Searcharxiv⌕ Search

arXiv subjects

Jonathan Geuter

Publications and source records attributed to Jonathan Geuter.

14 recordsLinked to original sources

K2-Horizon: A Frontier Fleet of Open Source Language Models

We present K2-Horizon, a fleet of six fully open language models, from a 0.9B dense model to a 375B-A23B MoE, built for general-purpose reasoning and agentic work across math and science, software engineering, deep research, office automation, and beyond. The models lead or are competitive with strong open models of their size on agentic and coding benchmarks. On the modeling side, to strengthen reasoning, we introduce Mixture-of-Value Attention (MoVA), applying MoE to attention value projections, and softplus attention gating, a more flexible, theoretically motivated gate. Our 36B-A4B model uses both and nearly matches a 32B dense model trained on the same data, with one-seventh the active parameters. We also introduce Uno, lossless diffusion adapters that speed up decoding by up to 2.2x without a separate draft model. Our release follows the LLM360 definition of open source for transparency and reproducibility: the full development history, including intermediate checkpoints from every training stage, pre-training data, training logs, and all training, evaluation, and infrastructure code, notably our production-scale xLLM framework for dense and MoE training and the RL360 reinforcement-learning stack. Beyond the data itself, we release and document in detail every pipeline that built it, the part of model development that open releases most rarely disclose. These pipelines produced roughly 10T synthetic pre-training tokens, showing that synthetic data can be a major ingredient of pre-training at scale. Four of the models share one 22T-token pre-training corpus and recipe, revealing consistent scaling behavior and enabling controlled studies of scale and sparsity. All artifacts are available at https://ifm.ai/k2/

cs.AI↗

Understanding Layer Patching in Model Size Interpolation

Layer patching is a key technique for model size interpolation in large language models (LLMs), in which a subset of a distilled student model's layers is replaced with corresponding teacher-model layers to construct intermediate-sized models without additional training. Prior work largely ignores the order in which layers are patched, which we show can dramatically affect the performance and reliability of the resulting models. In this work, we study the role of layer patching in model size interpolation. We cast finding the optimal layer subset for each model size as an optimization problem and prove it can be viewed as a shortest-path problem in a certain acyclic graph. In experiments, we show that patching strongly shapes interpolation behavior, with effects that vary substantially across model families. We find that simple sequential strategies---patching either from the first layer to the last or from the last to the first---often achieve surprisingly strong performance in practice. We further introduce KLPatch, a greedy patching algorithm based on KL divergence, which often improves over last-to-first patching and approximately solves the optimization problem. Together, our results provide a principled understanding of how layer patching affects model size interpolation and offer practical guidance for constructing near-optimal interpolated models. The code and models are available at https://github.com/dcml-lab/boomerang-distillation.

cs.LG↗

Beyond the Point Mass: Acceptance-Derived Weights and Distributional Losses for Speculative Decoding

In speculative decoding, a lightweight draft model proposes tokens that the target model verifies in parallel, speeding up generation without changing the target distribution. D-PACE (Wu et al., 2026) trains such drafters with a loss that weights each position by its contribution to the expected accepted length. Both its weights and its cross-entropy (CE) loss use only the draft's probability of the recorded target token, treating the target as a point mass. Under rejection sampling, however, a sampled draft token is accepted with probability equal to the overlap of the draft and target distributions, one minus their total variation (TV). We derive D-PAL (Dynamic Position-wise Acceptance Loss) from the exact expected accepted length, using this overlap as the acceptance signal. This yields overlap-based position weights with log acceptance (LA) as the per-position loss, and the objective reduces to D-PACE for a point-mass target. Since the weights are detached, any per-position loss can replace LA in practice. We show that forward KL divergence and Rényi divergence of order 1/2 match or exceed LA, and that Rényi-1/2 provably bounds the true acceptance rate from above and below. Replacing CE with Rényi-1/2 improves acceptance length under both D-PACE's weights and ours. On six benchmarks with a Qwen3-4B target, D-PAL raises mean acceptance length over D-PACE from 4.50 to 4.57 at T=0 and from 4.20 to 4.28 at T=1, and consistently achieves higher throughput at both temperatures.

cs.LG↗

Releasing the Inner Jev: A Simple Recipe for Open Decision Models

LLM-based decision-making systems, also called System One models, have recently gained popularity, most notably through Jev (Almeida, 2026). Given a typed input state, a question, and a set of options, they return a probability for each option in a single forward pass, without text generation. We show that existing open instruction-tuned language models can achieve comparable performance \textit{without any additional training}, by reading out answers from letters (A, B, C, etc.) in the model's output logits at the last token position. Furthermore, we show how a simple supervised fine-tuning recipe can improve their performance further. We evaluate four open instruction-tuned language models: Gemma 4 31B, Gemma 4 12B, Qwen3.6-27B, and Qwen3.5-4B. We fine-tune each using LoRA and a decision head initialized from a slice of the language model head. Furthermore, we introduce OpenDecisions, a novel 2,352-question validation dataset designed to reflect the task composition of JevBench (JevBench main- tainers, 2026) and which computes a Capability score based on accuracy and calibration metrics. Without fine-tuning, Gemma 4 31B scores 79.6 on OpenDecisions, compared to Jev's 80.1. After undergoing our fine-tuning recipe, it scores 84.4 on OpenDecisions and 81.5 on JevBench v1.7, outperforming Jev by 4.3 and 4.4 Capability points respectively and setting a new state-of-the-art on OpenDecisions. On a suite of public decision datasets, our models achieve accuracy comparable to Jev, with lower calibration error. Our results show that open language models already possess System One capabilities, and with simple fine-tuning they outperform Jev on a wide range of decision tasks. We release our models, serving code, and OpenDecisions.

cs.AI↗

Unlocking Lossless Speedups in LLMs via Discrete Diffusion

Large Language Models (LLMs) owe much of their success to next-token prediction (NTP), but their autoregressive (AR) structure requires slow, sequential token generation. To overcome this bottleneck, we introduce diffusion-augmented LLMs, a new class of models that defines an AR model distribution while using diffusion to draw multiple tokens in parallel from that distribution. We decouple the parameters of these models into two sets: AR weights, trained using the standard NTP objective, and lightweight diffusion weights, trained to generate multiple tokens simultaneously. The diffusion weights are learned through a simple Diffusion Distillation phase that adds negligible overhead to existing LLM training pipelines. We also introduce $Ψ$-Spec, a family of samplers that enables lossless acceleration and inference-time scaling at a fixed context length. Unlike speculative decoding, our method requires no separate draft model. Unlike diffusion LLMs (d-LLMs), it accelerates generation without sacrificing the quality of the underlying AR model. The resulting models, called Uno, can be trained from scratch or built by augmenting existing open-weight AR LLMs. Uno achieves higher throughput than leading speculative-decoding methods at every evaluated batch size and delivers up to $3\times$ speedups over the base AR model, including at the largest batch size supported by the device. Notably, our 8B Uno model outperforms the leading open d-LLM, the 26B DiffusionGemma, and the proprietary Mercury 2 across all evaluated benchmarks in agentic tool use, coding, and long-context reasoning. We release code and checkpoints at: https://s-sahoo.github.io/uno/

cs.LG↗

Thinking at the Right Size: Amortized Distillation Across Post-Trained LLMs

Practical deployment of large language models (LLMs) requires families of post-trained variants---instruction-tuned, reasoning-tuned, and chat-style models---each at multiple sizes to meet diverse latency and memory budgets. Producing each (variant, size) pair independently is prohibitive, so model families typically span only a handful of coarse-grained sizes per post-trained variant. Boomerang distillation (Kangaslahti et al., 2026) reduces this cost along the size axis for base models. Through model size interpolation, it constructs models of intermediate sizes from a single teacher-student pair without additional training. However, it still treats each post-trained variant as a separate object of optimization. We introduce ADAPT---Amortized Distillation Across Post-Trained LLMs---a framework for amortizing distillation across both axes of a model family: size and post-training variant, producing $L \times K$ models for $L$ interpolated sizes across $K$ post-trained variants with a single distillation run. ADAPT combines two components. First, a two-phase distillation procedure constructs post-trained students through pre-training alignment and supervised fine-tuning distillation, enabling smooth size--performance interpolation on generation and reasoning tasks. Second, weight-delta initialization approximates this construction across post-trained variants by transferring the distillation-induced weight change from the base model to students initialized from different post-trained variants. The resulting continuum of interpolated models also enables adaptive model-size selection at inference time, improving the compute--accuracy trade-off for long-form reasoning tasks.

cs.LG↗

Stop Training for the Worst: Progressive Unmasking Accelerates Masked Diffusion Training

Masked Diffusion Models (MDMs) have emerged as a promising approach for generative modeling in discrete spaces. By generating sequences in any order and allowing for parallel decoding, they enable fast inference and strong performance on non-causal tasks. However, this flexibility comes with a training complexity trade-off: MDMs train on an exponentially large set of masking patterns, which is not only computationally expensive, but also creates a train--test mismatch between the random masks used in training and the highly structured masks induced by inference-time unmasking. In this work, we propose Progressive UnMAsking (PUMA), a simple modification of the forward masking process that aligns training-time and inference-time masking patterns, thereby focusing optimization on inference-aligned masks and speeding up training. Empirically, PUMA speeds up pretraining at the 125M scale by $\approx 2.5\times$ and offers complementary advantages on top of common recipes like autoregressive initialization. We open-source our codebase at https://github.com/JaeyeonKim01/PUMA.

cs.LG↗

Guided Speculative Inference for Efficient Test-Time Alignment of LLMs

We propose Guided Speculative Inference (GSI), a novel algorithm for efficient reward-guided decoding in large language models. GSI combines soft best-of-$n$ test-time scaling with a reward model $r(x,y)$ and speculative samples from a small auxiliary model $π_S(y\mid x)$. We provably approximate both the optimal tilted policy $π_{β,B}(y\mid x) \propto π_B(y\mid x)\exp(β\,r(x,y))$ of soft best-of-$n$ under the base model $π_B$, as well as the expected reward under the optimal policy. In experiments on reasoning benchmarks (MATH500, OlympiadBench, Minerva Math, MMLU-STEM, GSM8K) and across different model families, our method achieves higher accuracy than standard soft best-of-$n$ with $π_S$ and reward-guided speculative decoding (Liao et al., 2025), and in certain settings even outperforms soft best-of-$n$ with $π_B$, while reducing end-to-end latency by up to $28\%$. The code is available at https://github.com/j-geuter/GSI .

cs.LG↗

Boomerang Distillation Enables Zero-Shot Model Size Interpolation

Large language models (LLMs) are typically deployed under diverse memory and compute constraints. Existing approaches build model families by training each size independently, which is prohibitively expensive and provides only coarse-grained size options. In this work, we identify a novel phenomenon that we call boomerang distillation: starting from a large base model (the teacher), one first distills down to a small student and then progressively reconstructs intermediate-sized models by re-incorporating blocks of teacher layers into the student without any additional training. This process produces zero-shot interpolated models of many intermediate sizes whose performance scales smoothly between the student and teacher, often matching or surpassing pretrained or distilled models of the same size. We further analyze when this type of interpolation succeeds, showing that alignment between teacher and student through pruning and distillation is essential. Boomerang distillation thus provides a simple and efficient way to generate fine-grained model families, dramatically reducing training cost while enabling flexible adaptation across deployment environments. The code and models are available at https://github.com/dcml-lab/boomerang-distillation.

cs.LG↗

RoBoN: Routed Online Best-of-n for Test-Time Scaling with Multiple LLMs

Best-of-$n$ is a widely used test-time scaling approach for LLM inference. Yet despite evidence that LLMs exhibit complementary strengths across tasks, traditionally best-of-$n$ relies on a single model to generate responses. We propose RoBoN (Routed Online Best-of-$n$), a sequential multi-LLM alternative to the prevailing single-model best-of-$n$. Given a suite of models $\{m_i\}_{i=1}^M$, RoBoN sequentially routes generations one-by-one across models, based on scores computed using a reward model and an agreement signal on the predicted responses. This online routing requires no additional training, keeps compute parity, and works with any plug-in reward model. Across reasoning benchmarks (MATH500, OlympiadBench, MinervaMath, GSM8K, MMLU), RoBoN consistently outperforms standard best-of-$n$ applied to each individual model for larger $n$, with gains of up to 3.4\% in absolute accuracy, and also improves over a uniform multi-model portfolio baseline. Our results indicate that diversity across models can be exploited at inference to improve best-of-$n$ performance over any constituent model alone, providing a simple, training-free path to test-time scaling with multiple LLMs.

cs.LG↗

Entropy-Driven Pre-Tokenization for Byte-Pair Encoding

Byte-Pair Encoding (BPE) has become a widely adopted subword tokenization method in modern language models due to its simplicity and strong empirical performance across downstream tasks. However, applying BPE to unsegmented languages such as Chinese presents significant challenges, as its frequency-driven merge operation is agnostic to linguistic boundaries. To address this, we propose two entropy-informed pre-tokenization strategies that guide BPE segmentation using unsupervised information-theoretic cues. The first approach uses pointwise mutual information and left/right entropy to identify coherent character spans, while the second leverages predictive entropy derived from a pretrained GPT-2 model to detect boundary uncertainty. We evaluate both methods on a subset of the PKU dataset and demonstrate substantial improvements in segmentation precision, recall, and F1 score compared to standard BPE. Our results suggest that entropy-guided pre-tokenization not only enhances alignment with gold-standard linguistic units but also offers a promising direction for improving tokenization quality in low-resource and multilingual settings.

cs.CL↗

Universal Neural Optimal Transport

Optimal Transport (OT) problems are a cornerstone of many applications, but solving them is computationally expensive. To address this problem, we propose UNOT (Universal Neural Optimal Transport), a novel framework capable of accurately predicting (entropic) OT distances and plans between discrete measures for a given cost function. UNOT builds on Fourier Neural Operators, a universal class of neural networks that map between function spaces and that are discretization-invariant, which enables our network to process measures of variable resolutions. The network is trained adversarially using a second, generating network and a self-supervised bootstrapping loss. We ground UNOT in an extensive theoretical framework. Through experiments on Euclidean and non-Euclidean domains, we show that our network not only accurately predicts OT distances and plans across a wide range of datasets, but also captures the geometry of the Wasserstein space correctly. Furthermore, we show that our network can be used as a state-of-the-art initialization for the Sinkhorn algorithm with speedups of up to $7.4\times$, significantly outperforming existing approaches.

cs.LG↗

DDEQs: Distributional Deep Equilibrium Models through Wasserstein Gradient Flows

Deep Equilibrium Models (DEQs) are a class of implicit neural networks that solve for a fixed point of a neural network in their forward pass. Traditionally, DEQs take sequences as inputs, but have since been applied to a variety of data. In this work, we present Distributional Deep Equilibrium Models (DDEQs), extending DEQs to discrete measure inputs, such as sets or point clouds. We provide a theoretically grounded framework for DDEQs. Leveraging Wasserstein gradient flows, we show how the forward pass of the DEQ can be adapted to find fixed points of discrete measures under permutation-invariance, and derive adequate network architectures for DDEQs. In experiments, we show that they can compete with state-of-the-art models in tasks such as point cloud classification and point cloud completion, while being significantly more parameter-efficient.

cs.LG↗

Jina Embeddings: A Novel Set of High-Performance Sentence Embedding Models

Jina Embeddings constitutes a set of high-performance sentence embedding models adept at translating textual inputs into numerical representations, capturing the semantics of the text. These models excel in applications like dense retrieval and semantic textual similarity. This paper details the development of Jina Embeddings, starting with the creation of high-quality pairwise and triplet datasets. It underlines the crucial role of data cleaning in dataset preparation, offers in-depth insights into the model training process, and concludes with a comprehensive performance evaluation using the Massive Text Embedding Benchmark (MTEB). Furthermore, to increase the model's awareness of grammatical negation, we construct a novel training and evaluation dataset of negated and non-negated statements, which we make publicly available to the community.

cs.CL↗