llmfit

Which open-weight language models will actually run well on this machine? llmfit answers that from the terminal. It inspects CPU cores, system RAM, discrete or integrated GPUs, VRAM and unified memory (NVIDIA CUDA, Apple Silicon, AMD ROCm, Intel OneAPI), then scores each model on quality, speed, fit and context across quantization formats such as GGUF, AWQ, GPTQ and EXL2. The default interface is an interactive TUI; a classic CLI mode, a web dashboard and a REST API are also included, and it understands multi-GPU setups and MoE architectures. A newer benchmark feature downloads a model, measures real tokens per second on your hardware and lets you submit the numbers back as a pull request from inside the TUI, so people with identical hardware see measured figures instead of estimates. Written in Rust and installable through Homebrew, Scoop, MacPorts, uv, a Docker image or signed release binaries for macOS, Linux and Windows.