Comment by airgapstopgap

Long-context tasks are not really the true gap between LLaMA and GPT series, but important result.