Views
No views yet

| Property | Value |
|---|---|
| Parameters | 450M |
| Precision | 4-bit |
| Group Size | 64 |
| Size | 0.35 GB |
| Context Length | 32K |
| Vision Encoder | SigLIP2 NaFlex (86M) |
| Native Resolution | up to 512x512 |
uv pip install 'mlx-vlm==0.3.9'1from mlx_vlm import load, generate
2from mlx_vlm.utils import load_image
3
4model, processor = load("LiquidAI/LFM2.5-VL-450M-MLX-4bit")
5
6image = load_image("photo.jpg")
7
8# Apply chat template (required for LFM2.5-VL)
9messages = [{"role": "user", "content": [
10 {"type": "image"},
11 {"type": "text", "text": "What do you see in this image?"},
12]}]
13prompt = processor.apply_chat_template(messages, add_generation_prompt=True)
14
15result = generate(
16 model,
17 processor,
18 prompt,
19 [image],
20 temp=0.1,
21 min_p=0.15,
22 repetition_penalty=1.05,
23 verbose=True,
24)
25print(result.text)| Parameter | Value |
|---|---|
| temperature | 0.1 |
| min_p | 0.15 |
| repetition_penalty | 1.05 |