Modelpublic

vanch007/Huihui-Qwen3.6-35B-A3B-abliterated-mlx-8bit

sourceHugging Faceupdated 5mo agoView on Hugging Face

0likes72downloads

Model Card

Huihui-Qwen3.6-35B-A3B-abliterated-mlx-8bit

MLX-VLM conversion of huihui-ai/Huihui-Qwen3.6-35B-A3B-abliterated.

Overview

—Format: MLX-VLM
—Precision: 8bit
—Size: 35G
—Source model type: Qwen3_5MoeForConditionalGeneration
—Source pipeline: image-text-to-text
—Intended runtime: mlx-vlm, LM Studio
—Quantization result: 8-bit affine

Validation

Local checks on Apple Silicon:

—model loading in mlx-vlm: passed
—text generation smoke test: passed
—image generation path smoke test: passed

Smoke-test logs were produced locally before upload.

Notes

This is an abliterated model conversion. Outputs may be less filtered than standard instruction-tuned models. Use responsibly and comply with applicable laws and platform policies.

Files

Important files in this repo:

—config.json
—generation_config.json
—chat_template.jinja
—tokenizer.json
—tokenizer_config.json
—model.safetensors.index.json
—model-*.safetensors

Usage

Text generation

mlxvlm.generate \ --model /path/to/Huihui-Qwen3.6-35B-A3B-abliterated-mlx-8bit \ --prompt "你好" \ --max-tokens 128 \ --trust-remote-code \ --processor-kwargs '{"enablethinking": false}'

Image prompt

mlxvlm.generate \ --model /path/to/Huihui-Qwen3.6-35B-A3B-abliterated-mlx-8bit \ --image /path/to/example.png \ --prompt "Describe this image." \ --max-tokens 128 \ --trust-remote-code \ --processor-kwargs '{"enablethinking": false}'