Mini-AGI – dynamic continual learning model trained from scratch on 8GB VRAM

Mini-AGI is a continual learning language model that trains from scratch on a single 8GB VRAM GPU, with weights stored on disk and dynamically loaded as needed, allowing parameter count to be bounded by available disk space rather than memory. The model can continuously learn from new data without catastrophic forgetting, growing its capacity during training and pruning unused components, making it accessible for individual users to train and customize on modest hardware. Currently a small experimental model, it demonstrates that personal language models capable of ongoing learning are feasible on consumer-grade GPUs.