azom dev

Where we're going, we don't need ram. Trying to make some SLMs (some not being transformers) with custom engines and infinite context (when possible).