nightmedia/Qwen3-14B-Researcher-qx64-hi-mlx

Qwen3-14B-Researcher-mxfp4-mlx

Performance

Researcher-qx86-hi-mlx  0.519,0.636,0.888,0.749,0.412,0.800,0.721
ScienceBlossom-qx86-hi  0.513,0.650,0.883,0.753,0.414,0.798,0.708

The qx64-hi metrics are being processed and will be posted when available.

The Researcher model has been abliterated by DavidAU using Heretic, bringing refusals down from 99 to 22/100, and small performance increase.

-G

Use with mlx

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("Qwen3-14B-Researcher-qx64-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, return_dict=False,
    )

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