Ollama-bench

A command-line tool to benchmark and compare the performance of Ollama language models. Measures tokens per second and total processing time.

Setup

1. Install Ollama

Choose your platform:

• Windows: Download Installer
• macOS: Download App
• Linux: Run:

  curl -fsSL https://ollama.com/install.sh | sh

• Docker: Pull and run:

  docker pull ollama/ollama
  docker run -d -v ollama:/root/.ollama -p 11434:11434 ollama/ollama

2. Start Ollama Server

Before running any benchmarks, make sure the Ollama server is running:

# On Linux/macOS terminal or Windows PowerShell
ollama serve

For Windows users, you can also run Ollama from the system tray after installation.

3. Install Benchmark Tool

Install globally:

npm install -g ollama-bench

Or run directly with npx:

npx ollama-bench <model1> [model2] [model3]

Usage

# Using global installation
ollama-bench smollm:135m qwen2.5:0.5b

# Using npx (no installation required)
npx ollama-bench smollm:135m qwen2.5:0.5b

Troubleshooting

If you encounter errors, check:

1. Is the Ollama server running? (ollama serve)
2. Can you access http://localhost:11434?
3. Do you have enough RAM for your chosen models?

Available Models

See ollama.com/library for all available models.

License

MIT