bobig/orpheus-finetuned-3b-mlx-6Bit

The Model bobig/orpheus-finetuned-3b-mlx-6Bit was converted to MLX format from srinivasbilla/orpheus-finetuned-3b using mlx-lm version 0.26.4.

Yeah, it works, but not very good. Choppy speech.

Here is an alternative that works well:

https://huggingface.co/mradermacher/orpheus-finetuned-3b-GGUF

the Q4_M can run 100% CPU on Mac M4 Max using 10 Performance cores.

Save your GPU for one of NightMedia's MLX quants.

Use with mlx

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("bobig/orpheus-finetuned-3b-mlx-6Bit")

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
    )

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