Ollama Parameter Explorer

A comprehensive web interface for experimenting with Ollama parameters and understanding how they affect model responses.

Features

• Interactive Parameter Controls: Sliders and toggles for all major Ollama parameters
• Real-time Parameter Adjustment: See parameter values change in real-time
• Streaming Responses: Choose between streaming and non-streaming responses
• Parameter Presets: Quick presets for Creative, Balanced, and Precise configurations
• Model Management: Automatic detection of available Ollama models
• Quick Test Prompts: Pre-made prompts to test different parameter effects
• Modern UI: Clean, responsive interface with visual feedback

Parameters Explained

Core Parameters

• Temperature (0.0-2.0): Controls randomness. Lower = more deterministic, higher = more creative
• Top P (0.0-1.0): Nucleus sampling - limits token choices by cumulative probability
• Top K (1-100): Limits to top K most likely tokens
• Repeat Penalty (0.5-2.0): Penalizes repetitive text (1.0 = no penalty)

Advanced Parameters

• Context Length: Number of tokens to keep in context window
• Max Response Length: Maximum tokens to generate in response
• Seed: For reproducible results (optional)

Mirostat (Advanced Sampling)

• Mirostat: Alternative to temperature/top_p sampling
• Mirostat Tau: Target entropy level
• Mirostat Eta: Learning rate for entropy adjustment

Setup Instructions

Prerequisites

1. Install Ollama:

   # On macOS
   brew install ollama
   
   # On Linux
   curl -fsSL https://ollama.ai/install.sh | sh
   
   # On Windows - download from https://ollama.ai

2. Start Ollama service:

   ollama serve

3. Pull at least one model:

   # Example models
   ollama pull llama2
   ollama pull codellama
   ollama pull mistral

Installation

1. Clone or create project directory:

   mkdir ollama-parameter-explorer
   cd ollama-parameter-explorer

2. Create the directory structure:

   mkdir templates

3. Save the files:

   • Save the Flask backend code as app.py
   • Save the HTML template as templates/index.html
   • Save requirements as requirements.txt

4. Install Python dependencies:

   pip install -r requirements.txt

5. Run the application:

   python app.py

6. Open your browser and go to: http://localhost:5000

Directory Structure

ollama-parameter-explorer/
├── app.py                 # Flask backend
├── requirements.txt       # Python dependencies
├── templates/
│   └── index.html        # Frontend template
└── README.md             # This file

Usage Guide

Getting Started

1. Ensure Ollama is running (ollama serve)
2. Start the Flask app (python app.py)
3. Open http://localhost:5000 in your browser
4. Select a model from the dropdown
5. Adjust parameters using the sliders and toggles
6. Enter a message and click send

Parameter Experimentation Tips

Understanding Temperature Through Examples

• Temperature 0.1: "The capital of France is Paris." (same answer every time)
• Temperature 1.5: "The capital of France is Paris, that magnificent city of lights and romance!" (varied, creative responses)
• Try asking "What is 2+2?" at different temperatures to see how it affects even simple questions

Understanding Top P and Top K

• High Top P (0.9) + Low Top K (10): Contradictory - gives inconsistent results
• Low Top P (0.5) + High Top K (50): Also contradictory - Top P will override Top K
• Best practice: Use either Top P OR Top K, not both at extreme values
• Try disabling one (set Top K to 100 or Top P to 1.0) to see the other's pure effect

Testing Repeat Penalty

• Ask for a story without repeat penalty (1.0) - notice natural repetition
• Set repeat penalty to 1.5+ - see how the AI awkwardly avoids repeating words
• Sweet spot is usually 1.1-1.3 for most use cases

Using Presets Effectively

• Creative: Perfect for "Write a story about..." prompts
• Balanced: Great for general questions and conversations
• Precise: Use for "Explain how..." or "What is..." questions

Troubleshooting

"Cannot connect to Ollama"

• Check if Ollama is running: ps aux | grep ollama
• Start Ollama: ollama serve
• Verify port 11434 is accessible

"No models available"

• Pull at least one model: ollama pull llama2
• Check available models: ollama list

Slow responses

• Reduce context length
• Reduce max response length
• Try a smaller model

Connection status always shows "error"

• Check if port 11434 is blocked by firewall
• Verify Ollama is running on localhost:11434
• Try restarting Ollama service

API Endpoints

• GET /api/models - Get available Ollama models
• POST /api/chat - Send chat request with parameters
• GET /api/health - Check Ollama connection status

Browser Compatibility

• Modern browsers with ES6+ support
• Chrome, Firefox, Safari, Edge
• Mobile responsive design

Contributing

Feel free to submit issues and enhancement requests!