unsloth-Qwen3-VL-32B-Instruct-qx86x-hi-mlx

This is the Deckard(qx) formula that uses data stores and most attention paths in low precision(6 bit), enhancing vital attention paths, head, context, and embeddings to 8 bit, and quantized with high precision(group size 32).

I am still evaluating this quant, here is a LinkedIn review of one of my pictures with the unsloth-Qwen3-VL-8B-Instruct-qx86x-hi-mlx

This model unsloth-Qwen3-VL-32B-Instruct-qx86x-hi-mlx was converted to MLX format from unsloth/Qwen3-VL-32B-Instruct using mlx-lm version 0.28.4.

Use with mlx

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("unsloth-Qwen3-VL-32B-Instruct-qx86x-hi-mlx")

prompt = "hello"

if tokenizer.chat_template is not None:
    messages = [{"role": "user", "content": prompt}]
    prompt = tokenizer.apply_chat_template(
        messages, add_generation_prompt=True
    )

response = generate(model, tokenizer, prompt=prompt, verbose=True)