Darwin-28B — local Q4_K_M¶
Status: ✅ local · HF: private · Local port:
:11401
Identity¶
| Full name | Darwin-28B (Q4_K_M) |
| Type | Local dense model (Darwin-merged) |
| Total params | 28B |
| Active per token | 28B (dense) |
| Quantization | Q4_K_M |
| Disk | ~16-18GB |
| VRAM | ~18-20GB (1× RTX 3090) |
What it is¶
One of the local Darwin-merged dense models that runs on the dual RTX 3090 rig. The specific 28B configuration is part of the Southpaw-models lineup (see local server infra for the full inventory).
Local use¶
- Server: llama-server on
:11401 - Profile:
sovthpawHermes profile uses it as one of the model options - Use case: when you need a 28B-class model for local reasoning, Darwin-28B is the default
See also¶
- Related: APEX-MTP I-Compact (the speculative-decode partner on GPU 1)
- Local server:
~/projects/LLM-Infra/southpaw-turbohaul/ - Blog:
../../blog/the-omni-family.md(for the broader family context)