A project to distribute UAE- and Saudi Arabia-origin LLMs as llamafiles — single-file executables that require no installation.
| Model | Developer | Architecture | Status |
|---|---|---|---|
| Falcon-H1 7B / 34B | TII (UAE) | Mamba-2 + Transformer | Ready |
| Falcon-H1-Arabic 7B | TII (UAE) | Mamba-2 + Transformer | Waiting for GGUF release |
| Jais 13B / 30B | G42 / MBZUAI (UAE) | Dense Transformer | Ready |
| ALLaM 7B | SDAIA / HUMAIN (Saudi Arabia) | Dense Transformer | Ready |
A single-file format that lets you run LLMs locally with no installation and no dependencies. Works on Windows, macOS, Linux, and ARM. For details, see docs/what-is-gguf.md.
See docs/how-to-run.md for model-specific commands and API examples.
# Set environment variables before running
export LLAMA_CPP_DIR=~/llama.cpp
export LLAMAFILE_DIR=~/llamafile
./scripts/convert.sh tiiuae/Falcon-H1-7B-Instruct falcon-h1-7b Q4_K_MFor details, see the comments in scripts/convert.sh.
This repository includes a Makefile for common tasks:
make help # Show all available commands
make lint # Lint shell scripts (shellcheck, or bash -n fallback)
make test # Run repository validation tests
make clean # Remove temporary work/ directoryTo build a llamafile via Make:
make convert ARGS="tiiuae/Falcon-H1-7B-Instruct falcon-h1-7b Q4_K_M".github/workflows/build.yml provides a manual GitHub Actions workflow that downloads an official GGUF file and bundles it into a llamafile. It requires a HF_TOKEN repository secret for HuggingFace uploads.
- PLAN.md — Project planning document
- docs/what-is-gguf.md — Explanation of the GGUF format
- docs/how-to-run.md — How to run the generated llamafiles