UI-Venus-1.5-30B-A3B-mxfp4-mlx

Brainwaves

          arc   arc/e boolq hswag obkqa piqa  wino
mxfp8     0.544,0.707,0.900,0.755,0.460,0.804,0.721
qx86-hi   0.557,0.715,0.899,0.764,0.452,0.806,0.699
qx64-hi   0.526,0.696,0.898,0.754,0.450,0.802,0.702
mxfp4     0.546,0.732,0.894,0.731,0.444,0.794,0.687


Perplexity:
mxfp8     3.815 ± 0.025
qx86-hi   3.748 ± 0.024
qx64-hi   3.805 ± 0.025
mxfp4     4.155 ± 0.028

Qwen3-VL-30B-A3B-Instruct
qx86-hi   0.439,0.541,0.894,0.619,0.430,0.764,0.592

Qwen3-VL-30B-A3B-Thinking
qx86-hi   0.393,0.466,0.751,0.648,0.366,0.776,0.667

Use with mlx

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("UI-Venus-1.5-30B-A3B-mxfp4-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)