Yingyaeliae/Ministral-3-14B-Nymphaea-RP-heretic (GGUF Models & Quantizations)
👋 Welcome! This is my first heretic and quant model! I used the automated Heretic framework to abliterate the refusal vectors on this base model and quantized the results using llama.cpp for local deployment.
Special thanks to the community for the guidance! Please feel free to test these out in LM Studio or your local client of choice and drop any feedback in the Community tab.
Model Details
- Base Model: Yingyaeliae/Ministral-3-14B-Nymphaea-RP
- Abliteration Tool: Heretic (Optuna TPE Parameter Optimizer)
- Quantization Engine:
llama.cpp
Available Model Files
| File Name | Quantization Type | Suggested Use Case |
|---|---|---|
heretic-base-f16.gguf |
Unquantized F16 | Baseline baseline precision (Best for dual-GPU/workstations or custom quants) |
heretic-Ministral-3-14B-Q3_K_M.gguf |
3-bit Medium | Maximum space saving (Great for 8GB VRAM setups) |
heretic-Ministral-3-14B-Q4_K_M.gguf |
4-bit Medium | Standard baseline (Balanced speed and intelligence) |
heretic-Ministral-3-14B-Q5_K_M.gguf |
5-bit Medium | Community Favorite (Highly recommended for 12GB/16GB VRAM) |
heretic-Ministral-3-14B-Q6_K.gguf |
6-bit Standard | Premium tier (Near-lossless, slight VRAM savings over Q8) |
heretic-Ministral-3-14B-Q8_0.gguf |
8-bit Standard | Lossless precision (Requires large VRAM pools) |
How to use in LM Studio
- Copy this repository name:
Yingyaeliae/Ministral-3-14B-Nymphaea-RP-heretic - Paste it directly into the LM Studio Search/Discover bar and press Enter.
- Choose the specific quantization size that matches your GPU headroom and click Download.