Internal Chain-of-Thought: Empirical Evidence for Layer-wise Subtask Scheduling in LLMs

Open in new window