one-pass inference

LLM2R marks a shift from large language models to three-dimensional reassemblers.

LLM2R casts 3D part assembly as sequence modeling, using a decoder-only LLM with summary tokens to predict SE(…