volotat/mini-AGI ? reverse-engineered prompt

Build me a small Python project for a continual learning language model that can train from scratch on a laptop with just 8 GB of VRAM and keep learning from a stream of text without forgetting everything.

I want one codebase that can both train and serve the model, with the same inference path used during training and at runtime. It should read plain text corpora, keep model weights on disk if needed, load them as it trains, and expand or prune capacity as the model needs it. Please include a simple way to point it at a corpus, start training, watch progress, and generate sample text from the current checkpoint.

Make it feel like an experimental toy project, not a huge production system. A basic dashboard or local status page would be great if that fits naturally. Save checkpoints and training history to disk so I can inspect the run later. If you need to check current docs for any libraries, go ahead and look them up online.