A transformer architecture alteration to incentivise externalised reasoning

1. 1Introduce an early-exit mechanism at intermediate layers of the transformer architecture
2. 2Train the model to exit at shallower layers when the next token can be predicted without deep computation
3. 3Calibrate the model to determine optimal exit points
4. 4Incentivize the model to exit as early as possible to promote externalized reasoning

AI engineers and ML researchers can benefit from this alteration as it enables more efficient and verbose reasoning in LLMs, potentially improving model performance and interpretability

💡 Introducing an early-exit mechanism can improve model efficiency and promote more verbose reasoning

💡 New transformer architecture alteration promotes externalized reasoning in LLMs