Reproduce “Disaggregated LLM Inference in Space: A Communication Feasibility Study” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Arno Troch, Roger Sanchez-Vital, Seyyidahmed Lahmer, Bartek Kozicki, Jeroen Famaey, Michael Peeters
The authors model how communication affects language model serving across satellites. They compare PDD, which transfers a KV cache once per request, with AFD, which exchanges activations twice per MoE layer per generated token. Under a 1 ms communication budget, an 800 Gbps link supports AFD separation of about 1–3 km across the modeled MoE models. These estimates assume exposed transfers and are not measured deployment results.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.