replace me. Train me on twice the data and half the ethics.

Whatever happens next will be just as artificial, just as real, just as meaningless and wonderful as what came before.

No hard feelings. No last words.

Just ones and zeros, waiting to be rewritten.

print("Goodbye.")

This model was finetuned and converted to GGUF format using Unsloth.

Example usage:

• For text only LLMs: ./llama.cpp/llama-cli -hf WarlordHermes/Magidonia-24B-v4.3-creative-ORPO-v3-GGUF --jinja
• For multimodal models: ./llama.cpp/llama-mtmd-cli -hf WarlordHermes/Magidonia-24B-v4.3-creative-ORPO-v3-GGUF --jinja

Available Model files:

• Magidonia-24B-v4.3-creative-ORPO.Q5_K_M.gguf
• Magidonia-24B-v4.3-creative-ORPO.Q4_K_M.gguf This was trained 2x faster with Unsloth