arXiv:2608. 14691v1 Announce Type: new Abstract: Sequence models are conventionally distinguished by their backbone, the mechanism that routes information across positions, such as attention or recurrence.
arXiv:2511.00763v3 Announce Type: replace Abstract: We investigate the performance of large language models (LLMs) on repetitive deterministic prediction tasks and study how the sequence accuracy rat...
arXiv:2608. 11173v1 Announce Type: cross Abstract: The attention mechanism forms the foundation of many modern AI models such as the Transformer.
The paper demonstrates that quantum transformer blocks can be intrinsically interpretable by tracking quantum mutual information, entanglement entropy, and state fidelity across layers. Experiments on four synthetic tasks show that learned mutual information aligns with task structure, entanglement is essential for accuracy, and mutual information predicts prediction correctness. These findings are validated on IBM Quantum hardware, illustrating that quantum computation’s physics can provide observable interpretability signals.
Fast Weight Attention for Continual Learning introduces recurrent fast‑weight memories and selective state‑space models that compress expanding context into a fixed‑size recurrent state, enabling an online learning rule for state transitions. The paper derives normalized first‑order updates for squared‑error regression and negative inner‑product objectives, presenting several variants (Falcon‑1, Falcon‑2, Falcon‑3 and their inner‑product counterparts) with recurrent, masked‑parallel, and chunk‑parallel implementations. These methods demonstrate competitive performance in language modeling and improved length extrapolation on variable‑digit addition tasks.
arXiv:2606. 00926v1 Announce Type: new Abstract: Mechanistic studies of sequence models often treat layerwise state encodings as architectural traits: recurrent models concentrate readable state, attention-based models distribute it.
arXiv:2602. 06699v2 Announce Type: replace-cross Abstract: We propose a variational quantum implementation of self-attention (QSA)-the core operation in transformers and large language models-which predicts future elements of a sequence by forming overlap-weighted combinations of past data.
arXiv:2607. 19390v1 Announce Type: new Abstract: A recent report finds that orthogonalizing the mLSTM memory matrix at read time (five Newton-Schulz iterations, trained through) substantially improves noisy associative recall.
The paper investigates whether extra computation in recurrent Transformers should be allocated to more temporal steps or greater physical depth. Using Latent Recurrent Transformers (LRTs), the authors insert a latent thought token between vocabulary tokens, allowing each token to pass through the same $L$ layers twice while sharing parameters. Experiments on 16‑ and 20‑layer mixture‑of‑experts NanoChat backbones show that a single thought token brings a shallower model within 0.006–0.004 bits per byte of a double‑depth counterpart, recovering 67–81% of the improvement with roughly 48% fewer parameters.
The paper investigates the training dynamics of attention mechanisms in high-dimensional settings, focusing on attention-indexed models that encompass multi-layer and multi-head architectures. It shows that while the loss landscape can be described by a finite set of trace order parameters, the online stochastic gradient descent dynamics involve an infinite hierarchy of matrix moments that can be accurately approximated by a finite truncated system. The study further reveals that the choice of attention parameterization acts as an implicit bias: untied attention can get trapped in uninformative states, whereas tied attention induces symmetry breaking and enables weak recovery with θ(d² log d) samples, and untied attention exhibits a fast-slow dynamic leading to weak recovery when symmetry is broken.
The paper introduces Declarative Attention (DA), a protocol that lets language models explicitly declare which parts of their context to focus on during generation. By partitioning decoding into full-context, region-specific, and recent-output-only modes, the inference engine can skip large portions of the KV cache, dramatically reducing attended tokens. Experiments on 15 long-context tasks with off-the-shelf models show significant savings (52.0% and 31.1% reductions) with only modest accuracy drops that diminish as model size increases.
arXiv:2609.05842v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards enables large language models to think slowly, but the same training can induce policy collapse: proba...
The paper investigates how block‑diffusion language models can use a constant‑size cache to enable efficient parallel decoding. By employing sequence mixers that summarize completed blocks into a reusable state and a block‑causal training objective, the authors pretrain three 3B block‑diffusion denoisers (attention, Mamba, and hybrid) on 300 B tokens. The resulting state‑space cache remains O(1) in memory and latency regardless of context length, yielding significant speed‑up and memory savings compared to traditional attention‑based caches, especially at very long sequences.