llmfit

Why we included this project

Picking a model for local inference is usually the slow part: you know your machine's specs, but not which of the hundreds of available models will actually run well on them. llmfit reads your RAM, CPU, and GPU, then scores a large catalog of models on memory fit, estimated speed, quality, and context, so you can check before you download. It comes as an interactive TUI or a scriptable CLI, supports multi-GPU setups, MoE architectures, and dynamic quantization, and works with the common local runtimes: Ollama, llama.cpp, MLX, and LM Studio. Speed estimates rest on a memory-bandwidth model and community measurements, and a built-in benchmark command measures real tok/s on your hardware and lets you contribute those results back to the project. For developers and small teams, it is a practical sizing tool rather than another model runner.