Llamatik uses GGUF models (the standard file format used by llama.cpp).

Picking a model#

• Choose a model that has a GGUF release.
• Choose a quantization that matches your device constraints:
  • Smaller quantizations run faster and use less memory, with some quality tradeoff.
  • Larger quantizations can be higher quality but require more RAM.

Where to get GGUF models#

Common places include Hugging Face model pages that provide .gguf files.

Recommended first tests#

• Tiny/mini models (very small) to validate that your build + loading works.
• Then move to your target model size/quality once everything is stable.

Shipping models#

Models can be large. Consider:

• Android: ship a smaller default model, or download on first run.
• iOS: Apple size limits apply; use on-demand resources or download after install.
• Desktop: bundle or download, depending on your distribution model.