spec2llm 0.1.0

Find the best LLMs for your hardware specs

Spec2LLM

Find the best LLMs for your hardware.

Detects your system specs (CPU, GPU, RAM, storage, OS) and recommends compatible LLMs ranked by performance fit. Works on Linux, Windows, and macOS — including Apple Silicon.

pip install spec2llm
spec2llm recommend

Quick Start

# See what models fit your system
spec2llm recommend

# Search for specific models
spec2llm search deepseek

# Compare two models
spec2llm compare llama-3.2-1b-q4 mistral-7b-q4

# Discover new models
spec2llm catalog update

# JSON output for scripting
spec2llm scan --json

Features

• Cross-platform hardware detection — CPU (cores, freq), GPU (NVIDIA VRAM, AMD, Apple Silicon), RAM, storage, OS
• Smart scoring — VRAM headroom (40%), RAM headroom (20%), GPU compute tier (20%), CPU cores (10%), Apple Silicon bonus (10%)
• Curated catalog — 40+ popular models (Llama 3.x, Mistral, Gemma, Qwen, DeepSeek, Phi, and more)
• Auto-discovery — Fetches new models from Ollama registry with estimated requirements
• Apple Silicon — Detects unified memory and adjusts scoring
• JSON output — --json flag on all commands

Commands

Command	Description
spec2llm scan	Detect and display all system hardware specs
spec2llm recommend	Find and rank best-matching LLMs
spec2llm search <query>	Search the model catalog
spec2llm list	Browse all models
spec2llm install <model>	Show install commands (Ollama, HuggingFace)
spec2llm compare <a> <b>	Compare two models vs your system
spec2llm catalog update	Fetch new models from Ollama registry

How It Works

1. Scan detects your CPU, RAM, GPU, storage, and OS
2. Match filters models that fit your VRAM, RAM, and storage
3. Score (0-100): VRAM headroom (40) + RAM headroom (20) + GPU tier (20) + CPU cores (10) + Apple Silicon bonus (10)
4. Recommend returns a sorted table with scores

Requirements

• Python 3.9+

Platform Support

Feature	Linux	Windows	macOS
CPU / RAM / Storage	✅	✅	✅
NVIDIA GPU (VRAM)	✅	✅	❌
AMD / Intel GPU	✅ lspci	✅ wmi	❌
Apple Silicon	N/A	N/A	✅

License

MIT