Note: command for conversion:

mlx_lm.convert --hf-repo openai/gpt-oss-20b ... -q --q-bits 4

The version without doing an added 4-bit quant was in a 16-bit format (either bfloat16 or float16, I assume; I didn't check).

Use with mlx

pip install mlx-lm

python

from mlx_lm import load, generate

model, tokenizer = load("sealad886/gpt-oss-20b-MLX-native")

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)