One email each morning — yesterday's AI, sortedGet it in your inbox

METAL LAB

AlexsJones/llmfit

33,132+1,545 this weekRust

One terminal command tells you which AI models will actually run on your computer

llmfit is a terminal tool that detects your machine's RAM, CPU, and GPU, then ranks hundreds of language models by how well they'd actually run on your hardware. A single command shows a scored table right in your terminal. It also lets you measure real speed on your own machine and share results back so estimates get better for everyone.

What it does

  1. Supports hundreds of models across multiple local runtime providers, including Ollama, llama.cpp, MLX, Docker Model Runner, and LM Studio
  2. Detects RAM, CPU, and GPU/VRAM, and scores each model across four dimensions: memory fit, estimated speed, quality, and context length
  3. Speed estimates use a memory-bandwidth-based formula combined with real measurements collected from the community
  4. Users can benchmark on their own hardware and submit results as a PR from the interactive TUI, with merged results becoming verified numbers for others with identical hardware in the next release
  5. Handles multi-GPU setups and MoE (Mixture-of-Experts) architectures, which only activate a subset of parameters

Why it matters

Anyone trying to run AI models locally usually has to guess or trial-and-error their way into finding a model that fits their hardware, and this tool automates that decision. As more real user measurements accumulate, the project's estimates shift from guesses to verified numbers that benefit the whole community.

Terms in this repo

  • TUI · A text-based interactive interface you navigate inside a terminal
  • MoE (Mixture-of-Experts) · A neural network design that activates only part of its total parameters at once, needing less memory than its full size suggests
  • VRAM · Dedicated memory built into a graphics card (GPU)
  • tok/s · Tokens processed per second, a speed measure for language models
  • TTFT · Time-to-first-token, how long it takes before a model starts producing its first output after a prompt

Repository description (English)

Hundreds of models & providers. One command to find what runs on your hardware.

Open on GitHub

Trending repos

All repos →

Latest from METAL LAB