Phi-1.5 exhibits many of the traits of much larger LLMs such as the ability
to “think step by step” or perform some rudimentary in-context learning.