Reproduce “CAMP: Context-Aware Mapping and Placement for LLMs on 3D-IMC Arrays” 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
Orhun Görkem, Athanasios Vasilopoulos, Georgios Pagonis, Victoria Clerico, Hadjer Benmeziane, Irem Boybat, Hsinyu Tsai, Abu Sebastian and 1 more
CAMP optimizes weight placement in simulated 3D-IMC accelerators for LLM inference, predicting overlap between matrix-vector multiplications and refining placements through simulator feedback. Simulation reports utilization rising from 42.4% to 60.3%. For Qwen-72B, MVM throughput reaches 2.0× Greedy-Util at batch size 16; gains emerge at larger batches, while dense models at batch size 1 match the TP-maximizing baseline. Results are simulated, not physical-chip measurements.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.