The paper introduces SelfMorph, a method that lets frozen large language models adapt their internal computation topology at inference time based on the input. It combines three morphing strategies—BASE, REPEAT, and BRANCH—into a unified space and uses a lightweight selector trained via distillation to choose the best morphology for each prompt. Experiments on 4B–27B models show that SelfMorph improves performance by an average of 8.06% on in‑distribution benchmarks and 4.45% on unseen benchmarks, outperforming other settings in most cases.
arXiv:2610.23455v1 Announce Type: cross Abstract: LLMs and LVLMs are increasingly deployed across heterogeneous hardware, ranging from data-center GPUs to resource-constrained edge devices. Their fix...
NeuronSpark is a self‑attention‑free language model that uses input‑dependent spiking membrane dynamics, allowing each neuron’s decay rate, gain, and firing threshold to adapt to the current token. This selective spiking approach enables billion‑parameter pretraining from random initialization without relying on backbone transfer, conversion, or distillation, achieving competitive zero‑shot performance and a fast–slow hierarchy. On a 1.1B‑parameter, 2.1B‑token training run, NeuronSpark attains a 40.0 eight‑task zero‑shot average, surpasses comparable‑budget Pythia/Mamba checkpoints, and demonstrates significant energy and cycle reductions on FPGA implementations.
The paper introduces recirculation, an inference‑time architectural enhancement for foundation models that reduces perplexity and improves accuracy on generation and reasoning tasks without adding significant latency. Recirculation adds a specific form of recurrence, enabling the model to function as a dynamical system that tracks belief states, and is distinct from chain‑of‑thought or depth‑recurrence methods. An adaptive variant requires minimal hyperparameter tuning and achieves notable gains on the Gemma3 family, including a 23% perplexity drop and a 21% accuracy increase on GSM8k.
The paper introduces EvoResearcher, a training‑free, inference‑time protocol that enables a frozen large language model to perform cost‑bounded self‑reflection and early stopping. By iterating through generate → self‑critique → revise steps until a maximum depth or a CONFIRMED sentinel is reached, the model can self‑verify its answers within a strict compute budget. The protocol incorporates four self‑reflective meta‑reward components—correctness, efficiency, reflection depth, and tool‑call diversity—implemented as prompt‑level mechanisms, and is validated on Big‑Bench Hard, GSM8K, and MATH benchmarks, achieving comparable accuracy while terminating 82‑88% of items early with only about 2.1 generations per question.
The paper introduces Untied Self-Conditioning, a sampler that corrects a train–inference mismatch in flow‑matching language models. By dampening redundant directions in the self‑conditioning input and approximating a step‑average prediction from history, the method improves generation quality without retraining. On LangFlow and ELF‑B datasets, it dramatically lowers perplexity and is preferred in the majority of pairwise comparisons.
The paper introduces CHASE, a cache‑hole‑adapted skip‑exit mechanism for looped state‑space language models, specifically Looped Mamba and Looped Hybrid Mamba‑Transformer. It shows that looping these architectures improves performance on controlled reasoning tasks and remains competitive in pre‑training benchmarks while using fewer distinct parameters. The cache‑hole adaptation allows selective skipping of recurrent steps during inference, maintaining perplexity close to full computation and achieving significant speedups.
LEV is a decision model that performs reasoning by looping its layers instead of generating text, allowing it to deepen its computation on a per-question basis. Trained jointly across all loop steps, a single LEV checkpoint can answer at any depth, improving accuracy—especially on hard questions—while keeping inference fast. On benchmark tests, LEV outperforms dense and cascade models by 5.4 to 8.8 accuracy points with sub‑50 ms latency, and on‑demand depth selection saves 15–24% of computation for minimal accuracy loss.
arXiv:2610.16301v1 Announce Type: new Abstract: Self-attention assigns two functions to one mechanism-where to read and how to mix-at the price of O(L^2) training, an O(T) decode state, and position...
Scaling Large Language Models (LLMs) has been driven mainly by enlarging the Transformer backbone, but for an already-strong model this requires another round of costly pretraining. We study whether an existing backbone can keep improving by allocating more computation to each token while leaving the Transformer backbone fixed.
arXiv:2608. 01651v1 Announce Type: cross Abstract: Hybrid-attention large language models combine full attention with recurrent linear attention to reduce long-context inference costs, yet their autoregressive decoding remains memory-bound.
arXiv:2610.19210v1 Announce Type: new Abstract: Transformers are efficient to pre-train at scale because attention permits parallel computation across sequence positions, but long-context inference r...
FlexEE is an early‑exiting framework designed for large language model inference that is constrained by computation and memory, particularly in offloading‑based deployments. It uses layer‑wise exit supervision, self‑speculative decoding over a Top‑K local vocabulary, and dynamic hidden‑state management to enable reliable intermediate‑layer predictions and memory‑aware execution. Experiments on Llama2‑7B and Llama3‑8B show that FlexEE achieves significant speedups—up to 1.27×/3.16× and 1.25×/2.83× respectively—while maintaining minimal accuracy loss.