overthinkos/notebook-ollama

notebook-ollama -- Ollama integration notebook data layer

Layer Properties

| Property | Value | |----------|-------| | Dependencies | (none) | | Packages | (none) | | Services | (none) | | Volumes | workspace -> /workspace (from jupyter) | | Data | data/ollama -> workspace volume, dest: ollama | | Install files | (none) |

How It Works

This is a data layer — it uses the data: field in layer.yml to map a directory of notebooks to a named volume with a subdirectory destination:

info: "Ollama integration notebook collection"

data:
  - src: data/ollama
    volume: workspace
    dest: ollama

At build time, the contents of data/ollama/ are staged into /data/workspace/ollama/ inside the image.

At deploy time, ov config or ov update copies the staged data into the workspace volume at <workspace>/ollama/. The dest: ollama field places the notebooks in a subdirectory rather than the volume root.

Included Notebooks

6 Jupyter notebooks covering different Ollama client libraries:

| Notebook | Library | Features | |----------|---------|----------| | 00_Ollama_Requests.ipynb | requests | Raw REST API: list, show, generate, chat, stream, embed, copy, delete | | 01_Ollama_GPU.ipynb | requests | GPU verification: nvidia-smi, inference metrics, VRAM monitoring | | 02_Ollama_OpenAI.ipynb | openai | OpenAI-compatible API: completions, chat, multi-turn, stream, embed | | 03_Ollama_Library.ipynb | ollama | Native Python library: all API operations + model management | | 04_Ollama_HuggingFace.ipynb | ollama | HuggingFace GGUF model import, verification, inference | | 05_Ollama_Anthropic.ipynb | anthropic | Anthropic-compatible API: chat, system prompts, streaming, tool calling, vision |

Manifest: notebooks.yaml — structured catalog with title, description, and ordering.

Network Connectivity

The notebooks default to http://localhost:11434 for the Ollama API endpoint. Each notebook reads:

OLLAMA_HOST = os.getenv("OLLAMA_HOST", "http://localhost:11434")

When the ollama image is deployed via ov config ollama, its env_provides automatically injects OLLAMA_HOST=http://ov-ollama:11434 into the global deploy.yml env. Any container configured after ollama (or reconfigured with --update-all) automatically gets the correct Ollama endpoint — no manual environment setup needed.

Setup:

ov config ollama --update-all    # deploys ollama + propagates OLLAMA_HOST to all
ov start ollama
ov start jupyter-ml-notebook   # OLLAMA_HOST already set via env_provides

Both containers must be on the same ov Podman network for DNS resolution to work.

See /ov-core:ov-config for --update-all and /ov-ollama:ollama for env_provides details.

Notebook Compatibility Notes

ollama Python library env var gotcha

The ollama Python library creates its HTTP client at import time by reading OLLAMA_HOST from os.environ. Module-level functions (ollama.list(), ollama.generate(), etc.) are bound methods on this cached client. Simply setting a Python variable doesn't propagate to the library.

The notebooks use this pattern to handle both fresh kernels and re-runs:

import os
OLLAMA_HOST = os.getenv("OLLAMA_HOST", "http://localhost:11434")
os.environ["OLLAMA_HOST"] = OLLAMA_HOST

import ollama
import importlib; importlib.reload(ollama)  # Rebinds module functions

Pydantic model changes (ollama >= 0.4)

The newer ollama library returns Pydantic models, not dicts. Model names are accessed via .model attribute (not ["name"]), and the list is .models (not .get("models", [])):

# Old (dict access)
model_names = [m.get("model", "") for m in models.get("models", [])]

# New (Pydantic attribute access)
model_names = [m.model for m in models.models]

Default models

• 5 notebooks use llama3.2:latest (3.2B, Q4_K_M)
• ollama_anthropic.ipynb uses ministral-3:3b-instruct-2512-q4_K_M

Pull models before running: ov shell ollama -c "ollama pull llama3.2"

Usage

# image.yml
jupyter-ml-notebook:
  layers:
    - jupyter-ml
    - notebook-templates
    - notebook-finetuning
    - notebook-ollama
    # ... other layers

# Deploy and start both services
ov config ollama && ov start ollama
ov config jupyter-ml-notebook && ov start jupyter-ml-notebook

# Pull a model, then open notebooks
ov shell ollama -c "ollama pull llama3.2"
# Open http://localhost:8888 -> navigate to ollama/

Used In Images

• /ov-jupyter:jupyter-ml-notebook

Related Skills

• /ov-image:layer — data field documentation and layer authoring rules
• /ov-core:ov-config — data provisioning during ov config setup
• /ov-core:deploy — volume backing configuration
• /ov-jupyter:notebook-finetuning — sibling data layer pattern (Unsloth fine-tuning notebooks)
• /ov-jupyter:notebook-templates — sibling data layer pattern (starter notebooks)
• /ov-ollama:ollama — the Ollama binary layer (server side)
• /ov-ollama:ollama — the standalone Ollama image (must be running for notebooks to connect)
• /ov-jupyter:jupyter-ml-notebook — the image that includes this layer

When to Use This Skill

Use when the user asks about:

• The notebook-ollama layer or its contents
• Ollama API tutorial notebooks or integration examples
• How Jupyter notebooks connect to a separate Ollama container
• The ollama Python library env var or Pydantic API gotchas
• Which client libraries and API features are covered

Related

• /ov-eval:eval — declarative testing (eval: block, ov eval image, ov eval live)