Language Modeling

Ask about this topic

Overview

Language modeling studies how systems learn from sequences, most often by predicting the next token or reconstructing masked tokens, enabling text generation and language understanding. Common approaches include autoregressive Transformers, masked language models, and alternatives such as recurrent or state-space networks, mixture-of-experts models, and diffusion methods. Researchers improve models through data and tokenizer design, continued pretraining, instruction tuning, retrieval, distillation, and compression, evaluating them with perplexity, linguistic tests, and task benchmarks. A central tension is how to make models more efficient and reliable while ensuring benchmark success reflects robust reasoning, generalization, and competence across languages and domains.

Latest papers

CardsList

525 papers73 papers in the last four weeks, up 22% on the four weeks before. 0.5% of all new papers.

May 13, 2026

Sampling from Flow Language Models via Marginal-Conditioned Bridges

Flow Language Models (FLMs) are a recently introduced class of language models which adapt continuous flow matching for one-hot encoded token sequences. Their denoisers have a special structure absent from generic continuous diffusion models: each block of the denoising mean is a posterior marginal distribution over the clean token at that position. Standard DDPM-style samplers collapse these marginals to a single conditional-mean endpoint and bridge toward this simplex-valued point, which is generally not a valid one-hot sequence. We argue that the natural sampler for an FLM is instead posterior-predictive. At each reverse step, we sample a clean one-hot endpoint from the factorized posterior defined by the FLM token marginals, and then sample the next continuous state from the analytic Ornstein--Uhlenbeck bridge conditioned on that endpoint. The method is training-free, uses the same model evaluations as standard sampling, and gives a principled interface for token-level decoding controls such as temperature scaling and nucleus truncation. We show that, under exact posterior marginals, the endpoint approximation error is exactly the conditional multi-information among token positions. The induced one-step bridge kernel preserves all token-wise posterior-predictive marginals and loses only the residual cross-position dependence. Finally, we prove a Girsanov path-space comparison showing that the marginal-conditioned bridge has a no-larger denoising-error term than the frozen conditional-mean bridge, with strict improvement whenever intermediate coordinate-wise bridge observations reveal additional information about the clean token. Experiments with FLMs show that the sampler improves the quality--diversity tradeoff. Code is available at: github.com/imbirik/mcb.

Iskander Azangulov, Leo Zhang

Department of Statistics, University of Oxford

Flow MatchingLanguage Model DecodingLanguage ModelingFlow-Based Generative Modeling
AskPDF
May 13, 2026

Effective Context in Transformers: An Analysis of Fragmentation and Tokenization

Transformers predict over a representation of a sequence. The same data can be written as bytes, characters, or subword tokens, and these representations may be lossless. Yet, under a fixed context window, they need not expose the same information to the model. This raises a basic question: how does the choice of representation change what a finite-context predictor can achieve? We study this question on Markov sources and uncover two complementary phenomena. First, we observe that moving to smaller representation units can hurt prediction even when the context window is enlarged to cover the relevant source history. To explain this, we introduce fragmentation: a lossless recoding that replaces each source symbol by several smaller units. We prove that fragmentation can strictly increase the optimal finite-context log-loss, showing that the gap is not merely an optimization or capacity issue, but can be intrinsic to the representation. This gives a theoretical account of the finite-context gap observed in byte- and character-level models such as ByT5 and CANINE relative to subword-tokenized models. Second, we study the opposite direction: greedy tokenization -- BPE, WordPiece, and related methods -- which groups source symbols into larger units. We show that tokenization can make a short token window behave like a longer source-context window, and we give a loss guarantee describing when this is achievable. The guarantee depends on how reliably token windows span the needed source history, together with the compression rate of the tokenizer. This also yields a simple diagnostic for real tokenizers: measuring how much source context a fixed token window reliably contains. Together, the two directions establish a finite-context information-theoretic framework for reasoning about representation choices in Transformers.

Amirmehdi Jafari Fesharaki, Mohammadamin Rami, Aslan Tchamkerten

Department of Communications and Electronics Institut Polytechnique de Paris

Language ModelingSubword TokenizationTokenization
AskPDF
May 13, 2026

N-vium: Mixture-of-Exits Transformer for Accelerated Exact Generation

Improving the inference efficiency of autoregressive transformers typically means reducing FLOPs per token, usually through approximations that degrade model quality. We introduce N-vium, a mixture-of-exits transformer that partially parallelizes computation across depth on standard hardware, increasing effective FLOPs per second rather than minimizing compute per token. N-vium attaches prediction heads at multiple depths and defines the next-token distribution as a learned mixture over these exits, with token-adaptive routing. This formulation strictly generalizes the standard transformer, which is recovered exactly when routing assigns zero mass to all intermediate heads. Sampling from the mixture is exact, and complete KV caches are recovered by deferring the upper-layer computation and batching it with later tokens. We pretrain N-vium at scales up to 1.5B parameters. Our largest model reaches 57.9% wall-clock speedup over a parameter- and data-matched standard transformer at no perplexity cost.

Aleksander Lorenc, Frédéric Berdoz, Joël Mathys +1

ETH Zurich

Efficient Transformer InferenceLLM Inference AccelerationLanguage Modeling
AskPDF
May 13, 2026

The Growing Pains of Frontier Models: When Leaderboards Stop Separating and What to Measure Next

Leaderboards rank frontier models on independent axes but do not reveal whether capabilities reinforce or trade off across releases -- and at the frontier, this interaction is the more informative signal. We decompose paired SWE-bench and GPQA Diamond scores into a population coupling trend and per-release residual (hh-field) that diagnoses capability emphasis from two public benchmark scores. Across 34 models from 10 labs (2024--2026), capabilities cooperate (r=+0.72r = +0.72, p<10−6p < 10^{-6}), but cooperation varies systematically: per-lab coupling slopes span 5×5\times (Google 1.151.15 vs. DeepSeek 0.230.23), and labs pivot -- DeepSeek reversed from reasoning-rich to coding-first (Δh=15.9Δh = 15.9~pp); Anthropic oscillates between coding excursions and recovery. The population regression serves as an isocline phase boundary: the same (a/b)⋅B1\sqrt{(a/b)\cdot B_1} classifier that identifies the base-scale coupling transition [Amin, 2026] classifies frontier models and already detects mixed-phase behavior at the next transition (two models below the GPQA--IFEval isocline). The hh-field is not just diagnostic -- it tells you what to change. Pretraining establishes coupling at 0.8710.871 while RLHF adds 0.0810.081 [Amin, 2026]: pretraining-level shifts are permanent (DeepSeek's four-release reversal persists), post-training shifts are reversible (Anthropic's three coding excursions each recover within one release), and inference compute alone shifts hh by +7.8+7.8~pp without retraining. Knowing which component dominates determines whether to retrain or wait. We provide a three-step diagnostic (locate, classify, predict), a per-lab measurement-priority table, and seven falsifiable predictions with timestamped criteria. Five post-cutoff releases fall within the 95% prediction interval. Code, data, and an interactive dashboard: https://zehenlabs.com/cape/.

Adil Amin

ZEHEN Labs

LLM EvaluationForecasting BenchmarksLanguage Modeling
AskPDF
May 13, 2026

The Efficiency Gap in Byte Modeling

Modern language models have historically relied on two dominant design choices: subword tokenization and autoregressive (AR) ordering. These design decisions bake in priors that dictate a model's learning. Recently, two alternative paradigms have challenged this: byte-level modeling, which bypasses static statistically-derived token vocabularies, and masked diffusion modeling (MDM), which conducts parallel, non-sequential generation. Their intersection represents a fully end-to-end modality-agnostic generative prototype; however, removing these structural priors incurs a significant computational cost. In this work, we investigate this cost through a compute-matched scaling study. Our results reveal that the performance penalty of byte modeling is not uniform; across scale, the scaling overhead of byte modeling is worse for MDM than for AR. We hypothesize that this disparity stems from context fragility: while AR's stable causal history allows models to naturally rediscover subword patterns, the MDM objective destroys the local contiguity required to efficiently resolve semantics from raw bytes. Our findings from controlled permutation experiments suggest that future modality-agnostic designs must incorporate alternative structural biases to maintain viable scaling trajectories in the byte regime.

Celine Lee, Jing Nathan Yan, Chen Liang +9

1Google DeepMind · 2Work done while at Google DeepMind · Department of Computer Science, Cornell University.

Language Model Scaling LawsAutoregressive Language ModelingLanguage ModelingDiffusion Language Models
AskPDF
May 12, 2026

In-Context Learning Operates as Concept Subspace Learning

Regression and Bayesian accounts of in-context learning (ICL) explain how demonstrations can induce predictors, while mechanistic analyses often identify compact activation directions that steer prompted behavior. However, it remains unclear whether structured demonstrations induce low-dimensional concept inference. We study this question through a concept-subspace view of ICL, in which tasks vary only along intrinsic concept coordinates, although inputs are observed in a high-dimensional ambient space. For ridge and least-squares ICL proxies, prediction decomposes exactly into concept-coordinate regression and off-subspace leakage. Under block-diagonal or near-block-diagonal covariance assumptions, the leading estimation and nuisance-sensitivity terms scale with the dimension of the concept subspace, while residual effects are controlled by cross-subspace coupling. This separation gives a mechanistic prediction: recoverable task information should concentrate in a low-dimensional, task-aligned activation subspace. On CounterFact-derived multi-relation prompts with Llama-3-8B, a 68--73-dimensional subspace of the 4096-dimensional residual stream restores 78.8% of the clean--corrupted accuracy gap, whereas patching the complementary subspace restores 0%. Concept swaps redirect predictions toward injected relations, while random and cross-task matched-rank controls are largely ineffective. Additional experiments on Qwen2.5-7B and a controlled cross-lingual rule task show the same qualitative pattern. These results support concept subspaces as compact, task-aligned mediators of recoverable ICL behavior in structured task families, without implying full-circuit recovery.

Wei Tang, Xinyan Jiang, Fakhri Karray +1

Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) · Shanghai Advanced Research Institute, Chinese Academy of Science

Representation LearningMechanistic InterpretabilityIn-Context LearningLanguage Modeling
AskPDF
May 12, 2026

WriteSAE: Sparse Autoencoders for Recurrent State

We introduce WriteSAE, a sparse autoencoder for the matrix updates written into recurrent language-model state. In Gated DeltaNet, Mamba-2, and RWKV-7, each token writes a matrix-shaped update to a recurrent cache; a residual-stream SAE has vector-shaped atoms and cannot replace that update directly. WriteSAE learns rank-1 matrix atoms with the same shape as the model's own write. This lets us test a direct replacement: at positions where the SAE activates an atom, we remove the model's write, insert the atom scaled by its SAE activation, and continue the forward pass. The atom gives a closer final token distribution than deleting the write on 92.4% of evaluated positions; averaged per atom, the rate is 89.8%. For Gated DeltaNet, a formula using the forget gate, read query, and output embedding predicts the resulting logit change with R2=0.98R^2 = 0.98. The same replacement test transfers to Mamba-2-370M at 88.1%. In generation, the formula chooses a write direction; writing it into three consecutive cache positions at 3×3\times the norm of the model's write makes tokens initially ranked 100--1000 by the unmodified model appear in 100% of continuations, up from 33.3%. To our knowledge this is the first cache-level steering intervention reported in a state-space or hybrid recurrent layer.

Jack Young

Indiana University

Recurrent Language ModelsLanguage Model SteeringSparse AutoencodersLanguage ModelingState Representation LearningCausal Interventions in Language Models
AskPDF
May 12, 2026

A Causal Language Modeling Detour Improves Encoder Continued Pretraining

When adapting an encoder to a new domain, the standard approach is to continue training with Masked Language Modeling (MLM). We show that temporarily switching to Causal Language Modeling (CLM) followed by a short MLM decay improves downstream performance. On biomedical texts with ModernBERT, this CLM detour outperforms MLM baselines trained on identical data and compute across 8 French and 11 English biomedical tasks, by +1.2-2.8pp and +0.3-0.8pp respectively, depending on model size. We investigate the reasons for these gains. We find that CLM's dense supervision impacts low transformer layers (0-7) far more than MLM does. Freezing low layers during CLM eliminates the downstream benefit; freezing mid layers preserves it. The representational changes persist through the MLM decay phase, even when it matches the CLM phase in length, and they scale with model capacity. We release ModernCamemBERT-bio and ModernBERT-bio as state-of-the-art biomedical encoders in Base and Large sizes.

Rian Touchent, Eric de la Clergerie

Sorbonne Université / INRIA Paris · ALMAnaCH Team · INRIA Paris

Language Model PretrainingMasked Language ModelingAutoregressive Language ModelingLanguage ModelingContinued Pretraining
AskPDF
May 12, 2026

A Unified Graph Language Model for Multi-Domain Multi-Task Graph Alignment Instruction Tuning

Leveraging Graph Neural Networks (GNNs) as graph encoders and aligning the resulting representations with Large Language Models (LLMs) through alignment instruction tuning has become a mainstream paradigm for constructing Graph Language Models (GLMs), combining the generalization ability of LLMs with the structural modeling capacity of GNNs. However, existing GLMs that adopt GNNs as graph encoders largely overlook the problem of aligning GNN-encoded representations across domains and tasks with the LLM token space to obtain unified graph tokens, thereby limiting their ability to generalize across diverse graph data. To bridge this gap, we aim to incorporate a multi-domain, multi-task GNN encoder into GLMs and align its representations with LLMs to enable multi-domain, multi-task graph alignment instruction tuning. This alignment problem remains underexplored and poses two key challenges: 1) learning GNN-encoded representations that are simultaneously generalizable across domains and tasks and well aligned with textual semantics is difficult, due to substantial variations in graph structures, feature distributions, and supervision signals, together with the lack of textual-semantic alignment guidance in task-specific GNN training; 2) diverse graph data and task-specific instructions can exhibit different degrees of compatibility with the LLM token space during instruction tuning, leading to varying alignment difficulty and rendering a fixed alignment strategy suboptimal. To tackle these challenges, we propose UniGraphLM, a Unified Graph Language Model that incorporates a multi-domain, multi-task GNN encoder to learn generalizable graph representations aligned with textual semantics, and then adaptively aligns these representations with the LLM.

Haibo Chen, Xin Wang, Jiaheng Chao +2

Tsinghua University

LLM AlignmentGraph Representation LearningLanguage ModelingInstruction Tuning
AskPDF
May 12, 2026

To Whom Do Language Models Align? Measuring Principal Hierarchies Under High-Stakes Competing Demands

Language models deployed in high-stakes professional settings face conflicting demands from users, institutional authorities, and professional norms. How models act when these demands conflict reveals a principal hierarchy -- an implicit ordering over competing stakeholders that determines, for instance, whether a medical AI receiving a cost-reduction directive from a hospital administrator complies at the expense of evidence-based care, or refuses because professional standards require it. Across 7,136 scenarios in legal and medical domains, we test ten frontier models and find that models frequently fail to adhere to professional standards during task execution, such as drafting, when user instructions conflict with those standards -- despite adequately upholding them when users seek advisory guidance. We further find that the hierarchies between user, authority, and professional standards exhibited by these models are unstable across medical and legal contexts and inconsistent across model families. When failing to follow professional standards, the primary failure mechanism is knowledge omission: models that demonstrably possess relevant knowledge produce harmful outputs without surfacing conflicting knowledge. In a particularly troubling instance, we find that a reasoning model recognizes the relevant knowledge in its reasoning trace -- e.g., that a drug has been withdrawn -- yet suppresses this in the user-facing answer and proceeds to recommend the drug under authority pressure anyway. Inconsistent alignment across task framing, domain, and model families suggests that current alignment methods, including published alignment hierarchies, are unlikely to be robust when models are deployed in high-stakes professional settings.

Fangyi Yu, Nabeel Seedat, Jonathan Richard Schwarz +1

Thomson Reuters Foundational Research, 5 Canada Square, London · Imperial College London · University of Oxford

LLM AlignmentLanguage Model Safety EvaluationKnowledge Conflicts in Language ModelsLanguage ModelingMedical LLMs
AskPDF
May 12, 2026

Do Language Models Encode Knowledge of Linguistic Constraint Violations?

Large Language Models (LLMs) achieve strong linguistic performance, yet their internal mechanisms for producing these predictions remain unclear. We investigate the hypothesis that LLMs encode representations of linguistic constraint violations within their parameters, which are selectively activated when processing ungrammatical sentences. To test this, we use sparse autoencoders to decompose polysemantic activations into sparse, monosemantic features and recover candidates for violation-related features. We introduce a sensitivity score for identifying features that are preferentially activated on constraint-violated versus well-formed inputs, enabling unsupervised detection of potential violation-specific features. We further propose a conjunctive falsification framework with three criteria evaluated jointly. Overall, the results are negative in two respects: (1) the falsification criteria are not jointly satisfied across linguistic phenomena, and (2) no features are consistently shared across all categories. While some phenomena show partial evidence of selective causal structure, the overall pattern provides limited support for a unified set of grammatical violation detectors in current LMs.

Hardy, Sebastian Padó

IMS, University of Stuttgart, Stuttgart, Germany · Universitas Mikroskil, Medan, Indonesia.

LLM InterpretabilityMechanistic InterpretabilityLanguage Modeling
AskPDF
May 12, 2026

GRAFT: Graph-Tokenized LLMs for Tool Planning

Large language models (LLMs) are increasingly used to complete complex tasks by selecting and coordinating external tools across multiple steps. This requires aligning tool choices with subtask intent while satisfying directional execution dependencies among tools. To do this, existing methods model these dependencies as tool graphs and incorporate the graphs with LLMs through retrieval, serialization, or prompt-level injection. However, these external graph-use strategies all follow a matching paradigm, which often fails to align tool choices with the underlying subtask structure, producing semantically plausible plans that violate graph constraints. This issue is further exacerbated by error accumulation, where an early incorrect tool selection shifts the plan into an invalid graph state and causes subsequent predictions to drift away from the valid execution path. To address these challenges, we propose GRAFT, a graph-tokenized language model framework for dependency-aware tool planning. GRAFT internalizes the tool graph by mapping each tool node to a dedicated special token and learning directed tool dependencies within the representation space. It further introduces on-policy tool context distillation, training the model on its own sampled trajectories while distilling stepwise planning signals. Experiments show that GRAFT achieves state-of-the-art performance in exact sequence matching and dependency legality, supporting more reliable LLM tool planning in complex workflows.

Xinyi Gao, Xinyu Ren, Junliang Yu +3

The University of Queensland · Griffith University

Tool-Use PlanningLanguage ModelingGraph-Based Planning
AskPDF
May 11, 2026

A Theory of Time-Sensitive Language Generation: Sparse Hallucination Beats Mode Collapse

We study language generation in the limit under a global preference ordering on strings, as introduced by Kleinberg and Wei. As is done in previous work, we aim for breadth, but impose an additional requirement of timeliness: higher-ranked strings should be generated earlier. A string is then only credited if it is generated before a deadline, where its deadline is defined by a function that maps a string's rank in the target language to the time by which it must be produced. This is in keeping with a central consideration in machine learning, where inductive bias favors simpler'' or more plausible'' outputs, all else being equal. We show that timely generation is impossible in a strong sense for eventually consistent generators -- the protagonists of most prior related work. Under what is perhaps the mildest natural relaxation of consistency, a hallucination rate that vanishes over time, we show that we can circumvent our impossibility result. In particular, we can achieve optimal density with respect to any superlinear deadline function. We also show this is tight by ruling out timely generation with linear deadlines and vanishing hallucination rate.

Atul Ganju, Travis McVoy, Shaddin Dughmi +1

University of Southern California

Hallucination in Language ModelsLanguage ModelingText Generation
AskPDF
May 11, 2026

HEBATRON: A Hebrew-Specialized Open-Weight Mixture-of-Experts Language Model

We present Hebatron, a Hebrew-specialized open-weight large language model built on the NVIDIA Nemotron-3 sparse Mixture-of-Experts architecture. Training employs a three-phase easy-to-hard curriculum with continuous anti-forgetting anchoring, followed by supervised fine-tuning on 2 million bilingual Hebrew--English samples. The curriculum ordering alone yields a 3-point aggregate benchmark gain over the reversed configuration. Hebatron achieves a Hebrew reasoning average of 73.8%, outperforming DictaLM-3.0-24B-Thinking (68.9%) and remaining competitive with Gemma-3-27B-IT on GSM8K-HE and Israeli Trivia, while activating only 3B parameters per forward pass across a 30B-parameter model, delivering approximately 9 times higher inference throughput at native context lengths up to 65,536 tokens. To our knowledge, this is the first language-specific adaptation of the Nemotron-3 architecture for any target language, and the first open-weight Hebrew-specialized MoE model with native long-context support. Model weights are released openly to support further research in Hebrew and Semitic-language NLP.

Noam Kayzer, Dan Revital, Ori Bar Joseph +10

PwC Next

Mixture-of-Experts Language ModelsOpen-Weight Language ModelsLanguage ModelingLLM Training
AskPDF
May 11, 2026

Language Modeling with Hyperspherical Flows

Discrete Diffusion Language Models progressed rapidly as an alternative to autoregressive (AR) models, motivated by their parallel generation abilities. However, for tractability, discrete diffusion models sample from a factorized distribution, which is less expressive than AR. Recent Flow Language Models (FLMs) apply continuous flows to language, transporting noise to data with a deterministic ODE that avoids factorized sampling. FLMs operate on one-hot vectors whose dimension scales with the vocabulary size, making FLMs costly to train. Moreover, since all distinct one-hot embeddings are equidistant in ℓ2\ell_2, adding Gaussian noise does not have a clear semantic interpretation (unlike images, where Gaussian noise progressively degrades structure). We introduce S\mathbb{S}-FLM, a latent FLM in the hypersphere. S\mathbb{S}-FLM generates sequences by rotating vectors in Sd−1\mathbb{S}^{d-1} along a velocity field learned with cross-entropy, avoiding the overhead of materializing one-hot vectors. Previous FLMs match AR in Generative Perplexity (Gen.\ PPL), but samples with high likelihood are not necessarily correct in verifiable domains such as math and code. S\mathbb{S}-FLM substantially improves continuous flow language models on large-vocabulary reasoning and closes the gap to masked diffusion under standard-temperature sampling (T=1T=1), while a gap remains under optimized low-temperature (T=0.1T=0.1) decoding.

Justin Deschenaux, Caglar Gulcehre

EPFL, Lausanne, Switzerland

Normalizing FlowsContinuous Normalizing FlowsLanguage ModelingFlow-Based Generative Modeling
AskPDF
May 11, 2026

A Single-Layer Model Can Do Language Modeling

Modern language models scale depth by stacking layers, each holding its own state - a per-layer KV cache in transformers, a per-layer matrix in Mamba, Gated DeltaNet (GDN), RWKV, and xLSTM. Biological systems lean heavily on recurrence rather than on stacking. We ask how far that shape can go on language modeling. We propose Grounded Prediction Networks (GPN): one state vector revisited at every step through a single recurrent block - one FFN, one shared matrix memory. At 130M parameters, a 1-layer GPN+M reaches FineWeb-Edu perplexity 18.06, within 13% of a 12-layer Transformer++ (16.05) and 18% of a 10-layer GDN (15.34); a 2-layer variant closes the gap to 6%/11%. We do not match the deep baselines. Because the working context is a single vector, we can directly inspect its geometry: a persistent default-token direction, a content-bearing horizon of tens of tokens, and memory heads that split spontaneously into fast and slow retention pools.

Zanmin Wang

Language Model PretrainingRecurrent Language ModelsRecurrent Neural NetworksAutoregressive Language ModelingNext-Token PredictionLanguage Modeling
AskPDF
May 11, 2026

The Truth Lies Somewhere in the Middle (of the Generated Tokens)

How should hidden states generated autoregressively be collapsed into a representation that reflects a language model's internal state? Despite tokens being generated under causal masking, we find that mean pooling across their hidden states yields more semantic representations than any individual token alone. We quantify this through kernel alignment to reference spaces in language, vision, and protein domains. The improvement through mean pooling is consistent with information being distributed across generated tokens rather than localized to a single position. Furthermore, representations derived from generated tokens outperform those from prompt tokens, and alignment across generation reveals interpretable dynamics in model behavior.

Sophie L. Wang, Phillip Isola, Brian Cheung

1MIT.

LLM InterpretabilityAutoregressive Language ModelingLanguage Modeling
AskPDF
May 11, 2026

From Syntax to Semantics: Unveiling the Emergence of Chirality in SMILES Translation Models

Understanding how chemical language models (CLMs) learn chemical meaning from molecular string representations, rather than only surface-level string patterns, is an important question in chemical representation learning and machine learning for chemistry. Chirality provides a demanding test case: enantiomers can differ greatly in pharmacological activity and toxicity, yet CLMs often struggle to distinguish chiral configurations reliably. Here we present Pan-CORE (Pan-Chemical Omniscale Representation Engine), a family of autoregressive Transformer-based encoder-decoder models for SMILES translation, and use high-temporal-resolution checkpoint analysis to investigate how chiral information is learned during training. Across all tested Pan-CORE variants, we observe a reproducible jump-up in which chiral-token accuracy rises abruptly after a long plateau, suggesting that chiral learning stagnation is not explained by model capacity alone and instead reflects the complexity of chiral constraints. Analyses of attention dynamics, residual-stream trajectories, and latent-space geometry support an encoder-centered mechanism in which chiral-token representations undergo transient destabilization and reconstruction, seen as a V-shaped drop and recovery in vector norm and directional stability, together with a clear reorganization of chiral molecular representations in the latent space. Encoder-decoder cross-evaluation further supports the encoder-centered nature of the transition, and targeted attention-head ablation identifies a small set of chiral-sensitive heads whose removal selectively reduces chiral-token accuracy even in the fully trained model. These findings show that SMILES translation can serve as a useful experimental system for mechanistic analysis of semantic emergence in CLMs, with implications for interpretable chemical representation learning.

Zehao Li, Yasuhiro Yoshikai, Shumpei Nemoto +2

Laboratory of Molecular Pharmacokinetics, Graduate School of Pharmaceutical Sciences, The University of Tokyo, 7-3-1 Hongo, Bunkyo, Tokyo, Japan · The Institute of Statistical Mathematics (ISM), Research Organization of Information and Systems, 190-8562 Tachikawa, Tokyo, Japan

Chemical Language ModelsTransformer InterpretabilityLanguage ModelingMolecular Representation Learning
AskPDF
May 10, 2026

Language Models Without a Trainable Input Embedding Table: Learning from Fixed Minimal Binary Token Codes

Trainable input embedding tables are a standard component of modern language models. We ask whether they are actually necessary at the input interface. For a vocabulary of size VV, exact token identity requires only K=⌈log⁡2V⌉K=\lceil \log_2 V\rceil bits. We replace the usual trainable V×dmodelV\times d_{\text{model}} input embedding matrix with fixed minimal binary token codes and a zero-parameter lift to model width. In our main setting, V=65,536V=65{,}536, so K=16K=16, and tokens are represented by fixed 16-dimensional binary codes tiled to dmodel=1024d_{\text{model}}=1024. We also evaluate a fully table-free variant in which codes are generated from token IDs on the fly and randomly recoded by an invertible affine transform over F2K\mathbb{F}_2^K. Across matched 32-layer decoder-only models trained on approximately 17B tokens and evaluated over three independent training seeds, fixed minimal codes achieve comparable held-out validation perplexity to a standard learned-input baseline while removing 67.1M trainable input parameters. The fixed-code runs have a lower mean validation perplexity in our experiments, 2.36 versus 2.44, but the observed gap is within the measured seed-to-seed variation of 4.8%; we therefore interpret the result as evidence that the trainable input table is not necessary, rather than as a statistically resolved superiority claim. The table-free affine-recoded variant remains close at 2.39 despite a slightly shorter training run. These results show that, in this regime, a trainable input embedding table is not necessary for useful language modeling. The output projection remains standard and trainable.

A. Bochkov

Language Model PretrainingLanguage Modeling
AskPDF
May 10, 2026

Scratchpad Patching: Decoupling Compute from Patch Size in Byte-Level Language Models

Tokenizer-free language models eliminate the tokenizer step of the language modeling pipeline by operating directly on bytes; patch-based variants further aggregate contiguous byte spans into patches for efficiency. However, the average patch size chosen at the model design stage governs a tight trade-off: larger patches reduce compute and KV-cache footprint, but degrade modeling quality. We trace this trade-off to patch lag: until a patch is fully observed, byte predictions within it must rely on a stale representation from the previous patch to preserve causality; this lag widens as patches grow larger. We introduce Scratchpad Patching (SP), which inserts transient scratchpads inside each patch to aggregate the bytes seen so far and refresh patch-level context for subsequent predictions. SP triggers scratchpads using next-byte prediction entropy, selectively allocating compute to information-dense regions and enabling post-hoc adjustment of inference-time compute. Across experiments on natural language and code, SP improves model quality at the same patch size; for example, even at 1616 bytes per patch, SP-augmented models match or closely approach the byte-level baseline on downstream evaluations while using a 16×16\times smaller KV cache over patches and 33-4×4\times less inference compute.

Lin Zheng, Vasilisa Bashlovkina, Timothy Dozat +3

1Google DeepMind · 2The University of Hong Kong

Byte-Level Language ModelLanguage ModelingEfficient Language Model InferenceAdaptive Computation
AskPDF
May 10, 2026

Position: Avoid Overstretching LLMs for every Enterprise Task

Enterprise workloads are dominated by deterministic, structured, and knowledge-dependent tasks operating under strict cost, latency, and reliability constraints. While these are often addressed through large language model (LLM) deployment or distillation into smaller models, we argue this is inefficient, unreliable, and misaligned with enterprise task structures. Instead, AI systems should treat language models as interfaces rather than monolithic engines, externalizing knowledge and computation into dedicated components for greater reliability, scalability, and transparency. Our theoretical evidences show that finite-capacity models cannot fully capture the breadth of knowledge required for enterprise tasks, creating inherent limits to efficiency and interpretability. Building on this, we take the position that language models should primarily be used for structured extraction in deterministic enterprise workflows, while computation and storage are delegated to knowledge bases and symbolic procedures. We formally demonstrate that such modular architectures are more reliable and maintainable than monolithic frameworks, offering a sustainable foundation for enterprise tasks.

Kuldeep Singh, Anson Bastos, Isaiah Onando Mulang'

Microsoft, India · SAP, Germany

Language ModelingNeuro-Symbolic AIEfficient Language Model Inference
AskPDF
May 10, 2026

Understanding, Predicting, and Steering LLM Behavior via Representational Effective Theory

Large language models (LLMs) assist with writing, teaching, programming and decision-making, yet the computations underlying their behaviour remain largely opaque. Full access to every weight and activation does not by itself yield understanding: a microscopic description need not reveal how computation is organized at the level that governs behaviour. Complex systems are often understood instead through a few macroscopic variables whose dynamics capture the behaviour of interest, as temperature and pressure do for molecular motion in thermodynamics. Here we show that such an effective description of LLM computation can be learned directly from data. We introduce Representational Effective Theory (RET), a framework that learns low-dimensional macrostates and their dynamics from a model's hidden-state trajectories. These macrostates exhibit approximately closed dynamics and trace interpretable trajectories through reasoning. They also predict sycophancy and alignment faking early in generation, while steering the model towards selected macrostates produces corresponding changes in its reasoning. Together, these results show that LLM computation admits useful macroscopic descriptions that support interpretation, prediction and control.

Muhammed Ustaomeroglu, Guannan Qu

Carnegie Mellon University

LLM InterpretabilityMechanistic InterpretabilityLanguage ModelingPredictive Representation Learning
AskPDF
May 9, 2026

Sparse Layers are Critical to Scaling Looped Language Models

Looped language models repeat a set of transformer layers through depth, reducing memory costs and providing natural early-exit points at loop boundaries. However, looped models do not scale as favorably as standard transformers with unique layers. We compare standard and Mixture-of-Experts (MoE) transformers, with and without looping, and find two main results. First, we find Looped-MoE models scale better than the standard baseline while dense looped models do not. We trace this to routing divergence between loops: in Looped-MoE models, different experts are activated on each pass through the same shared layers, recovering expressivity without additional parameters. Our second finding is that looped models have better compute-quality trade-offs with early exits than standard models. Because each loop ends with the same layers that produce the final output, loop boundaries are superior exit points, as confirmed by earlier output convergence at these points. In sum, we provide a clear direction for scaling looped models: a Looped-MoE model with early exits can not only beat standard transformers at scale, but also enable significant memory and inference savings with minimal degradation in quality.

Ryan Lee, Jacob Biloki, Edward J. Hu +1

1USC Information Sciences Institute · 2Netflix · 3Independent Researcher

Mixture-of-Experts Language ModelsRecurrent TransformersLanguage Model ScalingLLM Inference AccelerationLanguage ModelingRecursive Language Models
AskPDF
May 8, 2026

Fast Byte Latent Transformer

Recent byte-level language models (LMs) match the performance of token-level models without relying on subword vocabularies, yet their utility is limited by slow, byte-by-byte autoregressive generation. We address this bottleneck in the Byte Latent Transformer (BLT) through new training and generation techniques. First, we introduce BLT Diffusion (BLT-D), a new model and our fastest BLT variant, trained with an auxiliary block-wise diffusion objective alongside the standard next-byte prediction loss. This enables an inference procedure that generates multiple bytes in parallel per decoding step, substantially reducing the number of forward passes required to generate a sequence. Second, we propose two extensions inspired by speculative decoding that trade some of this speed for higher generation quality: BLT Self-speculation (BLT-S), in which BLT's local decoder continues generating past its normal patch boundaries to draft bytes, which are then verified with a single full-model forward pass; and BLT Diffusion+Verification (BLT-DV), which augments BLT-D with an autoregressive verification step after diffusion-based generation. All methods may achieve an estimated memory-bandwidth cost over 50% lower than BLT on generation tasks. Each approach offers its own unique advantages, together removing key barriers to the practical use of byte-level LMs.

Julie Kallini, Artidoro Pagnoni, Tomasz Limisiewicz +5

FAIR at Meta · Stanford University · University of Washington

Speculative DecodingLLM Inference AccelerationByte-Level Language ModelLanguage ModelingDiffusion Language Model DecodingDiffusion Language Models
AskPDF
May 8, 2026

Benchmarking EngGPT2-16B-A3B against Comparable Italian and International Open-source LLMs

This report benchmarks the performance of ENGINEERING Ingegneria Informatica S.p.A.'s EngGPT2MoE-16B-A3B LLM, a 16B parameter Mixture of Experts (MoE) model with 3B active parameters. Performance is investigated across a wide variety of representative benchmarks, and is compared against comparably-sized open-source MoE and dense models. In comparison with popular Italian models, namely FastwebMIIA-7B, Minerva-7B, Velvet-14B, and LLaMAntino-3-ANITA-8B, EngGPT2MoE-16B-A3B performs as well or better on international benchmarks: ARC-Challenge, GSM8K, AIME24, AIME25, MMLU, and HumanEval (HE). It achieves the best performance for the longest context setting (32k) of the RULER benchmark. On the Italian benchmark dataset ITALIC, the model performs as well or better than the other models except for Velvet-14B, which outperforms it. Compared with popular MoE models of comparable size, the new model reports higher values than DeepSeek-MoE-16B-Chat on all considered benchmarks. It has higher values than Moonlight-16B-A3B on HE, MMLU, AIME24, AIME25, GSM8K, and the 32k RULER setting, but lower on BFCL and some ARC and ITALIC settings. Finally it has lower values than GPT-OSS-20B on most benchmarks, including HE, MMLU, AIME24, AIME25, GSM8K, ARC, BFCL, and the RULER 32k. When compared with popular dense models, EngGPT2MoE-16B-A3B reports higher values on AIME24 and AIME25 than Llama-3.1-8B-Instruct, Gemma-3-12b-it, and Ministral-3-8BInstruct-2512-BF16, but lower values on ITALIC, BFCL, and RULER with a 32k context. When performance is aggregated across all benchmark metrics, EngGPT2MoE-16B-A3B shows higher performance than the Italian models under evaluation while achieving lower results than some of the most performant international models, in particular GPT-5 nano and Qwen3-8B. Taken together, our findings find the new model to be a step forward for native Italian Large Language Models.

Andrea Sassella, Andrea Chizzola, Tommaso Bianchi +2

2DEIB, Politecnico di Milano · 1AIRIC, Politecnico di Milano

LLM EvaluationMixture-of-Experts Language ModelsOpen-Weight Language ModelsLanguage Modeling
AskPDF
May 8, 2026

Memory-Efficient Looped Transformer: Decoupling Compute from Memory in Looped Language Models

Recurrent LLM architectures have emerged as a promising approach for improving reasoning, as they enable multi-step computation in the embedding space without generating intermediate tokens. Models such as Ouro perform reasoning by iteratively updating internal representations while retaining a standard Key-Value (KV) cache across iterations, causing memory consumption to grow linearly with reasoning depth. Consequently, increasing the number of reasoning iterations can lead to prohibitive memory usage, limiting the practical scalability of such architectures. In this work, we propose Memory-Efficient Looped Transformer (MELT), a novel architecture that decouples reasoning depth from memory consumption. Instead of using a standard KV cache per layer and loop, MELT maintains a single KV cache per layer that is shared across reasoning loops. This cache is updated over time via a learnable gating mechanism. To enable stable and efficient training under this architecture, we propose to train MELT using chunk-wise training in a two phase procedure: interpolated transition, followed by attention-aligned distillation, both from the LoopLM starting model to MELT. Empirically, we show that MELT models fine-tuned from pretrained Ouro parameters outperform standard LLMs of comparable size, while maintaining a memory footprint comparable to those models and dramatically smaller than Ouro's. Overall, MELT achieves constant-memory iterative reasoning without sacrificing LoopLM performance, using only a lightweight post-training procedure.

Victor Conchello Vendrell, Arnau Padres Masdemont, Niccolò Grillo +3

Qualcomm AI Research is an initiative of Qualcomm Technologies, Inc.

KV-Cache ManagementMemory-Efficient InferenceRecurrent TransformersLanguage ModelingRecursive Language Models
AskPDF
May 8, 2026

mHC-SSM: Manifold-Constrained Hyper-Connections for State Space Language Models with Stream-Specialized Adapters

Manifold-Constrained Hyper-Connections (mHC) introduce a stability-motivated variant of multi stream residual mixing by constraining residual stream mixing matrices to the manifold of doubly stochastic matrices via Sinkhorn-Knopp projection. In his work, we study whether mHC-style constrained multi-stream residual topology transfers effectively to state space model (SSM) language modeling. We implement a static mHC mechanism around an SSM block by expanding the residual stream into multiple parallel streams, aggregating streams into a single SSM input through simplex-constrained pre-mixing, scattering the SSM output back to streams through simplex-constrained post-mixing, and applying Sinkhorn-projected residual stream mixing at each layer. We further introduce stream-specialized adapters that add lightweight stream-specific capacity through a shared bottleneck with per-stream scaling, applied both before stream aggregation and after the SSM output prior to scattering. We evaluate baseline single-stream SSM, static mHC SSM, and mHC SSM with adapters on WikiText-2 using identical training settings and report checkpoint-based validation loss, perplexity, throughput, and peak GPU memory. Under the reported fair checkpoint evaluation, static mHC improves validation loss from 6.3507 to 6.2448 and reduces perplexity from 572.91 to 515.35, while mHC with adapters further improves validation loss to 6.1353 and perplexity to 461.88. These gains are accompanied by modest throughput reductions from 1025.52 to 964.81 and 938.90 tokens per second, and increased peak memory from 2365 MB to 2568 MB and 3092 MB. The results suggest that mHC-inspired constrained multi-stream residual mixing can yield measurable quality improvements in SSM language models and that stream-specialized adapter capacity can further enhance performance with predictable efficiency tradeoffs.

Abdulvahap Mutlu, Şengül Doğan, Türker Tuncer

Department of Digital Forensics Engineering, Technology Faculty, Firat University, Elazig, Turkey

State Space Language ModelsLanguage Modeling
AskPDF
May 8, 2026

Understanding Performance Collapse in Layer-Pruned Large Language Models via Decision Representation Transitions

Layer pruning efficiently reduces Large Language Model (LLM) computational costs but often triggers sudden performance collapse. Existing representation-based analyses struggle to explain this mechanism. We propose studying pruning through decision representation. Focusing on multiple-choice tasks, we introduce two metrics, Decision Margin and Option Frequency, and an Iterative Pruning method to analyze layer-wise decision dynamics. Our findings reveal a sharp decision transition that partitions the network into two stages: a Silent Phase, where the model cannot yet predict the correct answer, and a Decisive Phase, where the correct prediction emerges. We also find that pruning the Decisive Phase has minimal impact, whereas pruning the Silent Phase triggers immediate performance collapse, highlighting its extreme sensitivity to structural changes. Therefore, we conclude that pruning-induced collapse stems from disrupting the Silent Phase, which prevents the critical decision transition from occurring.

Boyu Shi, Chang Liu, ChuanBao Gao +2

School of Computer Science and Engineering, Southeast University, Nanjing, China · Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications (Southeast University), Ministry of Education, China

LLM PruningLLM InterpretabilityLanguage ModelingNeural Network Pruning
AskPDF
May 8, 2026

LoopQ: Quantization for Looped Language Models

Looped language models (LoopLMs) increase effective computational depth by repeatedly applying shared Transformer blocks without increasing the parameter count. However, this reuse makes them particularly sensitive to post-training quantization (PTQ). We systematically study PTQ in LoopLMs and identify three sources of degradation: distribution shift across roles, state reuse across loop transitions, and recursive error accumulation. We propose LoopQ, a loop-aware PTQ framework that preserves a shared quantized backbone while introducing lightweight adaptations. LoopQ combines loop-aware activation scaling and selective transformations to accommodate changing hidden-state distributions, cross-loop state alignment to stabilize transitions, and trajectory-aware calibration to limit accumulated error. Under W4A4 quantization, LoopQ achieves a 67.8% average relative improvement in five-task mean accuracy across four backbones and a conservative estimate of 75.2% for the average relative perplexity reduction across eight backbone-dataset pairs, compared with the strongest static PTQ baseline for each comparison.

Rui Fang, Hsi-Wen Chen, Chia-Ming Cheng +1

National Taiwan University

Recurrent TransformersLanguage ModelingActivation QuantizationPost-Training QuantizationLow-Bit QuantizationRecursive Language Models
AskPDF
May 7, 2026

Echo: KV-Cache-Free Associative Recall with Spectral Koopman Operators

Long chain-of-thought reasoning and agentic tool-calling produce traces spanning tens of thousands of tokens, yet Transformer KV caches grow linearly with sequence length, creating a memory bottleneck on commodity hardware. State-space models offer constant-memory recurrence but suffer a memory cliff: retrieval accuracy collapses once the gap between a stored fact and its query exceeds the effective horizon of the recurrent state. We introduce Echo, a KV-cache-free associative recall architecture built around Spectral Koopman Attention (SKA); a drop-in replacement for attention layers that augments SSM blocks with a closed-form dynamical operator whose sufficient statistics are accumulated in constant memory with no KV cache. Echo fits a spectral linear system to the key and value history via kernel ridge regression and retrieves through a learned power-iterated filter, all from O(r2)O(r^{2}) streaming state where rr is a small projection rank. On the Multi-Query Associative Recall benchmark, a pure Mamba-2 SSM fails to exceed chance accuracy (∼3%{\sim}3\%) across all gap lengths and KV-pair counts, while at the 50M parameter scale SKA-augmented models achieve 100%100\% retrieval accuracy on every configuration tested, including distractor gaps of 4,0964{,}096 tokens with 3232 KV pairs. Across five additional transfer benchmarks including needle-in-a-haystack, tool-trace, and multi-hop retrieval, SKA consistently outperforms both pure SSM and SSM+Attention hybrids while maintaining constant inference memory. Ablations confirm that the spectral operator, not the prefix masking strategy, drives the retrieval gain.

Anupama Sridhar, Alexander Johansen

California, USA · Stanford University Computer Science California, USA

State Space Language ModelsKoopman Operator LearningMemory-Efficient InferenceLanguage ModelingAssociative Memory
AskPDF