L3-8B-Stheno-v3.2-MLX

This is the MLX version of the fantastic L3-8B-Stheno-v3.2 by Sao10K. Achieves roughly 10-15 tok/sec faster on my system then the original 4 bit quant.

Use with mlx

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("synk/L3-8B-Stheno-v3.2-MLX")

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)