AITRADER/Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliterated-mlx-mxfp8

Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliterated MLX MXFP8

MXFP8 (Microscaling FP8) quantized MLX version of Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliterated.

Model Details

• Architecture: Qwen 3.5 27B (hybrid linear attention + full attention)
• Quantization: MXFP8 (E4M3 with block-level scaling), group_size=32
• Size: ~29 GB
• Context Length: 262,144 tokens
• Vision: Full image and video understanding via integrated vision tower (27 ViT blocks, kept in bf16)
• Tool Use: Native function calling support
• Thinking: Chain-of-thought reasoning mode

Why MXFP8?

MXFP8 uses floating-point (E4M3) representation with per-block scaling instead of fixed-point integer quantization. This gives:

• Better handling of outlier weights (exponent absorbs magnitude)
• Lower quantization error across varying tensor ranges
• Native hardware acceleration on modern chips

Capabilities

• Image understanding and description
• Video understanding
• Tool use / function calling
• Multi-step agent reasoning
• Thinking/reasoning mode
• Multilingual support
• Long context (262K tokens)

Usage

Works with LM Studio, MLX, and other MLX-compatible frameworks.