LFM2.5-VL-450M is a vision-language model built on the LFM2.5-350M backbone with a SigLIP2 NaFlex vision encoder (86M). It supports OCR, document comprehension, multilingual vision understanding, bounding box prediction, and function calling.

Model Details

Property	Value
Parameters	450M
Precision	bfloat16
Size	0.84 GB
Context Length	32K
Vision Encoder	SigLIP2 NaFlex (86M)
Native Resolution	up to 512x512

Quickstart

uv pip install 'mlx-vlm==0.3.9'

python

from mlx_vlm import load, generate
from mlx_vlm.utils import load_image

model, processor = load("LiquidAI/LFM2.5-VL-450M-MLX-bf16")

image = load_image("photo.jpg")

# Apply chat template (required for LFM2.5-VL)
messages = [{"role": "user", "content": [
    {"type": "image"},
    {"type": "text", "text": "What do you see in this image?"},
]}]
prompt = processor.apply_chat_template(messages, add_generation_prompt=True)

result = generate(
    model,
    processor,
    prompt,
    [image],
    temp=0.1,
    min_p=0.15,
    repetition_penalty=1.05,
    verbose=True,
)
print(result.text)

Recommended Sampling Parameters

Parameter	Value
temperature	0.1
min_p	0.15
repetition_penalty	1.05

License

This model is released under the LFM 1.0 License.