Hour 6 - Overview of Models

Lecture Notes: 

---

1. Concepts

What is a Model in Machine Learning?

• A model is a computational representation of a process used to make predictions or generate insights based on input data.
• In the context of language models, such as those used in Ollama, a model generates text or embeddings based on input prompts.

Key Components of a Model:

1. Architecture: Defines the structure of the model (e.g., transformers like GPT, BERT).
2. Parameters: Determines the model's capacity to learn from data (e.g., number of neurons, layers).
3. Weights: Encoded knowledge the model learns during training.
4. System Prompt: A predefined instruction guiding the model’s behavior.
5. Fine-Tuning: Adjusting the model for a specific task by retraining on a domain-specific dataset.

---

2. Key Aspects

Types of Models in Ollama:

1. Pre-trained Models: Models trained on large datasets for general-purpose tasks.
   • Example: GPT, Llama, Mistral.
2. Fine-Tuned Models: Pre-trained models further trained on specific datasets for specialized tasks.
   • Example: A chatbot fine-tuned for customer service.
3. Custom Models: Created by users with specific system prompts, templates, or datasets.

Why Use Models in Ollama?

• Enable text generation, summarization, translation, embedding generation, and more.
• Provide flexibility through CLI for creating, modifying, and deploying models.

---

3. Implementation

CLI Commands for Working with Models

Command	Description	Example
ollama run	Run a model to generate text based on a prompt.	ollama run llama --prompt "Hello!"
ollama create	Create a custom model from a base model.	ollama create mymodel -f ./modelfile
ollama pull	Download a model from Ollama’s repository.	ollama pull llama
ollama show	Display details of a model.	ollama show llama
ollama ls	List all available models locally.	ollama ls
ollama rm	Remove a model.	ollama rm llama
ollama cp	Copy a model (e.g., for renaming).	ollama cp llama llama_custom

---

4. Real-Life Example

Scenario: Customizing a Model for Technical FAQs

Imagine creating a custom model to answer FAQs for a software company.

• Base Model: Llama 2.
• Customization: Add a system prompt that aligns the model with the company’s tone and expertise.

---

5. Code Examples

Step 1: Pull the Base Model

Download a base model to use as the foundation for your customization.

# Pull a model from Ollama's repository
ollama pull llama

Step 2: Create a Custom Model

Define the behavior of your custom model in a modelfile.

Modelfile (modelfile.txt):

FROM llama
SYSTEM """
You are a technical support assistant for Acme Software.
Provide concise and accurate answers to customer queries.
"""

Use the ollama create command to generate the model.

# Create a custom model
ollama create acme_support -f ./modelfile.txt

Step 3: Run the Custom Model

Test your custom model by running it with a prompt.

# Run the custom model
ollama run acme_support --prompt "What are the system requirements for Acme Pro 3.0?"

Step 4: Show Model Details

Inspect the properties of the newly created model.

# Display model details
ollama show acme_support

---

6. Example Output

Input Prompt:

"What are the system requirements for Acme Pro 3.0?"

Model Response:

"Acme Pro 3.0 requires Windows 10 or later, 8GB RAM, and 20GB of free disk space. For macOS, it supports version 11.0 or newer."

---

7. Summary

• Concepts Covered: Overview of models, their types, and customization in Ollama.
• Key Aspects: Pre-trained, fine-tuned, and custom models, along with the CLI commands.
• Implementation: Demonstrated creating, running, and inspecting models.
• Real-Life Example: Built a custom technical support model.

---

8. Homework/Practice

1. Pull a base model and test its default behavior.
2. Create a custom model with a system prompt of your choice.
3. Run your custom model to answer specific questions.
4. Experiment with fine-tuning a model by modifying the modelfile or adding new training data.

This lecture provides a solid foundation in understanding and working with models in Ollama, emphasizing practical usage with CLI commands and real-life applications.