# No Clustering, No Routing: How Transformers Actually Process Rare Tokens

arXiv: https://arxiv.org/abs/2509.04479 · HTML: https://mlsanity.com/paper/2509.04479/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 7 of its 12 references that are in the map.
- First posted on arXiv 30 Aug 2025.

## Builds on (most cited first, 7 of 7)

- [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://mlsanity.com/paper/1910.10683/) (2019)
- [Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer](https://mlsanity.com/paper/1701.06538/) (2017)
- [Transformer Feed-Forward Layers Are Key-Value Memories](https://mlsanity.com/paper/2012.14913/) (2020)
- [Large Language Models Struggle to Learn Long-Tail Knowledge](https://mlsanity.com/paper/2211.08411/) (2022)
- [The Curse of Recursion: Training on Generated Data Makes Models Forget](https://mlsanity.com/paper/2305.17493/) (2023)
- [Towards Tracing Factual Knowledge in Language Models Back to the Training Data](https://mlsanity.com/paper/2205.11482/) (2022)
- [Emergent Specialization: Rare Token Neurons in Language Models](https://mlsanity.com/paper/2505.12822/) (2025)

## Built on by (most cited first, 1 of 1)

- [Circuit-Aware Reward Training: A Mechanistic Framework for Longtail Robustness in RLHF](https://mlsanity.com/paper/2509.24713/) (2025)

## Abstract (arXiv)

Large language models struggle with rare token prediction, yet the mechanisms driving their specialization remain unclear. Prior work identified specialized ``plateau'' neurons for rare tokens following distinctive three-regime influence patterns citeliu2025emergent, but their functional organization is unknown. We investigate this through neuron influence analyses, graph-based clustering, and attention head ablations in GPT-2 XL and Pythia models. Our findings show that: (1) rare token processing requires additional plateau neurons beyond the power-law regime sufficient for common tokens, forming dual computational regimes; (2) plateau neurons are spatially distributed rather than forming modular clusters; and (3) attention mechanisms exhibit no preferential routing to specialists. These results demonstrate that rare token specialization arises through distributed, training-driven differentiation rather than architectural modularity, preserving context-sensitive flexibility while achieving adaptive capacity allocation.