Qwen3-0.6B-dwq6b-mlx

Tuned for nice output, 165 tok/sec

Performance evaluation:

arc_challenge:
acc 0.263, acc_norm 0.294, stderr: 0.013

arc_easy:
acc 0.398, acc_norm 0.362, stderr 0.0098

boolq:
acc 0.378, stderr 0.0084

hellaswag:
acc 0.363, acc_norm 0.423, stderr: 0.0049

openbookqa:
acc 0.168, acc_norm 0.340, stderr: 0.0212

piqa:
acc 0.655, acc_norm 0.643, stderr: 0.011

winogrande:
acc 0.531, stderr: 0.014

This model Qwen3-0.6B-dwq6b-mlx was converted to MLX format from Qwen/Qwen3-0.6B using mlx-lm version 0.26.0.

Use with mlx

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("Qwen3-0.6B-dwq6b-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)