Models · family
Granite
2 MLX aliases · IBM Granite 4 · hybrid state-space + attention.
IBM's Granite 4 line — hybrid state-space + attention small models. Both aliases run hybrid cache so long-context cost stays bounded. Useful as draft models or as low-RAM agents on base-spec Macs.
- family
- Granite
- aliases
- 2
- lines
- 1
- install
- rapid-mlx serve <alias>
- OpenAI base URL
- http://localhost:8000/v1
Granite 4 · 2 aliases
H-Micro + Tiny variants. Both run hybrid cache — long-context cost bounded by SSM step, not by KV cache size.
parser: hermes
| alias | hf repo | tool parser | reasoning | flags |
|---|---|---|---|---|
| granite4-h-micro-4bit | mlx-community/granite-4.0-h-micro-4bit | hermes | — | hybrid |
| granite4-tiny-4bit | mlx-community/granite-4.0-h-tiny-4bit | hermes | — | hybrid |
Notes & caveats
granite4-h-micro-4bitis sub-2 GB on disk — runs anywhere.- Hybrid cache enabled — long-context cost bounded by SSM step, not by KV cache size.