Developers from Mozilla presented The first release of the utility LLAMAFILE , which allows you to create universal executable files to launch large language models of machine learning (LLM). Using Llamafile, you can take a file with machine learning model parameters in GGUF format and convert it to an executable file that can be launched in six different operating systems on equipment with AMD64 and ARM64 processors. The toolkit is written on C/C ++ and spreads under the license Apache 2.0.
Framwork llama.cpp with the implementation of the model llama on C/C ++, which is compiled into a universal executable file in format ape (Actoally Portable Executable), combined with a model file. The compiled code is associated with the standard SI-Bibludeck cosmopolitan that gives the opportunity to create assemblies of applications launched in Linux, FreeBSD, OpenBSD, NETBSD and Windows. The code is stored in the options for AMD64 and ARM64 architectures, and also includes components for removing calculations on GPU NVIDIA (used by CUDA).