POCKET-35B — a 35B model that runs on iPhone and on PC with no GPU, using stock llama.cpp POCKET-35B A 35B model that runs on your iPhone and on your PC with no GPU. Just stock llama.cpp. No fork, no CUDA, no cloud. iPhone-ready 5 GB · Korean & English CPU-only 27 tok/s · no GPU needed Stock runtime LM Studio · Ollama · PocketPal 34.66B total · ~3B active/token · sparse Mixture-of-Experts, Korean-tuned 35B runs here. no GPU