This is the repo for the Layer_Gradient project, in which we try to understand the layer-wise gradient behaviors when LLMs are finetuned on Fast vs. Slow Thinking. What makes a difference in the ...