TensorFold Living Weights: Does It Really Change the Model?
The short answer is yes: TensorFold’s Living Weights really does modify a model checkpoint. It is not retrieval-augmented generation, a prompt cache, or merely a LoRA that must remain attached during inference. During learning, TensorFold uses a temporary low-rank update as scaffolding. If its checks accept the lesson, it folds that update into selected dense projection matrices and rewrites the model’s safetensors checkpoint. Restarting from that folder without the live-learning flag still loads the changed parameters. ...