--- license: apache-2.0 base_model: mlx-community/Qwen3.6-27B-4bit tags: - mlx - lora - fine-tuned - vram-optimized - 36gb-mac --- # Qwen 3.6 27B Claude Opus Reasoning Distilled (MLX Adapters) This repository contains the LoRA adapter weights fine-tuned on [mlx-community/Qwen3.6-27B-4bit](https://huggingface.co/mlx-community/Qwen3.6-27B-4bit). These adapters are kept in high-precision `float16` format to prevent the quality degradation and quantization noise that occurs when fusing adapters directly into a 4-bit quantized base model. ## How to use in MLX ```python from mlx_lm import load, generate # Loading the base model and applying the high-precision adapters at runtime model, tokenizer = load( "mlx-community/Qwen3.6-27B-4bit", adapter_path="ethicalvats/Qwen3.6-27B-Claude-Opus-Reasoning-Distilled-MLX" ) response = generate( model, tokenizer, prompt=" user\nWhat is the oldest tree? \n assistant\n \n", verbose=True )
