Before you run
The official app requires macOS 15.2+ or iOS 18.2+ and a compatible Xcode toolchain. Download the model before building. Use an empty model folder for the initial download; keep any existing custom model elsewhere before changing variants.
Check your first result
Build and run in Xcode. Select the target device, configure signing for iPhone, then ask a question about an image or camera frame. Check the generated answer and the app’s TTFT display.
Stage2 / Stage3 and Apple Silicon files →Source checked September 12, 2026. Runtime performance depends on your device.
Step by step
- 1
Confirm your Mac uses Apple Silicon.
- 2
Clone the official repository and inspect the app directory.
- 3
Download the documented MLX-compatible model assets.
- 4
Start with 0.5B before moving to a larger model.
Example
git clone https://github.com/apple-aiml-research/ml-fastvlm.git
cd ml-fastvlm
chmod +x app/get_pretrained_mlx_model.sh
app/get_pretrained_mlx_model.sh --model 0.5b --dest app/FastVLM/model
open app/FastVLM/FastVLM.xcodeproj Verify against the primary source
APIs, model files and dependency versions can change. Treat this page as an implementation map, then confirm the current command and license in the linked official documentation.
Open official source