Skip to content
/ universal-ollama-optimizer Public

Professional bash script for launching and managing Ollama AI models with parameter profile suggestions, system monitoring, and automated configuration. Not an automatic optimizer - provides suggested /set commands for manual application.

License

MIT license
0 stars 0 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

jtgsystems/universal-ollama-optimizer

BranchesTags

Repository files navigation

Banner

Ollama Launcher & Manager

Developed by JTGSYSTEMS.COM | JointTechnologyGroup.com

A professional bash script for launching and managing Ollama AI models with intelligent profile suggestions, system monitoring, and automated configuration.

License: MIT Platform Ollama Bash

What This Actually Does

This is a model launcher and management toolkit for Ollama. It provides:

  • Interactive model selection from 55+ pre-configured 2025 models
  • System validation before launching (network, Ollama service, disk space, GPU/RAM checks)
  • Parameter profile suggestions with recommended settings for different use cases
  • Automatic model downloading with validation and retry logic
  • Clean terminal UI with colors and progress indicators

Important Note on "Optimization"

This script does NOT automatically apply parameters to models. Ollama's CLI doesn't support passing parameters like temperature or top_p at launch time. Instead, this tool:

  1. Suggests optimized profiles based on your use case
  2. Shows you the recommended parameters for that profile
  3. Provides copy-paste ready /set commands to apply manually during your session

For automatic parameter application, you would need to use Ollama's HTTP API directly or create custom Modelfiles (which this script can help with).

Latest Updates (September 2025)

Added 5 Critical Missing Models Based on Community Research:

  • GLM-4 - Ranks 3rd overall on hardcore benchmarks, beats Llama 3 8B
  • Magicoder - OSS-Instruct trained coding specialist with 75K synthetic instruction data
  • Gemma3n - Multimodal model optimized for everyday devices (phones, tablets, laptops)
  • Granite 3.3 - IBM's improved models with 128K context (2B and 8B variants)
  • Gemma3:270M - Ultra-compact 270M model with 0.75% battery usage on mobile devices

Enhanced Menu Organization:

  • Expanded model selection from 50+ to 55+ latest 2025 models
  • Added new performance benchmarks and descriptions
  • Updated all optimization profiles and system recommendations

Features

  • ** Universal Model Support** - Works with any Ollama model (llama3.2, mistral, gemma2, etc.)
  • ** 6 Profile Suggestions** - Pre-configured parameter suggestions: Balanced, Technical, Creative, Code, Reasoning, Roleplay
  • ** System Resource Monitoring** - Real-time GPU, RAM, and disk space monitoring
  • ** Auto-Download & Validation** - Intelligent model downloading with space checking
  • ** Professional Interface** - Clean, colorful terminal UI with progress indicators
  • ** Configuration Management** - File-based config with runtime overrides
  • ** Custom Parameters** - Full manual parameter control and Modelfile creation
  • ** Error Logging** - Comprehensive logging to ~/.config/universal-ollama-optimizer/errors.log

Requirements

  • Linux (Ubuntu 20.04+, other distros compatible)
  • Bash 5.0+ (pre-installed on most systems)
  • Ollama (Download here)
  • 8GB+ RAM recommended
  • GPU optional but recommended (NVIDIA with CUDA support)

Quick Start

One-Command Install & Run

# Download and run
curl -fsSL https://raw.githubusercontent.com/jtgsystems/universal-ollama-optimizer/main/universal-ollama-optimizer.sh -o universal-ollama-optimizer.sh
chmod +x universal-ollama-optimizer.sh
./universal-ollama-optimizer.sh

Manual Installation

# Clone repository
git clone https://github.com/jtgsystems/universal-ollama-optimizer.git
cd universal-ollama-optimizer

# Make executable
chmod +x universal-ollama-optimizer.sh

# Run
./universal-ollama-optimizer.sh

Usage

Interactive Mode (Default)

./universal-ollama-optimizer.sh
  1. Select Model - Choose from available models or enter new model name
  2. Choose Profile - Select parameter suggestion profile (1-9)
  3. Launch Model - Model starts with parameter instructions displayed
  4. Apply Parameters - Use /set commands manually during the session

Example Session

Enter model name: llama3.2:latest

Model Information:
  • Model: llama3.2:latest
  • Size: 4.7GB
  • Parameters: 3.2B

System Status:
  • GPU Memory: 16380 MB
  • System RAM: 32GB
  • Available Disk: 150GB

Parameter Profiles:
  1) Balanced          - General purpose (temp: 0.5)
  2) Technical/Factual - Precise answers (temp: 0.2)
  3) Creative Writing  - Imaginative content (temp: 1.0)
  4) Code Generation   - Programming tasks (temp: 0.15)
  5) Reasoning/Logic   - Problem solving (temp: 0.3)
  6) Roleplay/Chat     - Conversational (temp: 0.8)

Select profile [1-6]: 1

Starting llama3.2:latest with Balanced profile suggestion...

To apply these parameters during your session, use:
/set parameter temperature 0.5
/set parameter top_p 0.85
/set parameter top_k 30
/set parameter repeat_penalty 1.08

Configuration

Config File Location

~/.config/universal-ollama-optimizer/config.conf

Example Configuration

# Default profile (balanced, technical, creative, code, reasoning, roleplay)
DEFAULT_PROFILE="balanced"

# Auto-start Ollama service if not running
AUTO_START_OLLAMA=true

# System monitoring
SHOW_SYSTEM_INFO=true
SHOW_GPU_INFO=true

# Download settings
DOWNLOAD_TIMEOUT=1800
MIN_DISK_SPACE_GB=5

Parameter Profiles (Suggestions)

Profile Temperature Top-P Top-K Best For
Balanced 0.5 0.85 30 General use, Q&A
Technical 0.2 0.8 20 Documentation, facts
Creative 1.0 0.95 50 Stories, brainstorming
Code 0.15 0.7 15 Programming, debugging
Reasoning 0.3 0.75 25 Logic, analysis
Roleplay 0.8 0.9 40 Character chat

Advanced Features

Custom Parameters

Choose option 7 for manual parameter configuration:

  • Temperature (0.0-2.0)
  • Top-P (0.0-1.0)
  • Top-K (1-100)
  • Context length
  • Max tokens per response
  • Repeat penalty

Modelfile Creation

Choose option 8 to create and save custom model configurations:

# Creates permanent optimized model variants
# Example: llama3.2-coding, mistral-creative

Runtime Commands

Once model is running, use these commands:

/set parameter temperature 0.7    # Adjust parameters
/set system "You are a coding assistant"  # Change system prompt
/show parameters                  # View current settings
/save my-conversation            # Save session
/load my-conversation            # Load session

Troubleshooting

Common Issues

Ollama Not Found

# Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh

Permission Denied

# Make script executable
chmod +x universal-ollama-optimizer.sh

Model Download Fails

# Check internet connection and disk space
df -h
ping ollama.ai

GPU Not Detected

# Check NVIDIA drivers
nvidia-smi

Project Structure

universal-ollama-optimizer/
├── universal-ollama-optimizer.sh    # Main script
├── README.md                        # This file
├── LICENSE                          # MIT License
└── .github/
    ├── workflows/
    │   └── test.yml                 # CI/CD tests
    ├── ISSUE_TEMPLATE/              # Issue templates
    └── PULL_REQUEST_TEMPLATE.md     # PR template

Contributing

We welcome contributions! Please see our Contributing Guidelines.

Development Setup

# Fork and clone
git clone https://github.com/jtgsystems/universal-ollama-optimizer.git
cd universal-ollama-optimizer

# Test the script
./universal-ollama-optimizer.sh

# Run with debug mode
bash -x universal-ollama-optimizer.sh

Reporting Issues

  • Use the issue tracker
  • Include OS version, Ollama version, and error messages
  • Provide steps to reproduce

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

  • Jesus Christ - Our Lord and Saviour, for all gifts and abilities
  • Ollama Team - For creating an excellent local AI platform
  • Community Contributors - For feedback and improvements
  • JTGSYSTEMS.COM - For development and maintenance

Support

Recommended Ollama Models (September 2025)

Based on latest community recommendations and performance benchmarks

** Top-Tier Models (September 2025)**

** Best Overall Performance**

  • llama3.3:70b - Meta's flagship 2025 model, rivals GPT-4 performance locally
  • glm4:latest - Ranks 3rd overall, beats Llama 3 8B in benchmarks
  • llama3.1:8b - Community favorite, best balance of performance and efficiency
  • llama3.1:70b - High-performance for complex reasoning and enterprise use
  • deepseek-r1 - Powerhouse for deep logical reasoning and analysis

** Premier Coding Models**

  • deepseek-coder:33b - #1 coding model, excels at complex programming tasks
  • magicoder:latest - OSS-Instruct trained specialist, 75K synthetic data
  • qwen3-coder:30b - Alibaba's latest 2025 coding model with major improvements
  • codellama:34b - Meta's specialized coding model with excellent context understanding
  • qwen2.5-coder:32b - Previous Alibaba model with solid code generation

** Resource-Efficient Champions**

  • granite3.3:8b - IBM's improved model with 128K context, rivals Llama 3.1
  • mistral:7b-instruct - Community-recommended for beginners, excellent performance/resource ratio
  • phi4:14b - Microsoft's 2025 state-of-the-art efficiency model
  • granite3.3:2b - IBM's efficient enterprise model for edge deployment
  • llama3.2:3b - Compact Llama for lightweight deployments
  • gemma3:270m - Ultra-compact 270M model, 0.75% battery usage on mobile
  • gemma2:9b - Google's efficient model, great for general tasks

** Creative & Multimodal**

  • gemma3n:latest - Multimodal optimized for everyday devices (phones, tablets)
  • llava:latest - Leading vision model for image analysis and VQA
  • qwen-vl - Advanced multimodal model for document and image processing
  • gemma2:27b - Excellent for creative writing and content generation

** Performance Matrix (September 2025)**

Use Case Top Model Alternative RAM Required Best Profile
General Chat llama3.3:70b glm4:latest 64GB / 9GB Balanced
Code Development deepseek-coder:33b magicoder:latest 32GB / 7GB Code
Reasoning Tasks deepseek-r1 glm4:latest 16GB / 9GB Reasoning
Creative Writing gemma2:27b gemma3n:latest 32GB / 8GB Creative
Resource-Limited granite3.3:2b gemma3:270m 2GB / 300MB Balanced
Vision/Multimodal llava:latest gemma3n:latest 16GB / 8GB Technical
Enterprise/128K Context granite3.3:8b llama3.1:8b 8GB Technical
Edge/Mobile gemma3:270m granite3.3:2b 300MB / 2GB Balanced

** Quick Download Commands (September 2025)**

# Most recommended overall (2025 flagship)
ollama pull llama3.3:70b-instruct

# High-performance alternative (ranks 3rd overall)
ollama pull glm4:latest

# Specialized coding assistant with OSS-Instruct training
ollama pull magicoder:latest

# Premier coding powerhouse
ollama pull deepseek-coder:33b

# Enterprise model with 128K context
ollama pull granite3.3:8b

# Ultra-efficient for mobile/edge (270M parameters)
ollama pull gemma3:270m

# Multimodal for everyday devices
ollama pull gemma3n:latest

# Advanced reasoning powerhouse
ollama pull deepseek-r1

# Vision and image analysis
ollama pull llava:latest

# Best for beginners/limited hardware
ollama pull mistral:7b-instruct

** 2025 Community Insights**

  • Llama 3.3 has become the gold standard for local deployment
  • DeepSeek models dominate coding and reasoning benchmarks
  • Mistral 7B remains the go-to recommendation for newcomers
  • Vision models like LLaVA are gaining massive adoption
  • Over 1,700+ models now available in Ollama ecosystem

Note: Model availability and performance may vary. Check ollama.com/library for the latest models.

Repository: universal-ollama-optimizer | Developer: JTGSYSTEMS.COM | Technology: Bash, Linux, Ollama, AI

About

Professional bash script for launching and managing Ollama AI models with parameter profile suggestions, system monitoring, and automated configuration. Not an automatic optimizer - provides suggested /set commands for manual application.

Topics

linux shell bash automation tools ai optimization launcher model-management terminal-ui model-optimizer ai-models llm local-ai ollama

Resources

Readme

License

MIT license

Contributing

Contributing
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases 1

🚀 Universal Ollama Optimizer v1.0.0 - Initial Release Latest
Sep 16, 2025

Packages

No packages published

Contributors 3

  •  
  •  
  •  

Languages