On the Generalizability of "Competition of Mechanisms: Tracing How Language Models Handle Facts and Counterfactuals"
This is an incremental study that reproduces and extends prior work on mechanistic interpretability in language models, highlighting generalizability issues across models, prompts, and domains.
This paper reproduces a study on how language models handle facts and counterfactuals, confirming findings about attention mechanisms in GPT-2 and Pythia 6.9B, and extends it by showing reduced attention head specialization in Llama 3.1 8B, decreased counterfactual logits with prompt variations, and domain-specific limitations in the original method's effectiveness.
We present a reproduction study of "Competition of Mechanisms: Tracing How Language Models Handle Facts and Counterfactuals" (Ortu et al., 2024), which investigates competition of mechanisms in language models between factual recall and counterfactual in-context repetition. Our study successfully reproduces their primary findings regarding the localization of factual and counterfactual information, the dominance of attention blocks in mechanism competition, and the specialization of attention heads in handling competing information. We reproduce their results on both GPT-2 (Radford et al., 2019) and Pythia 6.9B (Biderman et al., 2023). We extend their work in three significant directions. First, we explore the generalizability of these findings to even larger models by replicating the experiments on Llama 3.1 8B (Grattafiori et al., 2024), discovering greatly reduced attention head specialization. Second, we investigate the impact of prompt structure by introducing variations where we avoid repeating the counterfactual statement verbatim or we change the premise word, observing a marked decrease in the logit for the counterfactual token. Finally, we test the validity of the authors' claims for prompts of specific domains, discovering that certain categories of prompts skew the results by providing the factual prediction token as part of the subject of the sentence. Overall, we find that the attention head ablation proposed in Ortu et al. (2024) is ineffective for domains that are underrepresented in their dataset, and that the effectiveness varies based on model architecture, prompt structure, domain and task.