Ollama – Complete Learning

A hands-on, structured guide to Ollama covering local LLM usage, LangChain integration, REST APIs, tool calling, Ollama Cloud, and advanced experiments with models like Gemma 3 (4B), including vision and multimodal workflows.

---

📌 What is Ollama?

Ollama is a platform that allows you to run, manage, and experiment with large language models (LLMs) locally and in the cloud. It provides a simple CLI, REST API, and library support to work with modern open-source models like LLaMA, Gemma, Qwen, Mistral, and more.

Key idea: Local-first AI with optional cloud scalability.

---

🧠 Why Use Ollama?

• 🚀 Run LLMs locally (privacy + speed)
• 🔒 No data leaves your machine (local mode)
• 🧩 Easy integration with Python, LangChain, REST APIs
• 🛠️ Built-in support for tool calling & function execution
• ☁️ Ollama Cloud for scalable inference
• 🧪 Perfect for learning, experimentation, and R&D

---

📂 Project Structure

.
├── Ollama.ipynb                    # Core Ollama usage & CLI experiments
├── Ollama using Rest API.ipynb     # REST API calls (generate, chat, models)
├── Ollama Using LangChain.ipynb    # LangChain + Ollama integration
├── Tool Calling.ipynb              # Function / tool calling with Ollama
├── Ollama Cloud.ipynb              # Cloud-based inference concepts
├── Modelfile.txt                   # Custom model configuration
├── Ollama Short notes.docx         # Quick theory notes
├── README.md                       # This documentation

---

⚙️ Ollama Core Concepts

🧾 Common Ollama Commands

🔹 Ollama CLI Commands

# Check Ollama version
ollama --version

# List available models
ollama list

# Pull a model
ollama pull gemma3:4b

# Run a model interactively
ollama run gemma3:4b

# Run with a prompt
ollama run gemma3:4b "Explain LLMs in simple words"

# Remove a model
ollama rm gemma3:4b

---

🔹 Ollama REST API (curl examples)

# Generate text
curl http://localhost:11434/api/generate \
  -d '{"model": "gemma3:4b", "prompt": "What is GenAI?"}'

# Chat API
curl http://localhost:11434/api/chat \
  -d '{"model": "gemma3:4b", "messages": [{"role": "user", "content": "Hello"}]}'

# List models via API
curl http://localhost:11434/api/tags

---

🔹 Python (Ollama Library)

from ollama import chat

response = chat(
    model="gemma3:4b",
    messages=[{"role": "user", "content": "Explain tool calling"}]
)
print(response["message"]["content"])

---

🔹 LangChain + Ollama

from langchain_community.llms import Ollama

llm = Ollama(model="gemma3:4b")
print(llm.invoke("What is RAG?"))

---

🔹 Tool Calling (Conceptual Command)

• Define tools (functions)
• Pass tool schema to model
• Model decides when to call tools

Used for:

• Calculations
• API calls
• Database queries

---

🔹 Ollama Cloud (Conceptual Commands)

Prompt → Cloud Endpoint → GPU Inference → Response

Used when:

• Large models (30B+)
• High traffic apps
• Production workloads

---

1️⃣ Ollama CLI & Library

• Pull models (ollama pull gemma3:4b)
• Run models locally (ollama run gemma3:4b)
• Manage models (list, delete, update)
• Python library for programmatic access

Benefit: Simple developer experience with production-ready models.

---

2️⃣ Ollama REST API

Ollama exposes a local REST server:

• /api/generate
• /api/chat
• /api/tags (list models)

Use cases:

• Backend integration
• Web apps
• Microservices

Benefits:

• Language agnostic
• Easy to scale
• Works with Docker & cloud infra

---

3️⃣ Ollama with LangChain

LangChain enables:

• Prompt templates
• Chains & agents
• Memory
• Tool usage

Workflow:

User → LangChain → Ollama Model → Response

Benefits:

• Build RAG pipelines
• AI agents
• Conversational memory

---

🛠️ Tool Calling with Ollama

Tool calling allows models to:

• Call Python functions
• Execute APIs
• Perform structured reasoning

Examples:

• Calculator tools
• Database queries
• File operations
• External API calls

Why it matters:

Turns LLMs into AI agents, not just chatbots.

---

☁️ Ollama Cloud – How Large Models Work in Cloud

How Cloud LLMs Work

1. User sends prompt
2. Request hits cloud inference server
3. Model runs on GPU/TPU
4. Response streamed back

Benefits of Cloud Models

• ⚡ High performance GPUs
• 📈 Auto scaling
• 🧠 Large models (70B+)
• 🔄 No local hardware dependency

Local vs Cloud

Feature	Local Ollama	Ollama Cloud
Privacy	✅ High	⚠️ Medium
Cost	✅ Low	💰 Usage based
Speed	⚡ Fast (small models)	🚀 Fast (large models)
Offline	✅ Yes	❌ No

---

🧪 Model Experiments – Gemma 3 (4B)

Why Gemma 3 (4B)?

• Lightweight
• Fast inference
• High-quality reasoning
• Ideal for laptops

Experiments Covered

• Text generation
• Instruction following
• Tool calling compatibility
• Vision & multimodal prompts (image → text)

---

🖼️ Vision & Image → Text (Multimodal)

Supported workflows:

• Image captioning
• OCR-like text extraction
• Visual reasoning

Use cases:

• Document processing
• Image understanding
• AI assistants with vision

---

🤖 Model Capability Matrix

Feature	Gemma 3	LLaMA	Qwen	Mistral
Tool Calling	✅	✅	✅	⚠️
Vision	✅	❌	⚠️	❌
Cloud Support	✅	✅	✅	✅
RAG Friendly	✅	✅	✅	✅

---

📚 Learning Path (Recommended)

Beginner

1. Ollama CLI basics
2. Pull & run models
3. Simple text generation

Intermediate

4. REST API usage
5. LangChain integration
6. Prompt engineering

Advanced

7. Tool calling
8. Vision models
9. RAG pipelines
10. Cloud deployment

---

🎯 Benefits of This Setup

• Learn LLMs practically
• Build production-ready AI apps
• No vendor lock-in
• Works locally & in cloud
• Ideal for students & professionals

---

🚀 Future Extensions

• RAG with vector databases
• Multi-agent systems
• Fine-tuning custom models
• Production deployment (Docker, Kubernetes)

---

Happy Building 🚀