LFM2.5-VL-450M-MLX-bf16
MLX export of LFM2.5-VL-450M for Apple Silicon inference.
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'
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.