arXiv AI
Before Attention Sinks Form: Hidden-State Geometry Reveals and Modulates Their Formation
Read the original on arXiv AI →The Flow has not summarised this story yet — read it at arXiv AI.
Related stories
arXiv AI
Sep 10
Do New Attention Mechanisms Actually Fix Attention Sinks at Million-Token Context?
The paper investigates whether recent attention‑mechanism improvements—specifically gated attention, Kimi K3, Kimi Delta Attention, and Attention Residuals—effectively eliminate the attention‑sink problem when scaling language models to a one‑million‑token context window. Using a new diagnostic suite called SinkProbe, the authors evaluate sink mass, massive activation, position‑resolved recall, and the recency gap across four small models that vary only in token mixing and depth. Their findings show that the training objective, rather than the architecture, drives the emergence of attention sinks; gating did not replicate its previously reported benefits at the larger scale, and sink mass, activations, and positional bias behaved independently.
arXiv Computation and Language
Aug 25
Massive Activations in Hybrid Linear Attention Large Language Models: Pre-Attention Spikes and Inter-Spike Plateaus
arXiv:2608.12149v2 Announce Type: replace Abstract: We present the first systematic study of Massive activations (MAs) in layer-interleaved HLA LLMs and uncover two architecture-aligned morphologies:...
arXiv AI
Oct 7
When Attention Closes: How LLMs Lose the Thread in Multi-Turn Interaction
arXiv:2605.12922v2 Announce Type: replace Abstract: Large language models can follow complex instructions in a single turn, yet over long multi-turn interactions they often lose the thread of instruc...
arXiv Machine Learning
Jun 9
A Unifying View of Attention Sinks: Two Algorithms, Two Solutions
arXiv:2606. 08105v1 Announce Type: new Abstract: When attention concentrates on a single token, a sink, what is the model actually computing?
arXiv AI
Oct 16
Learning to Read Out: Unembedding Dynamics in Language Model Pretraining
The paper investigates when a language model actually learns task information versus when its readout (unembedding matrix) learns to express that information. Using a new parameter‑trajectory cross‑coding method, the authors find that hidden states can encode task data early in training, but the readout reorganizes only in a narrow window (0.4–1.1% of Pythia’s pretraining). By varying the readout learning rate, they show that faster readout learning reduces the gap between hidden‑state encoding and output expression without affecting final performance.
arXiv Machine Learning
Aug 4
Understanding Sparse Attention Selectivity in Long-Context Foundation Models via Counterfactual Evaluation
arXiv:2608. 01676v1 Announce Type: cross Abstract: Sparse attention is widely deployed in long-context serving stacks, yet no framework audits how discarding blocks changes the influence of specific content on model output.