What needs to go right for an induction head? A mechanistic study of in-context learning circuits and their formation

Aaditya K Singh, Ted Moskovitz, Felix Hill, Stephanie C.Y. Chan, Andrew M Saxe
Proceedings of the 41st International Conference on Machine Learning, PMLR 235:45637-45662, 2024.

Abstract

In-context learning is a powerful emergent ability in transformer models. Prior work in mechanistic interpretability has identified a circuit element that may be critical for in-context learning – the induction head (IH), which performs a match-and-copy operation. During training of large transformers on natural language data, IHs emerge around the same time as a notable phase change in the loss. Despite the robust evidence for IHs and this interesting coincidence with the phase change, relatively little is known about the diversity and emergence dynamics of IHs. Why is there more than one IH, and how are they dependent on each other? Why do IHs appear all of a sudden, and what are the subcircuits that enable them to emerge? We answer these questions by studying IH emergence dynamics in a controlled setting by training on synthetic data. In doing so, we develop and share a novel optogenetics-inspired causal framework for modifying activations throughout training. Using this framework, we delineate the diverse and additive nature of IHs. By "clamping" subsets of activations throughout training, we then identify three underlying subcircuits that interact to drive IH formation, yielding the phase change. Furthermore, these subcircuits shed light on data-dependent properties of formation, such as phase change timing, already showing the promise of this more in-depth understanding of subcircuits that need to "go right" for an induction head.

Cite this Paper

BibTeX
@InProceedings{pmlr-v235-singh24c, title = {What needs to go right for an induction head? {A} mechanistic study of in-context learning circuits and their formation}, author = {Singh, Aaditya K and Moskovitz, Ted and Hill, Felix and Chan, Stephanie C.Y. and Saxe, Andrew M}, booktitle = {Proceedings of the 41st International Conference on Machine Learning}, pages = {45637--45662}, year = {2024}, editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix}, volume = {235}, series = {Proceedings of Machine Learning Research}, month = {21--27 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/singh24c/singh24c.pdf}, url = {https://proceedings.mlr.press/v235/singh24c.html}, abstract = {In-context learning is a powerful emergent ability in transformer models. Prior work in mechanistic interpretability has identified a circuit element that may be critical for in-context learning – the induction head (IH), which performs a match-and-copy operation. During training of large transformers on natural language data, IHs emerge around the same time as a notable phase change in the loss. Despite the robust evidence for IHs and this interesting coincidence with the phase change, relatively little is known about the diversity and emergence dynamics of IHs. Why is there more than one IH, and how are they dependent on each other? Why do IHs appear all of a sudden, and what are the subcircuits that enable them to emerge? We answer these questions by studying IH emergence dynamics in a controlled setting by training on synthetic data. In doing so, we develop and share a novel optogenetics-inspired causal framework for modifying activations throughout training. Using this framework, we delineate the diverse and additive nature of IHs. By "clamping" subsets of activations throughout training, we then identify three underlying subcircuits that interact to drive IH formation, yielding the phase change. Furthermore, these subcircuits shed light on data-dependent properties of formation, such as phase change timing, already showing the promise of this more in-depth understanding of subcircuits that need to "go right" for an induction head.} }
Endnote
%0 Conference Paper %T What needs to go right for an induction head? A mechanistic study of in-context learning circuits and their formation %A Aaditya K Singh %A Ted Moskovitz %A Felix Hill %A Stephanie C.Y. Chan %A Andrew M Saxe %B Proceedings of the 41st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Ruslan Salakhutdinov %E Zico Kolter %E Katherine Heller %E Adrian Weller %E Nuria Oliver %E Jonathan Scarlett %E Felix Berkenkamp %F pmlr-v235-singh24c %I PMLR %P 45637--45662 %U https://proceedings.mlr.press/v235/singh24c.html %V 235 %X In-context learning is a powerful emergent ability in transformer models. Prior work in mechanistic interpretability has identified a circuit element that may be critical for in-context learning – the induction head (IH), which performs a match-and-copy operation. During training of large transformers on natural language data, IHs emerge around the same time as a notable phase change in the loss. Despite the robust evidence for IHs and this interesting coincidence with the phase change, relatively little is known about the diversity and emergence dynamics of IHs. Why is there more than one IH, and how are they dependent on each other? Why do IHs appear all of a sudden, and what are the subcircuits that enable them to emerge? We answer these questions by studying IH emergence dynamics in a controlled setting by training on synthetic data. In doing so, we develop and share a novel optogenetics-inspired causal framework for modifying activations throughout training. Using this framework, we delineate the diverse and additive nature of IHs. By "clamping" subsets of activations throughout training, we then identify three underlying subcircuits that interact to drive IH formation, yielding the phase change. Furthermore, these subcircuits shed light on data-dependent properties of formation, such as phase change timing, already showing the promise of this more in-depth understanding of subcircuits that need to "go right" for an induction head.
APA
Singh, A.K., Moskovitz, T., Hill, F., Chan, S.C. & Saxe, A.M.. (2024). What needs to go right for an induction head? A mechanistic study of in-context learning circuits and their formation. Proceedings of the 41st International Conference on Machine Learning, in Proceedings of Machine Learning Research 235:45637-45662 Available from https://proceedings.mlr.press/v235/singh24c.html.

Related Material