What it does
llamafile is a Mozilla Builders project, now revamped by Mozilla.ai, that aims to make open LLMs more accessible to developers and end users. It combines llama.cpp with Cosmopolitan Libc into one framework, so a model and its runtime become a single executable, called a llamafile. To get started, you download a prebuilt llamafile from Hugging Face, mark it executable on macOS or Linux, and run it. On Windows you rename the file to add the .exe extension, and only executables under 4GB can run. You can also run the llamafile binary against your own external GGUF weights, and the documentation covers creating llamafiles. The project also includes whisperfile, a single-file speech-to-text tool built on whisper.cpp that transcribes and translates audio files. Versions from 0.10.0 use a new build system that tracks recent llama.cpp more closely, which means newer model support but possibly fewer of the older features; earlier versions remain on the releases page.
Capabilities
- Models packaged as single-file executables that need no installation
- Built from llama.cpp and Cosmopolitan Libc
- Runs on most operating systems and CPU architectures
- Prebuilt llamafiles, or run your own GGUF weights
- Whisperfile for audio transcription and translation
- Documentation for creating your own llamafiles
When to use it
- Trying a local LLM without installing anything
- Handing others a model as one self-contained file
- Running GGUF models locally on your own hardware
- Transcribing audio files on your own machine with whisperfile
