Downloading New Models
Expand Noema with additional GGUF and MLX models curated for iOS and iPadOS. Follow these steps to browse, evaluate, and manage downloads without cluttering your device.
Compatibility
Choose the right build for your hardware
Models run best on A13 Bionic or newer devices with at least 5 GB of free storage. Older hardware can still participate—just prioritize smaller quantizations so inference stays responsive.
Library tour
Explore the built-in catalog
- Open the Stored tab to browse Noema’s curated recommendations for GGUF and MLX formats.
- Scan the capability badges to quickly match models with your device.
- Tap a model to review its description, compatibility notes, and storage footprint.
Quantization primer
Balance quality and footprint
Quantization compresses weights into smaller data types. Higher numbers like Q8_0 retain more detail but consume extra storage and RAM, while Q4_K_M or Q3_K_M dramatically shrink downloads for everyday use.
- Q8_0: best fidelity for top-tier iPads and Macs.
- Q5_K_M: balanced blend of quality and size.
- Q4_K_M: default pick for most portable devices.
- Q3_K_M: ideal when RAM is limited and speed matters.
Download flow
From selection to installation
- Choose a model from search or the featured list.
- Pick a quantization that fits your performance budget.
- Review the size estimate and ensure you have enough storage.
- Tap Download and keep the app open until the status flips to Ready.
- Tap Activate to make it available in chat.
Manage your library
- Inspect metadata to confirm architecture, context window, and last update.
- Archive or delete inactive models to reclaim storage instantly.
- Keep multiple quantizations of the same model to swap depending on task complexity.
Troubleshooting tips
- Downloads stalling? Confirm Wi‑Fi strength and verify you have at least 20% battery.
- Model fails to load? Close heavy apps to free memory, then relaunch Noema.
- Responses feel slow? Switch to a lighter quant or smaller model.