The paper investigates how nonlinearity in attention mechanisms affects scaling laws in large language models. It finds that nonlinear attention leads to an inverse-depth decay of loss across various data spectra, enabling models to learn both strong and weak spectral directions simultaneously. This behavior contrasts with linear-attention models, which rely on a power-law data spectrum and learn directions sequentially.
By Zirui Peng, Yizhou Liu, Ziming Liu, Jeff Gore
arXiv:2606. 07473v1 Announce Type: cross Abstract: Whisper, a widely adopted ASR model, is known to suffer from hallucinations - coherent transcriptions generated for non-speech audio entirely disconnected from the input.
By Georgii Aparin, Vadim Popov, Tasnima Sadekova, Assel Yermekova
arXiv:2607. 09753v1 Announce Type: cross Abstract: Diffusion models have achieved remarkable success across diverse domains, with performance closely related to the denoising backbones that parameterize the score function.
By Haksoo Lim, Myeongjin Lee, Wonjoon Chang, Jaesik Choi
arXiv:2610.16745v1 Announce Type: cross
Abstract: Attention sinks are usually studied after they have formed. We ask whether a measurable hidden-state change appears beforehand and whether changing t...
By Fan Yang, Yue Song
arXiv:2608. 09417v2 Announce Type: replace Abstract: Deep decoder-only Transformers often replace the original Post-Norm architecture with Pre-Norm variants because Post-Norm training is highly sensitive to warmup and learning rate under conventional initialization schemes.
By Xingjian Wang, Qingyu Han, Xiaodong Luo, Yin Zhang
arXiv:2510. 05554v2 Announce Type: replace Abstract: As large language models scale to longer contexts, attention layers suffer from a fundamental pathology: attention scores collapse toward uniformity as context length $n$ increases, causing tokens to cluster excessively, a phenomenon known as rank-collapse.
By Shi Chen, Zhengjiang Lin, Yury Polyanskiy, Philippe Rigollet
arXiv:2608. 09417v1 Announce Type: new Abstract: Deep decoder-only Transformers often replace the original Post-Norm architecture with Pre-Norm variants because Post-Norm training is highly sensitive to warmup and learning rate under conventional initialization schemes.
By Xingjian Wang, Qingyu Han, Xiaodong Luo, Yin Zhang
arXiv:2610.20439v1 Announce Type: new
Abstract: We study classification dynamics under small initialization and identify a pronounced non-monotone behavior of accuracy that we call the accuracy well:...
By Yaroslav Gusev, Eugene Golikov, Dmitry Yarotsky
arXiv:2606. 17120v1 Announce Type: new Abstract: Deep neural networks (DNNs) exhibit first order phase transitions under variations of the L2 regularization strength, with each transition marking the onset of a new learnable feature.
By Ibrahim Talha Ersoy, Karoline Wiesner
The paper investigates why automatic speech recognition (ASR) systems sometimes generate fluent text that does not correspond to the input audio, a phenomenon termed hallucination. By examining two independently trained Conformer‑Large models—one using CTC and the other RNN‑T—under conditions of environmental noise and speaker‑background shift, the authors identify the final encoder stage as a critical boundary. Bypassing this final block leads to divergence on almost all utterances, while bypassing earlier blocks has minimal effect; at this stage, representations become compact, the decoder can read the text, and grapheme information becomes explicit, yet the output is garbled or repetitive rather than fluent fabrication. The study thus pinpoints a mechanistic precondition for hallucination—failure to produce adequately grounded output—though it does not fully explain naturally occurring hallucinations, and it highlights a consistent terminal‑stage dependency across decoder families and distribution shifts.
By Hamees Sayed, Apoorv Singh, Kumar Aman, Akshat Mandloi
We investigate how each component of the Transformer feedforward block architecture design determines how much rank survives across depth at initialization. We reinterpret skip connections and normalization, long understood as controlling magnitude, as mechanisms for preserving gradient rank across depth, since the very matrix multiplications and nonlinear activations that make the network expressive also reduce the rank.
arXiv:2607. 14018v1 Announce Type: cross Abstract: We investigate how each component of the Transformer feedforward block architecture design determines how much rank survives across depth at initialization.
By Katie Everett