# Rocinante

Heavy AI inference server.

> Planned direction: Rocinante becomes optional if the upgraded PD build receives the 4090. If PD does not get the 4090, Rocinante remains the heavy-inference specialist.

## Hardware

| Component | Details |
|-----------|---------|
| **IP** | 10.5.30.112 |
| **GPU** | NVIDIA RTX 4090 (24 GB VRAM) |
| **Role** | Heavy AI inference (Ollama) |

## Services

### Ollama

Ollama runs on port `11434` and hosts the large-model ("heavy") tier.

| Model | Use |
|-------|-----|
| `qwen3:latest` | Fast reasoning / general chat |
| `qwen2.5:32b` | Large-context tasks |
| `gemma3:27b` | Multimodal / vision |

> **LiteLLM tier:** Rocinante is the **"heavy"** tier in LiteLLM `config.yaml`.
> Heavy-tier requests from OpenWebUI and n8n route to `http://10.5.30.112:11434`.

## Current operational note

- VLAN access to Rocinante from N.O.M.A.D. was re-validated on 2026-05-27.
- Rocinante Ollama now answers from N.O.M.A.D. at `http://10.5.30.112:11434`; `/api/tags`, `/api/generate`, and OpenAI-compatible `/v1/chat/completions` all succeeded.
- Hermes configs that still pointed at the legacy `10.5.1.112` endpoint were corrected to `10.5.30.112`.
- Re-check CUDA Whisper on `http://10.5.30.112:8787/health` separately if Whisper-specific reachability is needed; this pass validated Ollama end-to-end.

## Notes

- No SSH config needed for routine use — Ollama API is accessed over LAN only
- Do not run multiple large VRAM allocations simultaneously on the 4090