llm-checker

v2.7.2

Published

3 months ago

Intelligent CLI tool with AI-powered model selection that analyzes your hardware and recommends optimal LLM models for your system

LLM Checker - Intelligent Ollama Model Selector

Advanced CLI tool that analyzes your hardware and intelligently recommends the optimal Ollama LLM models for your system with automatic installation detection.

Designed specifically for Ollama - Integrates with 177+ models from the complete Ollama model library to find the best models for your hardware configuration.

✨ Key Features

🎯 Multiple Model Recommendations

--limit flag: Show multiple compatible models instead of just one
Ranked display: See top 3, 5, or 10 models with compatibility scores
Smart alternatives: Get backup options with unique installation commands
Hardware-aware filtering: Automatically excludes unreasonably large models

✅ Automatic Installation Detection

Real-time detection: Automatically detects already installed Ollama models
Smart Quick Start: Shows ollama run for installed models, ollama pull for new ones
Status indicators: Clear "Already installed" vs "Available for installation" status
No duplicate suggestions: Won't suggest installing models you already have

🧠 Intelligent Use Case Categories

7 specialized categories: coding, creative, reasoning, multimodal, embeddings, talking, general
Typo tolerance: Handles common misspellings (e.g., "embedings" → "embeddings")
Smart filtering: Each category shows models optimized for that specific use case
Category-aware scoring: Different scoring weights for different use cases

📊 Real Model Data

177+ models with accurate size data from Ollama Hub
Real file sizes: Uses actual model sizes instead of parameter estimates
Pre-classified categories: All models categorized by capabilities
Static database: Stable, reliable model information without dynamic updates

🚀 Advanced Algorithm

Multi-objective ranking with hardware-size matching
Hardware utilization scoring: Penalizes models that underutilize high-end hardware
Smart size filtering: Filters out models too large for your system
Cross-platform compatibility: macOS, Windows, Linux with GPU detection

🚀 Quick Start

Installation

npm install -g llm-checker

Prerequisites

Node.js 16+
Ollama installed and running (Download here)

Basic Usage

# Get the best model for your hardware
llm-checker check

# Show top 5 compatible models
llm-checker check --limit 5

# Get coding-specific models
llm-checker check --use-case coding --limit 3

# Find creative writing models
llm-checker check --use-case creative --limit 5

📋 Available Use Cases

| Use Case | Description | Example Models | |----------|-------------|----------------| | coding | Programming and code generation | CodeLlama, DeepSeek Coder, CodeQwen | | creative | Creative writing and content | Dolphin, Wizard, Uncensored models | | reasoning | Logic and mathematical reasoning | DeepSeek-R1, Phi4-reasoning, Llama3.2-vision | | multimodal | Image analysis and vision tasks | Llama3.2-vision, LlaVa | | embeddings | Text vectorization and search | BGE, E5, embedding models | | talking | General conversation and chat | Llama, Mistral, Qwen (excluding specialized) | | general | Balanced, versatile models | Mixed selection prioritizing chat/reasoning |

🛠️ Command Reference

Main Commands

# Hardware analysis with model recommendations
llm-checker check [options]

# Get intelligent recommendations
llm-checker recommend [options]

# List available models
llm-checker list-models

# AI-powered model evaluation
llm-checker ai-check

# Ollama integration info
llm-checker ollama

Options

# Show multiple models
--limit <number>          Number of models to show (default: 1)

# Use case filtering
--use-case <case>         Specify use case (coding, creative, reasoning, etc.)

# Output control
--no-verbose              Clean, minimal output
--include-cloud           Include cloud-based models

# Filtering
--filter <type>           Filter by model type
--ollama-only             Only show Ollama-available models

📖 Examples

Basic Recommendations

# Single best model
llm-checker check
# Output: Shows #1 model with installation command

# Multiple options
llm-checker check --limit 5
# Output: Shows top 5 ranked models with scores

Use Case Specific

# Coding models
llm-checker check --use-case coding --limit 3
# Output: CodeLlama, DeepSeek Coder, CodeQwen with install commands

# Creative writing
llm-checker check --use-case creative --limit 5
# Output: Dolphin, Wizard, creative-optimized models

# Reasoning tasks
llm-checker check --use-case reasoning --limit 3
# Output: DeepSeek-R1, Phi4-reasoning, specialized reasoning models

Installation Detection

llm-checker check --limit 5 --use-case coding

Example output:

TOP 5 COMPATIBLE MODELS

#1 - CodeLlama 7B
Size: 3.8GB
Compatibility Score: 84.88/100
Status: Already installed in Ollama

#2 - Qwen 2.5 7B  
Size: 5.2GB
Compatibility Score: 83.78/100
Status: Available for installation

QUICK START
1. Start using your installed model:
   ollama run codellama:7b

Alternative options:
   2. ollama pull qwen2.5:7b
   3. ollama pull codeqwen

🔧 Advanced Features

Hardware Tier Detection

Flagship: RTX 5090/H100 tier → 30B-175B models (new!)
Ultra High: RTX 4090/A100 tier → 20B-105B models
High: RTX 4080/Apple Silicon 32GB → 8B-50B models
Medium: RTX 4070/Apple Silicon 16GB → 3B-20B models
Low: Budget systems → 1B-8B models
Ultra Low: Very limited systems → <3B models

Smart Filtering

Automatically excludes models >25GB for systems with <32GB RAM
Penalizes tiny models on high-end hardware
Prioritizes models in the "sweet spot" for your hardware tier
Removes duplicate commands from alternatives

Cross-Platform Support

macOS: Apple Silicon optimization with unified memory
Windows: NVIDIA/AMD GPU detection with device ID mapping
Linux: Full GPU compatibility with proper driver detection

🔒 Security & Caching

Installation guidance: For Ollama on Linux, prefer official package managers or documented methods rather than piping remote scripts into the shell. See the official installation docs: https://github.com/ollama/ollama/blob/main/docs/linux.md
Cache location: The Ollama model cache used by LLM Checker is stored at ~/.llm-checker/cache/ollama.
Backward compatibility: Existing cache files in the legacy path (src/ollama/.cache) are still read if present, but new cache writes go to the home directory.

🚀 What's New in v2.7.0

🎯 Complete Windows High-End GPU Optimization

NEW Flagship Tier: RTX 5090, H100, A100 now properly recognized
Enhanced RTX 50xx Support: Up to 50% RAM offload capacity (was 30%)
Smarter Memory Utilization: 95% VRAM efficiency for flagship GPUs
Better Model Range: Flagship systems now handle 30B-175B models

🔄 Improved Compatibility Classification

Realistic Thresholds: Compatible 65%+ (was 75%+), Marginal 45-64%
Better Category Filtering: All use cases now work correctly on both platforms
Cross-Platform Parity: Windows and Mac now have similar model counts

🧠 Enhanced Multi-Objective Scoring

Hardware Match Priority: Increased from 5% to 30% weight for better sizing
Platform-Specific Optimization: Apple Silicon vs Windows GPU paths
Quality-Speed Balance: Reduced speed emphasis for high-end hardware

🛠️ Bug Fixes

Chat Category Filter: Now correctly excludes coding models
Embeddings Fallback: Proper filtering when no compatible models found
Score Display: Fixed 5/100 score bug in CLI output
Platform Detection: Now uses hardware OS for simulation support

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Development

git clone https://github.com/Pavelevich/llm-checker.git
cd llm-checker
npm install

# Run locally
node bin/enhanced_cli.js check --limit 5

📄 License

MIT License - see LICENSE file for details.

👨‍💻 Author

Pavel Chmirenko - GitHub | Email

⭐ Support

If you find LLM Checker useful, please consider:

Starring the repository ⭐
Contributing improvements 🛠️
Reporting issues 🐛
Sharing with others 📢

Buy me a coffee: buymeacoffee.com/pavelchmirenko