Qwen3-4B-Engineer12-qx86-hi-mlx

Brainwaves

         arc   arc/e boolq hswag obkqa piqa  wino
qx86-hi  0.573,0.775,0.860,0.696,0.416,0.772,0.642

This is a model merge based on the new Jan-v3-4B-base-instruct with the following components:

• janhq/Jan-v3-4B-base-instruct
• TeichAI/Qwen3-4B-Instruct-2507-Polaris-Alpha-Distill
• TeichAI/Qwen3-4B-Thinking-2507-Gemini-2.5-Flash-Distill
• Alibaba-Apsara/DASD-4B-Thinking (GPT-OSS traces)

Metrics for the individual step models

         arc   arc/e boolq hswag obkqa piqa  wino
Jan-v3-4B-base-instruct
qx86-hi  0.452,0.600,0.846,0.457,0.392,0.699,0.563

Qwen3-4B-Instruct-2507-Polaris-Alpha-Distill
qx86-hi  0.482,0.651,0.848,0.512,0.386,0.686,0.567

Qwen3-4B-Thinking-2507-Gemini-2.5-Flash-Distill
qx86-hi  0.386,0.447,0.685,0.582,0.362,0.723,0.593

DASD-4B-Thinking
qx86-hi  0.395,0.452,0.380,0.565,0.356,0.694,0.590

This model Qwen3-4B-Engineer12-qx86-hi-mlx was converted to MLX format from nightmedia/Qwen3-4B-Engineer12 using mlx-lm version 0.30.4.

Use with mlx

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("Qwen3-4B-Engineer12-qx86-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)