Try LFM • Documentation • LEAP

LFM2-24B-A2B-MLX-8bit

MLX export of LFM2-24B-A2B for Apple Silicon inference.

Model Details

Property	Value
Total Parameters	24B
Active Parameters	~2B per token
Architecture	Mixture of Experts (64 experts, top-4)
Layers	40 (30 conv + 10 full attention)
Precision	8-bit
Group Size	64
Size	23.6 GB
Context Length	128K

Recommended Sampling Parameters

Parameter	Value
temperature	0.1
top_k	50
top_p	0.1
repetition_penalty	1.05
max_tokens	512

Use with mlx

pip install mlx-lm

from mlx_lm import load, generate
from mlx_lm.sample_utils import make_sampler, make_logits_processors

model, tokenizer = load("LiquidAI/LFM2-24B-A2B-MLX-8bit")

prompt = "What is the capital of France?"

if tokenizer.chat_template is not None:
    messages = [{"role": "user", "content": prompt}]
    prompt = tokenizer.apply_chat_template(
        messages, tokenize=False, add_generation_prompt=True
    )

sampler = make_sampler(temp=0.1, top_k=50, top_p=0.1)
logits_processors = make_logits_processors(repetition_penalty=1.05)

response = generate(
    model,
    tokenizer,
    prompt=prompt,
    max_tokens=512,
    sampler=sampler,
    logits_processors=logits_processors,
    verbose=True,
)

License

This model is released under the LFM 1.0 License.