CoolFace
Modelpublic

0xA50C1A1/Ministral-3-14B-Nymphaea-RP

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face
7likes2.5kdownloads
Model Card

Ministral-3-14B-Nymphaea-RP

A fine-tune of Ministral 3 14B Instruct 2512 for roleplay and creative writing.

[!Tip] The SillyTavern preset is available here. For custom presets, please use the Mistral V7-Tekken instruct template.

<details> <summary><b>Chat Example</b></summary><p>

Tested at Q6_K quantization with the Web Search extension (via SearXNG) in SillyTavern.

SillyTavern Screenshot </p></details>

GGUF

Here is my custom mixed-quant GGUF, which I use regularly. It fits fine into 16GB VRAM with a 16K context window (using Q8 KV cache). If you need mmproj, it's available here.

<details> <summary>GGUF recipe</summary>

llama-quantize \
  --imatrix imatrix.gguf \
  --token-embedding-type q8_0 \
  --output-tensor-type q8_0 \
  --tensor-type ".*attn_q.weight=q8_0" \
  --tensor-type ".*attn_k.weight=q8_0" \
  --tensor-type ".*attn_output.weight=q5_k" \
  --tensor-type ".*attn_v.weight=iq4_nl" \
  --tensor-type ".*ffn_up.weight=iq4_nl" \
  --tensor-type ".*ffn_gate.weight=iq4_nl" \
  Ministral-3-14B-Nymphaea-RP.F16.gguf \
  Ministral-3-14B-Nymphaea-RP.Q5_Mix.gguf \
  q5_k

</details>

Imatrix file for making your own quants is available here. I used this calibration dataset to create it, expanding it with RP and creative writing data (about 400k tokens).

Training Notes

Trained on the latest iteration of my Darkmere dataset. This version features expanded genre variety, built upon a mix of manually curated synthetics and human-written stories.

[!IMPORTANT] The base weights are abliterated via Heretic prior to fine-tuning, so this fine-tune is quite uncensored.

<details> <summary>Training Specs</summary><p>

Method:

  • —Training Method: DoRA (Weight-Decomposed LoRA)
  • —Target Modules all-linear
  • —LoRA Rank: 64
  • —LoRA Alpha: 64
  • —LoRA Dropout: 0.05

Hyperparameters:

  • —Batch Size: 2 (Per-device)
  • —Gradient Accumulation: 2
  • —Epochs: 2
  • —Learning Rate: 1e-4
  • —Optimizer: adamw_torch_fused
  • —LR Scheduler: cosine
  • —Noise Level: neftune_noise_alpha=5
[!Note] The vision encoder was frozen during training, so the model retains its native vision capabilities.

</p></details>

Special Thanks

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

  • —[p-e-w](https://huggingface.co/p-e-w) for developing [Heretic](https://github.com/p-e-w/heretic) - an essential tool for censorship removal.
  • —[SicariusSicariiStuff](https://huggingface.co/SicariusSicariiStuff) for developing [SLOP_Detector](https://github.com/SicariusSicariiStuff/SLOP_Detector) script.
  • —[Mistral AI](https://mistral.ai/) for their [Ministral 3](https://huggingface.co/collections/mistralai/ministral-3) weights.
  • —[AMD](https://oneclickamd.ai/) for their Instinct™ MI300X GPU.