Updated  7 hours ago

A 60M-parameter GPT trained from scratch on a single 8GB-RAM NVIDIA Jetson, offering a native 2048-token context for raw text completion. Base pretrained checkpoint, not instruction-tuned.