Gemma-4-12B-OBLITERATED-MLX-BF16

This is an unquantized BF16 MLX-LM conversion of OBLITERATUS/Gemma-4-12B-OBLITERATED.

Source revision: b68161d97ca96f6c354ddd614b1f4c08ad07a973

The source config declares gemma4_unified; the local conversion copy was patched to gemma4 so MLX-LM can load the text-generation model. No -q/quantization option was used, and the conversion was saved with --dtype bfloat16.

Use with MLX-LM

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("bsisduck/Gemma-4-12B-OBLITERATED-MLX-BF16")

messages = [{"role": "user", "content": "What is the capital of France?"}]
prompt = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_dict=False,
)

response = generate(model, tokenizer, prompt=prompt, max_tokens=128, verbose=True)