from mlx_lm import load, generate

model, tokenizer = load("SiddhJagani/gpt-oss-safeguard-20b-mlx-Q3")

prompt = "hello"

if hasattr(tokenizer, "apply_chat_template") and tokenizer.chat_template is not None:
    messages = [{"role": "user", "content": prompt}]
    prompt = tokenizer.apply_chat_template(
        messages, tokenize=False, add_generation_prompt=True
    )
    # Force our custom assistant marker
    prompt = force_generation_prompt(prompt)

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