v4.0 neural engine — GatedConv+Q-NFRE, BPE tokenizer, 4.5M/10M models, terminal 320d56f verified FerrellSyntheticIntelligence commited on 14 days ago