Revisiting In-context Learning Inference Circuit in Large Language Models

2025-01-22 · ICLR 2025 Poster · anchor

Findings

IC-578LLMs encode input text as linearly separable representations in forerunner token hidden states, emerging in early layers and enhanced by in-context demonstrations
IC-579ICL hidden states exhibit positional bias: representations of the same input are more similar when the input appears at similar positions in the sequence
IC-580The 3-step ICL inference circuit (text encoding, semantics merge, feature retrieval) is a dominant causal mechanism, as ablating the corresponding attention connections significantly degrades ICL accuracy
IC-581Induction heads for ICL operate on task-specific attention subspaces, with partial overlap across tasks, and the geometry of these subspaces explains demonstration saturation