Darkmere-14B-v0.1

A fine-tune of Ministral 3 14B Instruct 2512 for roleplay and creative writing. The 8B version is available here.

Why this exists

There's a noticeable gap between 8B and 24B LLMs fine-tuned for roleplay. Aside from old Mistral Nemo 12B fine-tunes, there are very few options that comfortably fit into 16GB VRAM at decent quantization. This model is an attempt to utilize that 14B "sweet spot" for a better RP experience.

The SillyTavern preset is available here.

Training Notes

Trained on a small private dataset I've been building for the last month or two. It's a mix of manually cleaned synthetic data, human-written stories, and RP logs. While the dataset leans toward a Dark Fantasy aesthetic, this fine-tune is versatile and behaves well across various genres.

• Training method: Full fine-tuning (not LoRA)
• Context Length: 16384
• Learning Rate: 5e-6
• Vision: the vision encoder was frozen during training, so the model retains its native vision capabilities.

Special Thanks

This fine-tune wouldn't be possible without the incredible work of the community:

• p-e-w for developing Heretic - an essential tool for censorship removal.
• Mistral AI for their Ministral 3 weights.
• AMD for their Instinct™ MI300X GPU.