LLllmfit
A hardware-inspection tool that estimates model fit and supports terminal, CLI and web/API workflows.
About this project
llmfit inspects a computer's CPU, system memory, graphics hardware and accelerator configuration, then estimates how selected language-model sizes and quantizations fit that hardware. It supports an interactive terminal interface, a conventional CLI and a web dashboard or HTTP API. Estimates and measured benchmark results are different kinds of information.
The README documents Windows x86_64, Intel and Apple Silicon macOS, and x86_64 or ARM64 Linux. Installation options include official release binaries, Scoop, Homebrew, MacPorts and uv or pip, with a local web/API mode started through llmfit serve. Actual inference uses the chosen model runtime and downloaded model rather than the inspection tool alone.
The code is MIT-licensed. Model weights have separate licenses and requirements, and an estimated fit or speed is not a successful local run. Benchmarking needs a model and serving runtime; optional sharing contributes measurements to the project. Windows artifacts are signed only when the relevant signing job succeeds, so users should check the artifact they download.
Why it is worth a look
- Inspect CPU, memory, GPUs and accelerator configuration
- Estimate model memory fit and generation speed
Before you use it
- Fit and speed estimates are not a compatibility guarantee
- Actual benchmarking needs a chosen runtime and model download
- Model licenses remain separate
- Check the signing status of the downloaded Windows artifact
Fit and speed estimates are not a compatibility guarantee; Actual benchmarking needs a chosen runtime and model download; Model licenses remain separate; Check the signing status of the downloaded Windows artifact; Confirmed requirements: Command line; Some requirements have not been verified. Check the official setup instructions.
What it can do
- Inspect CPU, memory, GPUs and accelerator configuration
- Estimate model memory fit and generation speed
- Terminal, CLI and web/API workflows
Who it fits
Shortlisting model sizes for existing hardware; Comparing estimated fit with personal benchmark results
A guaranteed prediction of local model performance; Assuming the inspection command alone serves every listed model
Getting started
- Install the official tool using the documented path for your platform
- Inspect the detected hardware and estimated model-fit table
- If needed, configure a separate runtime and model to obtain your own benchmark rather than relying on the estimate