Modelpublic

dawncr0w/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-oQ8

sourceHugging Faceapache-2.0updated 4mo agoView on Hugging Face

4likes378downloads

Model Card

Qwen3.6-35B-A3B Uncensored HauhauCS Aggressive oQ8

This is an oMLX oQ8 VLM quantization of ericpandev/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-bf16.

Quantization

—Quantizer: oMLX oQ
—Level: oQ8
—Group size: 64
—Non-quantized/scales dtype: bfloat16
—Output size: about 35 GB
—Vision weights: preserved
—Native MTP: disabled

Notes

The source checkpoint is a VLM-style BF16 conversion. This repository preserves vision_config and model.visual.* tensors. Native MTP was disabled because the BF16 source did not include mtp.* tensors.

Local validation confirmed the vision tensors are present, but the bundled MLX-VLM runtime in oMLX did not support the qwen3_5_moe_vision model type at the time of conversion.