GenAI Portfolio Suite – Phase 0: Ollama Runtime

Shared Ollama LLM runtime for the GenAI Portfolio Suite.

This repository provides a GPU-accelerated Ollama container and shared Docker network that all portfolio phases connect to.

Part of the GenAI Portfolio Suite.

Phase: 0 – LLM Runtime & Infrastructure

---

Table of Contents

• Overview
• Quick Start
• Architecture
• Features
• Available Models
• Requirements
• Usage
• Project Structure
• Tech Stack
• Author
• License

---

Overview

ollama-runtime is the shared LLM runtime for the suite.
It runs a single Ollama container that:

• Exposes the standard Ollama API on port 11434
• Hosts all configured models (pulled once, reused by all phases)
• Provides a shared Docker network so other services can reach Ollama at http://ollama:11434

---

Quick Start

./scripts/start.sh
curl http://localhost:11434/api/tags

To stop:

./scripts/stop.sh

---

Architecture

graph LR
    Ollama["Ollama Container<br/>:11434"] --> Network["ollama-runtime-network<br/>(bridge)"]
    Network --> Phase1["Phase 1: LLM Server"]
    Network --> Phase2["Phase 2: RAG Console"]
    Network --> Phase3["Phase 3+: Future Phases"]

Loading

Ollama runs as an independent service that other phases connect to via the shared Docker network ollama-runtime-network.

---

Features

Model Serving 1

Model Serving 2

Model Serving 3

• Shared LLM Runtime – single Ollama instance used by all portfolio phases
• GPU Acceleration – optional NVIDIA GPU support for faster inference
• Model Management – centralized model storage, pulled once and reused
• Network Isolation – dedicated Docker network for secure inter-container communication
• Lifecycle Independence – can be started/stopped independently of other phases

---

Available Models

Default model across Phase 0-1-2: llama3.2:3b

All models are 3B-class Q4_K_M quantized for consistent performance.

Family	Model	Size	Notes
Meta	llama3.2:3b	2.0 GB	Default -- general-purpose
Alibaba	qwen2.5:3b	1.9 GB	Strong multilingual support
Microsoft	phi3.5:3.8b	2.2 GB	Reasoning, code, structured

These are pulled via ./scripts/pull_models.sh and then reused by all phases.

---

Requirements

• Docker with Compose v2
• NVIDIA GPU + nvidia-container-toolkit (for GPU acceleration, optional but recommended)

---

Usage

Command	Description
./scripts/start.sh	Start Ollama container
./scripts/stop.sh	Stop Ollama container
./scripts/restart.sh	Restart Ollama container
./scripts/pull_models.sh	Download all configured models

Ports

Service	Port	URL
Ollama API	11434	http://localhost:11434

Port scheme across the suite:

• Phase 0: 11434 (standard Ollama port)
• Phase 1: 1xxx range (e.g. 1080, 1501)
• Phase 2: 2xxx range (e.g. 2080, 2501)

Network Integration

This repository creates the shared Docker network:

• Network name: ollama-runtime-network
• Driver: bridge
• Ollama is reachable for other containers at: http://ollama:11434

This repository provides the shared Ollama service for:

• Phase 1: ollama-multi-llm-server – Multi-model inference API + playground
• Phase 2: rag-operator-console – RAG pipeline + operator debugging UI
• Phase 3+: Future phases in the suite

Each phase's docker-compose.yaml declares ollama-runtime-network as an external network:

networks:
  ollama-network:
    external: true
    name: ollama-runtime-network

This keeps Ollama's lifecycle independent while allowing all phases to share the same LLM runtime.

---

Project Structure

ollama-runtime/
├── scripts/
│   ├── start.sh           Start Ollama container
│   ├── stop.sh            Stop Ollama container
│   ├── restart.sh          Restart Ollama container
│   └── pull_models.sh     Download all configured models
├── docker-compose.yaml     Ollama service + network definition
├── LICENSE
└── README.md

---

Tech Stack

• Runtime: Ollama
• Containerization: Docker + Docker Compose
• Networking: ollama-runtime-network (bridge)
• Hardware Acceleration: NVIDIA GPU (optional)

---

Author

Adityo Nugroho – github.com/adityonugrohoid

---

License

MIT License – see LICENSE.