---
base_model: huihui-ai/Huihui-Nex-N2-mini-abliterated
license: apache-2.0
pipeline_tag: image-text-to-text
tags: [gguf, llama.cpp, multimodal, vision, abliterated, qwen3_5_moe, imatrix]
---
# Huihui-Nex-N2-mini-abliterated-GGUF

GGUF de [`huihui-ai/Huihui-Nex-N2-mini-abliterated`](https://huggingface.co/huihui-ai/Huihui-Nex-N2-mini-abliterated) (arch `qwen3_5_moe`, MoE ~35B, **multimodal**), via `llama.cpp`.

> Conversao corrigida: o `block_count` (camada MTP/nextn fantasma) foi ajustado de 41 para 40 para o modelo carregar.

## Arquivos
| Arquivo | O que e |
|---|---|
| `Huihui-Nex-N2-mini-abliterated-UD-Q4_K_XL.gguf` | LLM texto **Q4_K_M + imatrix** (din., estilo Unsloth UD) |
| `Huihui-Nex-N2-mini-abliterated-Q4_K_M.gguf` | LLM texto Q4_K_M padrao |
| `mmproj-Huihui-Nex-N2-mini-abliterated-f16.gguf` | **Projetor de visao** (necessario p/ imagem) |
| `imatrix-Huihui-Nex-N2-mini-abliterated.dat` | importance matrix (calibration_datav3) |

## Uso (visao)
```bash
llama-server -m Huihui-Nex-N2-mini-abliterated-UD-Q4_K_XL.gguf \
  --mmproj mmproj-Huihui-Nex-N2-mini-abliterated-f16.gguf -c 8192
```