LLM.f90 - Minimal Large Language Model Inference Framework

A low‑dependency Fortran framework for LLM inference, showing zero‑dependency implementation, matrix operations, and support for Llama, Phi, and other models.

Overview

This is a zero-dependency LLM inference framework written in Fortran. The motivation was to have a compact, fast, framework that can run modern LLMs that is easy to understand and modify. It’s written in Fortran because the language provides numpy-style matrix manipulation in a fast, compiled language. Currently supports Llama, Phi, and a few other models. This is an open source project.

Links

https://github.com/rbitr/llm.f90
Enables pure Fortran LLM inference, CPU-optimized, with zero external dependencies.

Tech stack