ollama-nuggets

Coding nuggets about using Ollama and its LLM models

Context Length Experiments

This repository contains experiments exploring Ollama's behavior with different context and prediction length settings, using the llama3.2 model.

Key Findings

• Demonstrated how to use different context lengths (num_ctx) and prediction lengths (num_predict) with Ollama
• Compared output quality between default and expanded context settings
• Explored the relationship between prediction length and output completeness
• Analyzed token counts and model behavior under different configurations

Setup

pip install -r requirements.txt