Meta Announces a Fine-Tuned Llama to Deploy in Space

Meta Platforms Inc. has partnered with Booz Allen Holding Corp., a U.S. government contractor, to create an AI system for the International Space Station.

The project introduces "Space Llama," an AI system built on a fine-tuned version of Meta’s Llama 3.2 language model series. Space Llama is designed to assist with scientific research conducted in the ISS National Laboratory, the research hub located aboard the station.

The goal in building the tech stack is to not only replace paper documents onboard the ISS National Lab, but also decrease reliance on instructions received from the ground while deployed at the edge.

The result will be lower costs, less computing power required, and faster outcomes when responding to onboard anomalies or maintenance issues, without depending on Earth-bound internet.

Space Llama evolves and builds upon successful efforts by Booz Allen to deploy and operate the first known large language model (LLM) in space.

This moment sees the new addition of a fine-tuned version of Meta’s Llama vision AI capabilities (recently made available to US government agencies and private sector partners) to help process and interpret images and text at the farthest reaches of space.

This allows users to quickly and reliably access essential technical references, manuals, and instructions leveraging the power of generative and vision AI to address on-board issues at speed.

“In the power and communications constrained environment of space, the ability to quickly assess, research and solve on-board issues is paramount. This proof of concept overcomes the historical limitations of low-power, low-capability space-based computing by enabling simple, power-efficient AI use in space without relying on Earth network connectivity," Bill Vass, chief technology officer at Booz Allen, said.

The Space Llama tech stack, undertaken as internal research and development, is comprised of Booz Allen’s modular open architecture platform A2E2™ (AI for Edge Environments) and a fine-tuned version of Meta’s Llama 3.2, using Hewlett Packard Enterprise’s (HPE) Spaceborne Computer-2 and NVIDIA CUDA software. The result: a compact, energy-efficient system similar to those used in satellites that can be deployed locally on the ISS National Lab without requiring power, communication or computation from Earth.

This design – open architecture to ensure flexibility in our solution design, now and in the future – results in lower costs and strong technical performance, in addition to enabling predictive maintenance and faster repairs, freeing up users to do mission critical work.

“This is just the beginning. We see a future where open source AI models like Llama will play a vital role in fueling space exploration and research – enabling astronauts to undertake the next level of experimentation to solve complex scientific problems and drive innovation in entirely new ways," Ahmad Al-Dahle, Meta VP and Head of GenAI, said.