Updated  5 hours ago

A 60M-parameter GPT pretrained and instruction-tuned on a single 8GB-RAM NVIDIA Jetson, offering a native 2048-token context and real multi-turn conversation support.