Efficient Language Model Training

Follow this topic Ask about this topic

This field asks how to train language models with fewer resources while preserving their quality, making model development more accessible and large-scale training more practical. Approaches include selecting and scheduling data, changing tokenization or model architecture, improving optimizers and gradient computation, and coordinating parallelism, memory, and hardware; related work also targets fine-tuning and reasoning-model training. Researchers evaluate these methods with measures such as training loss, downstream benchmarks, throughput, memory use, and energy across different model sizes and hardware settings. A central tension is that gains depend on the task and system, and lower training cost may come with losses in generalization, robustness, or reproducibility, so efficiency must be assessed alongside model quality.

Latest papers

162 papers27 papers in the last four weeks, up 108% on the four weeks before. 0.2% of all new papers.

Apr 4, 2026

Rethinking Token Prediction: Tree-Structured Diffusion Language Model

Discrete diffusion language models have emerged as a competitive alternative to auto-regressive language models, but training them efficiently under limited parameter and memory budgets remains challenging. Modern architectures are predominantly based on a full-vocabulary token prediction layer, which accounts for a substantial fraction of model parameters (e.g., more than 20% in small scale DiT-style designs) and often dominates peak GPU memory usage. This leads to inefficient use of both parameters and memory under constrained training resources. To address this issue, we revisit the necessity of explicit full-vocabulary prediction, and instead exploit the inherent structure among tokens to build a tree-structured diffusion language model. Specifically, we model the diffusion process with intermediate latent states corresponding to a token's ancestor nodes in a pre-constructed vocabulary tree. This tree-structured factorization exponentially reduces the classification dimensionality, makes the prediction head negligible in size, and enables reallocation of parameters to deepen the attention blocks. Empirically, under the same parameter budget, our method reduces peak GPU memory usage by half while matching the perplexity performance of state-of-the-art discrete diffusion language models.

Zihao Wu, Haoming Yang, Juncheng Dong +1

Department of Electrical & Computer Engineering, Duke University

Diffusion Language Models Efficient Language Model Training

Apr 1, 2026

Scalable Pretraining of Large Mixture of Experts Language Models on Aurora Super Computer

Pretraining Large Language Models (LLMs) from scratch requires massive amount of compute. Aurora super computer is an ExaScale machine with 127,488 Intel PVC (Ponte Vechio) GPU tiles. In this work, we showcase LLM pretraining on Aurora at the scale of 1000s of GPU tiles. Towards this effort, we developed Optimus, an inhouse training library with support for standard large model training techniques. Using Optimus, we first pretrained Mula-1B, a 1 Billion dense model and Mula-7B-A1B, a 7 Billion Mixture of Experts (MoE) model from scratch on 3072 GPU tiles for the full 4 trillion tokens of the OLMoE-mix-0924 dataset. We then demonstrated model scaling by pretraining three large MoE models Mula-20B-A2B, Mula-100B-A7B, and Mula-220B-A10B till 100 Billion tokens on the same dataset. On our largest model Mula-220B-A10B, we pushed the compute scaling from 384 to 12288 GPU tiles and observed scaling efficiency of around 90% at 12288 GPU tiles. We significantly improved the runtime performance of MoE models using custom GPU kernels for expert computation, and a novel EP-Aware sharded optimizer resulting in training speedups up to 1.71x. As part of the Optimus library, we also developed a robust set of reliability and fault tolerant features to improve training stability and continuity at scale.

Dharma Teja Vooturi, Dhiraj Kalamkar, Dipankar Das +1

Parallel Computing Lab (India) Intel Corporation

Language Model Pretraining LLM Training Efficient Language Model Training Mixture-of-Experts Language Models

Mar 31, 2026

Training-Free Dynamic Upcycling of Expert Language Models

Large Language Models (LLMs) have achieved remarkable performance on a wide range of specialized tasks, exhibiting strong problem-solving capabilities. However, training these models is prohibitively expensive, and they often lack domain-specific expertise because they rely on general knowledge datasets. Expertise finetuning can address this issue; however, it often leads to overspecialization, and developing a single multi-domain expert remains difficult due to diverging objectives. Furthermore, multitask training is challenging due to interference and catastrophic forgetting. Existing work proposes combining the expertise of dense models within a Mixture of Experts (MoE) architecture, although this approach still requires multitask finetuning. To address these issues, we introduce Dynamic Upcycling MoE (DUME), a novel approach that reuses dense experts trained on different domains to construct a unified MoE model. Our method builds a single multitask model that preserves the capabilities of the original dense experts without requiring additional training. DUME is both cost-efficient and scalable: by leveraging the closed-form solution of ridge regression, it eliminates the need for further optimization and enables experts to be added dynamically while maintaining the model's original performance. We demonstrate that DUME consistently outperforms baseline approaches in both causal language modeling and reasoning settings. Finally, we also show that the DUME model can be fine-tuned to further improve performance. We show that, in the causal language modeling setting, DUME can retain up to 97.6% of a dense expert model specialized in one particular domain, and that it can also surpass it in the reasoning setting, where it can achieve 102.1% of the dense expert performance. Our code is available at: github.com/gensyn-ai/dume.

Eros Fanì, Oğuzhan Ersoy

Gensyn

Mixture-of-Experts Language Models Efficient Language Model Training Model Merging

Mar 29, 2026

Time is Not Compute: Scaling Laws for Wall-Clock Constrained Training on Consumer GPUs

Scaling laws relate model quality to compute budget (FLOPs), but practitioners face wall-clock time constraints, not compute budgets. We study optimal model sizing under fixed time budgets from 5 minutes to 24 hours on consumer GPUs (RTX 4090). Across 70+ runs spanning 50M--1031M parameters, we find: (1)~at each time budget a U-shaped curve emerges where too-small models overfit and too-large models undertrain; (2)~optimal model size follows

N^* \propto t^{0.60}

, growing faster than Chinchilla's

N^* \propto C^{0.50}

, with

α= 0.60 \pm 0.07

robustly exceeding compute-optimal across all sensitivity analyses; (3)~a dual U-shape mechanism: short-budget U-curves arise from compute bottlenecks, while long-budget U-curves emerge from data bottlenecks (overfitting), with an intermediate regime where the U-curve temporarily disappears. These findings have immediate implications for researchers training on consumer hardware, where wall-clock time -- not FLOPs -- is the binding constraint. We release all code, logs, and 70+ experimental configurations.

Yi Liu

Independent Researcher

Scaling Laws Language Model Scaling Laws Efficient Language Model Training

Mar 27, 2026

Throughput Optimization as a Strategic Lever in Large-Scale AI Systems: Evidence from Dataloader and Memory Profiling Innovations

The development of large-scale foundation models, particularly Large Language Models (LLMs), is constrained by significant computational and memory bottlenecks. These challenges elevate throughput optimization from a mere engineering task to a critical strategic lever, directly influencing training time, operational cost, and the feasible scale of next-generation models. This paper synthesizes evidence from recent academic and industry innovations to analyze key advancements in training efficiency. We examine architectural solutions to dataloader bottlenecks, such as the OVERLORD framework, which has demonstrated a 4.5% improvement in end-to-end training throughput. We investigate memory optimization techniques designed to overcome the GPU memory wall, including CPU offloading strategies like DeepSpeed's ZeRO-Offload, which enable the training of models far exceeding single-accelerator capacity. Furthermore, we explore the growing importance of compiler-centric optimizations, exemplified by Triton-distributed, which enables the joint optimization of computation, memory, and communication for substantial performance gains. The analysis is contextualized by advanced profiling tools and hardware characterization studies that identify and mitigate previously overlooked overheads like Dynamic Voltage and Frequency Scaling (DVFS). Findings indicate that a holistic, system-level approach, integrating innovations across data pipelines, memory management, network fabrics, and compiler technologies, is essential for accelerating AI development, managing costs, and pushing the boundaries of model scale.

Mayank Jha

Amazon Robotics

Efficient Language Model Training LLM Training Memory-Efficient Training

Mar 26, 2026

MAGNET: Autonomous Expert Model Generation via Decentralized Autoresearch and BitNet Training

We present MAGNET (Model Autonomously Growing Network), a decentralized system for autonomous generation, training, and serving of domain-expert language models across commodity hardware. MAGNET integrates four components: (1) autoresearch, an autonomous ML research pipeline that automates dataset generation, hyperparameter exploration, evaluation, and error-driven iteration; (2) BitNet b1.58 ternary training, enabling CPU-native inference via bitnet.cpp without GPU hardware; (3) DiLoCo-based distributed merging for communication-efficient aggregation of domain specialists; and (4) on-chain contribution tracking on the HOOTi EVM chain. We validate autoresearch through three case studies: video safety classification (balanced accuracy 0.9287 to 0.9851), cryptocurrency directional prediction (41% to 54.9% hit rate), and BitNet hyperparameter optimization (10-phase sweep, -16.7% validation loss).

Yongwan Kim, Sungchul Park

Holo Studio Co., Ltd.

Automated Machine Learning LLM Training Decentralized Learning Efficient Language Model Training

Mar 21, 2026

SozKZ: Training Efficient Small Language Models for Kazakh from Scratch

Kazakh, a Turkic language spoken by over 22 million people, remains underserved by existing multilingual language models, which allocate minimal capacity to low-resource languages and employ tokenizers ill-suited to agglutinative morphology. We present SozKZ, a family of Llama-architecture language models (50M-600M parameters) trained entirely from scratch on 9 billion tokens of Kazakh text with a dedicated 50K BPE tokenizer. We evaluate all models on three Kazakh benchmarks -- multiple-choice cultural QA, reading comprehension (Belebele), and topic classification (SIB-200) -- alongside five multilingual baselines ranging from 500M to 3B parameters. Our 600M model achieves 30.3% accuracy on Kazakh cultural QA, approaching the 32.0% of Llama-3.2-1B (2x larger), and 25.5% on SIB-200 topic classification, surpassing all evaluated multilingual models up to 2B parameters. We observe consistent scaling from 50M to 600M, with MC QA accuracy rising from 22.8% to 30.3%, suggesting that further scaling remains beneficial. These results demonstrate that small, dedicated models trained from scratch with a language-appropriate tokenizer offer a viable path for low-resource language technology, achieving competitive performance at a fraction of the computational cost. All models and the tokenizer are released under open licenses.

Saken Tukenov

Independent Researcher

Language Model Pretraining Small Language Models Low-Resource Language Processing Efficient Language Model Training

Mar 20, 2026

SAGE: Sustainable Agent-Guided Expert-tuning for Culturally Attuned Translation in Low-Resource Southeast Asia

The vision of an inclusive World Wide Web is impeded by a severe linguistic divide, particularly for communities in low-resource regions of Southeast Asia. While large language models (LLMs) offer a potential solution for translation, their deployment in data-poor contexts faces a dual challenge: the scarcity of high-quality, culturally relevant data and the prohibitive energy costs of training on massive, noisy web corpora. To resolve the tension between digital inclusion and environmental sustainability, we introduce Sustainable Agent-Guided Expert-tuning (SAGE). This framework pioneers an energy-aware paradigm that prioritizes the "right data" over "big data". Instead of carbon-intensive training on unfiltered datasets, SAGE employs a reinforcement learning (RL) agent, optimized via Group Relative Policy Optimization (GRPO), to autonomously curate a compact training set. The agent utilizes a semantic reward signal derived from a small, expert-constructed set of community dialogues to filter out noise and cultural misalignment. We then efficiently fine-tune open-source LLMs on this curated data using Low-Rank Adaptation (LoRA). We applied SAGE to translation tasks between English and seven low-resource languages (LRLs) in Southeast Asia. Our approach establishes new state-of-the-art performance on BLEU-4 and COMET-22 metrics, effectively capturing local linguistic nuances. Crucially, SAGE surpasses baselines trained on full datasets while reducing data usage by 97.1% and training energy consumption by 95.2%. By delivering high-performance models with a minimal environmental footprint, SAGE offers a scalable and responsible pathway to bridge the digital divide in the Global South.

Zhixiang Lu, Chong Zhang, Yulong Li +5

Xi’an Jiaotong-Liverpool University Suzhou, China · University of Liverpool Liverpool, United Kingdom · Mohamed bin Zayed University of Artificial Intelligence Abu Dhabi, United Arab Emirates

Machine Translation Low-Resource MT LLM Fine-Tuning Efficient Language Model Training

Mar 19, 2026

Optimal Splitting of Language Models from Mixtures to Specialized Domains

Language models achieve impressive performance on a variety of knowledge, language, and reasoning tasks due to the scale and diversity of pretraining data available. The standard training recipe is a two-stage paradigm: pretraining first on the full corpus of data followed by specialization on a subset of high quality, specialized data from the full corpus. In the multi-domain setting, this involves continued pretraining of multiple models on each specialized domain, referred to as split model training. We propose a method for pretraining multiple models independently over a general pretraining corpus, and determining the optimal compute allocation between pretraining and continued pretraining using scaling laws. Our approach accurately predicts the loss of a model of size N with D pretraining and D' specialization tokens, and extrapolates to larger model sizes and number of tokens. Applied to language model training, our approach improves performance consistently across common sense knowledge and reasoning benchmarks across different model sizes and compute budgets.

Skyler Seto, Pierre Ablin, Anastasiia Filippova +4

Apple · National University of Singapore

Efficient Language Model Training Domain-Adaptive Pretraining Language Model Scaling Laws

Mar 19, 2026

Anatomical Heterogeneity in Transformer Language Models

Current transformer language models are trained with uniform computational budgets across all layers, implicitly assuming layer homogeneity. We challenge this assumption through empirical analysis of SmolLM2-135M, a 30-layer, 135M-parameter causal language model, using five diagnostic metrics: weight predictability (R2), ablation degradation, recovery speed, weight manipulation robustness, and structural analysis. We find profound anatomical heterogeneity: (1) Layer weights follow strong mathematical regularity (R2 = 0.91) with a universal oscillatory delta pattern (correlation ~= -0.50), yet predicted weights cause catastrophic failure due to nonlinear error accumulation. (2) Layer importance spans a 10^7 range, from a critical core (L8-11, up to +63,419% PPL degradation) to anti-layers (L14, L17) whose removal improves performance. (3) Recovery speed correlates with layer importance, indicating differential training requirements. (4) Only weight scaling (alpha = 0.9) preserves model quality among five tested manipulation strategies. (5) Growth Transformer Training, allocating budget by layer importance, achieves ~54% cost reduction. A proof-of-concept experiment confirms this: 4.7x lower validation loss than uniform training at identical parameter count, while being 13% faster.

Tomasz Wietrzykowski

Independent Researcher, Wroclaw, Poland

Transformer Efficient Language Model Training Neural Network Interpretability Decoder-Only Language Models

Mar 19, 2026

SpecForge: A Flexible and Efficient Open-Source Training Framework for Speculative Decoding

Large language models incur high inference latency due to sequential autoregressive decoding. Speculative decoding alleviates this bottleneck by using a lightweight draft model to propose multiple tokens for batched verification. However, its adoption has been limited by the lack of high-quality draft models and scalable training infrastructure. We introduce SpecForge, an open-source, production-oriented framework for training speculative decoding models with full support for EAGLE-3. SpecForge incorporates target-draft decoupling, hybrid parallelism, optimized training kernels, and integration with production-grade inference engines, enabling up to 9.9x faster EAGLE-3 training for Qwen3-235B-A22B. In addition, we release SpecBundle, a suite of production-grade EAGLE-3 draft models trained with SpecForge for mainstream open-source LLMs. Through a systematic study of speculative decoding training recipes, SpecBundle addresses the scarcity of high-quality drafts in the community, and our draft models achieve up to 4.48x end-to-end inference speedup on SGLang, establishing SpecForge as a practical foundation for real-world speculative decoding deployment.

Shenggui Li, Chao Wang, Yikai Zhu +14

Nanyang Technological University · SpecForge Team · Meituan +2

Speculative Decoding LLM Inference Acceleration Efficient Language Model Training

Mar 18, 2026

Learning to Reason with Curriculum I: Provable Benefits of Autocurriculum

Chain-of-thought reasoning, where language models expend additional computation by producing thinking tokens prior to final responses, has driven significant advances in model capabilities. However, training these reasoning models is extremely costly in terms of both data and compute, as it involves collecting long traces of reasoning behavior from humans or synthetic generators and further post-training the model via reinforcement learning. Are these costs fundamental, or can they be reduced through better algorithmic design? We show that autocurriculum, where the model uses its own performance to decide which problems to focus training on, provably improves upon standard training recipes for both supervised fine-tuning (SFT) and reinforcement learning (RL). For SFT, we show that autocurriculum requires exponentially fewer reasoning demonstrations than non-adaptive fine-tuning, by focusing teacher supervision on prompts where the current model struggles. For RL fine-tuning, autocurriculum decouples the computational cost from the quality of the reference model, reducing the latter to a burn-in cost that is nearly independent of the target accuracy. These improvements arise purely from adaptive data selection, drawing on classical techniques from boosting and learning from counterexamples, and requiring no assumption on the distribution or difficulty of prompts.

Nived Rajaraman, Audrey Huang, Miro Dudik +3

Microsoft Research · University of Illinois Urbana Champaign

Efficient Language Model Training Curriculum RL RL for Language Model Reasoning RL Fine-Tuning

Mar 17, 2026

EngGPT2: Sovereign, Efficient and Open Intelligence

EngGPT2-16B-A3B is the latest iteration of Engineering Group's Italian LLM and it's built to be a Sovereign, Efficient and Open model. EngGPT2 is trained on 2.5 trillion tokens - less than Qwen3's 36T or Llama3's 15T - and delivers performance on key benchmarks, including MMLU-Pro, GSM8K, IFEval and HumanEval, comparable to dense models in the 8B-16B range, while requiring one-fifth to half of the inference power, and between one-tenth to one-sixth of the training data and consequent needed training power. Designed as a trained-from-scratch Mixture-of-Experts (MoE) architecture, EngGPT2 features 16 billion parameters with 3 billion active per inference, with expert sizes positioned between those used in GPT-OSS and Qwen3. Approximately 25% of its training corpus consists of Italian-language data, to deliver strong capabilities for European and Italian NLP tasks among models of similar scale. This efficiency aims to position EngGPT2 as a key contributor to the growing portfolio of open-weight European models, combining performance and efficiency with full alignment to the EU AI Act. EngGPT2 is also a single model capable of multiple reasoning modes: non-reasoning, reasoning in Italian or English, and turbo-reasoning (a concise, bullet-point style reasoning available in both languages designed for real-time reasoning use cases). EngGPT2 aims to set a new standard for resource-conscious, high-performance LLMs tailored to European and Italian contexts.

G. Ciarfaglia, A. Rosanova, S. Cipolla +17

Language Modeling Open-Weight Language Models Efficient Language Model Inference Efficient Language Model Training Mixture-of-Experts Language Models

Mar 16, 2026

HO-SFL: Hybrid-Order Split Federated Learning with Backprop-Free Clients and Dimension-Free Aggregation

Fine-tuning large models on edge devices is severely hindered by the memory-intensive backpropagation (BP) in standard frameworks like federated learning and split learning. While substituting BP with zeroth-order optimization can significantly reduce memory footprints, it typically suffers from prohibitively degraded convergence speed. To resolve this dilemma, we propose Hybrid-Order Split Federated Learning (HO-SFL). By reformulating the split learning process within a Lagrangian framework, HO-SFL decouples the optimization landscape: The server performs precise first-order updates (i.e., BP), whereas clients conduct memory-efficient zeroth-order optimization. This hybrid design not only eliminates the need for client-side BP but also enables dimension-free model aggregation, drastically lowering communication costs. Crucially, we provide a theoretical convergence analysis, demonstrating that HO-SFL mitigates the dimension-dependent convergence slowdown of zeroth-order optimization, achieving a convergence rate comparable to first-order methods. Extensive experiments on tasks across vision and language modalities validate that HO-SFL achieves convergence speeds comparable to first-order baselines while significantly reducing communication costs and client memory footprints.

Qiyuan Chen, Xian Wu, Yi Wang +1

Department of Electrical and Computer Engineering, The University of Hong Kong, Hong Kong SAR, China.

Communication-Efficient Distributed Training Zeroth-Order Optimization Split Federated Learning Efficient Language Model Training Memory-Efficient Training Federated Learning

Mar 16, 2026

Towards Next-Generation LLM Training: From the Data-Centric Perspective

Large language models (LLMs) have demonstrated remarkable performance across a wide range of tasks and domains, with data playing a central role in enabling these advances. Despite this success, the preparation and effective utilization of the massive datasets required for LLM training remain major bottlenecks. In current practice, LLM training data is often constructed using ad hoc scripts, and there is still a lack of mature, agent-based data preparation systems that can automatically construct robust and reusable data workflows, thereby freeing data scientists from repetitive and error-prone engineering efforts. Moreover, once collected, datasets are often consumed largely in their entirety during training, without systematic mechanisms for data selection, mixture optimization, or reweighting. To address these limitations, we advocate two complementary research directions. First, we propose building a robust, agent-based automatic data preparation system that supports automated workflow construction and scalable data management. Second, we argue for a unified data-model interaction training system in which data is dynamically selected, mixed, and reweighted throughout the training process, enabling more efficient, adaptive, and performance-aware data utilization. Finally, we discuss the remaining challenges and outline promising directions for future research and system development.

Hao Liang, Zhengyang Zhao, Zhaoyang Han +8

Peking University, Beijing, China · Hong Kong University of Science and Technology, Hong Kong SAR, China · Institute for Advanced Algorithms Research, Shanghai, China +2

LLM Training Efficient Language Model Training Training Data Selection Data Mixture Optimization

Mar 10, 2026

Lost in Backpropagation: The LM Head is a Gradient Bottleneck

The last layer of neural language models (LMs) projects output features of dimension

D

to logits in dimension

V

, the size of the vocabulary, where usually

D \ll V

. This mismatch is known to raise risks of limited expressivity in neural LMs, creating a so-called softmax bottleneck. We show the softmax bottleneck is not only an expressivity bottleneck but also an optimization bottleneck. Backpropagating

V

-dimensional gradients through a rank-

D

linear layer induces unavoidable compression, which alters the training feedback provided to the vast majority of the parameters. We present a theoretical analysis of this phenomenon and measure empirically that 95-99% of the gradient norm is suppressed by the output layer, resulting in vastly suboptimal update directions. We conduct controlled pretraining experiments showing that the gradient bottleneck makes trivial patterns unlearnable, and drastically affects the training dynamics of LLMs. We argue that this inherent flaw contributes to training inefficiencies at scale independently of the model architecture, and raises the need for new LM head designs.

Nathan Godey, Yoav Artzi

Cornell University, New York, USA

LLM Training Efficient Language Model Training

Mar 5, 2026

Progressive Residual Warmup for Language Model Pretraining

Transformer architectures serve as the backbone for most modern Large Language Models, therefore their pretraining stability and convergence speed are of central concern. Motivated by the logical dependency of sequentially stacked layers, we propose Progressive Residual Warmup (ProRes) for language model pretraining. ProRes implements an "early layer learns first" philosophy by multiplying each layer's residual with a scalar that gradually warms up from 0 to 1, with deeper layers taking longer warmup steps. In this way, deeper layers wait for early layers to settle into a more stable regime before contributing to learning. We demonstrate the effectiveness of ProRes through pretraining experiments across various model scales, as well as normalization and initialization schemes. Comprehensive analysis shows that ProRes not only stabilizes pretraining but also introduces a unique optimization trajectory, leading to faster convergence, stronger generalization and better downstream performance. Our code is available at https://github.com/dandingsky/ProRes.

Tianhao Chen, Xin Xu, Lu Yin +4

The Hong Kong University of Science and Technology, Hong Kong, China · University of Surrey, Guildford, UK · The University of Hong Kong, Hong Kong, China +1

Language Model Pretraining LLM Training Efficient Language Model Training

Mar 5, 2026

Bidirectional Curriculum Generation: A Multi-Agent Framework for Data-Efficient Mathematical Reasoning

Enhancing mathematical reasoning in Large Language Models typically demands massive datasets, yet data efficiency remains a critical bottleneck. While Curriculum Learning attempts to structure this process, standard unidirectional approaches (simple-to-complex) suffer from inefficient sample utilization: they blindly escalate complexity even when foundational gaps persist, leading to wasted computation on unsolvable problems. To maximize the instructional value of every training sample, we introduce a novel Bidirectional Curriculum Generation framework. Unlike rigid trajectories, our multi-agent ecosystem mimics adaptive pedagogy to establish a closed feedback loop. It dynamically generates data by either complicating problems to challenge the model or, crucially, simplying them to repair specific reasoning failures. This mechanism ensures that the model consumes only the most effective data at any given stage. Grounded in the Optimal Pacing Theorem, our approach optimizes the learning trajectory, significantly outperforming baselines while achieving superior reasoning performance with substantially fewer instruction samples.

Boren Hu, Xiao Liu, Boci Peng +4

Zhejiang University · University of Macau · Peking University +3

Curriculum Learning LLM Mathematical Reasoning Efficient Language Model Training Synthetic Data Generation for Language Models

Feb 27, 2026

Distribution-Aware Companding Quantization of Large Language Models

Large language models such as GPT and Llama are trained with a next-token prediction loss. In this work, we suggest that training language models to predict multiple future tokens at once results in higher sample efficiency. More specifically, at each position in the training corpus, we ask the model to predict the following n tokens using n independent output heads, operating on top of a shared model trunk. Considering multi-token prediction as an auxiliary training task, we measure improved downstream capabilities with no overhead in training time for both code and natural language models. The method is increasingly useful for larger model sizes and keeps its appeal when training for multiple epochs. Gains are especially pronounced on generative benchmarks like coding, where our models consistently outperform strong baselines by several percentage points. Our 13B parameter models solves 12 % more problems on HumanEval and 17 % more on MBPP than comparable next-token models. Experiments on small algorithmic tasks demonstrate that multi-token prediction is favorable for the development of induction heads and algorithmic reasoning capabilities. As an additional benefit, models trained with 4-token prediction are up to 3X times faster at inference, even with large batch sizes.

Athul Radhakrishnan, Siddhant Mohan, Mahima Sachdeva

LLM Training Multi-Token Prediction Efficient Language Model Training

Feb 27, 2026

Preference Packing: Efficient Preference Optimization for Large Language Models

Resource-efficient training optimization techniques are becoming increasingly important as the size of large language models (LLMs) continues to grow. In particular, batch packing is commonly used in pre-training and supervised fine-tuning to achieve resource-efficient training. We propose preference packing, a method to enhance resource efficiency in training techniques that use data with different responses for the same input prompt, such as reward models or Direct Preference Optimization (DPO). Preference packing improves resource efficiency by reducing the attention operations for duplicate input prompts and decreasing KV cache memory usage. We conducted experiments on text-only datasets and image-included datasets and achieved at least 37% reduction in training time. Notably, this method can be applied alongside existing optimization techniques such as batch sorting, resulting in a 3.22x speedup.

Jaekyung Cho

AWS GenAI Innovation Center

Efficient Language Model Training Preference Optimization Memory-Efficient Training

Feb 26, 2026

RLHFless: Serverless Computing for Efficient RLHF

Reinforcement Learning from Human Feedback (RLHF) has been widely applied to Large Language Model (LLM) post-training to align model outputs with human preferences. Recent models, such as DeepSeek-R1, have also shown RLHF's potential to improve LLM reasoning on complex tasks. In RL, inference and training co-exist, creating dynamic resource demands throughout the workflow. Compared to traditional RL, RLHF further challenges training efficiency due to expanding model sizes and resource consumption. Several RLHF frameworks aim to balance flexible abstraction and efficient execution. However, they rely on serverful infrastructures, which struggle with fine-grained resource variability. As a result, during synchronous RLHF training, idle time between or within RL components often causes overhead and resource wastage. To address these issues, we present RLHFless, the first scalable training framework for synchronous RLHF, built on serverless computing environments. RLHFless adapts to dynamic resource demands throughout the RLHF pipeline, pre-computes shared prefixes to avoid repeated computation, and uses a cost-aware actor scaling strategy that accounts for response length variation to find sweet spots with lower cost and higher speed. In addition, RLHFless assigns workloads efficiently to reduce intra-function imbalance and idle time. Experiments on both physical testbeds and a large-scale simulated cluster show that RLHFless achieves up to 1.35x speedup and 44.8% cost reduction compared to the state-of-the-art baseline.

Rui Wei, Hanfei Yu, Shubham Jain +5

Stevens Institute of Technology · Northeastern University · Stony Brook University +1

Efficient Language Model Training RL from Human Feedback

Feb 26, 2026

Semantic Tube Prediction: Beating LLM Data Efficiency with JEPA

Large Language Models (LLMs) obey consistent scaling laws -- empirical power-law fits that predict how loss decreases with compute, data, and parameters. While predictive, these laws are descriptive rather than prescriptive: they characterize typical training, not optimal training. Surprisingly few works have successfully challenged the data-efficiency bounds implied by these laws -- which is our primary focus. To that end, we introduce the Geodesic Hypothesis, positing that token sequences trace geodesics on a smooth semantic manifold and are therefore locally linear. Building on this principle, we propose a novel Semantic Tube Prediction (STP) task, a JEPA-style regularizer that confines hidden-state trajectories to a tubular neighborhood of the geodesic. STP generalizes JEPA to language without requiring explicit multi-view augmentations. We show this constraint improves signal-to-noise ratio, and consequently preserves diversity by preventing trajectory collisions during inference. Empirically, STP allows LLMs to match baseline accuracy with 16

\times

less training data on the NL-RX-SYNTH dataset, directly violating the data term of Chinchilla-style scaling laws and demonstrating that principled geometric priors can surpass brute-force scaling. Code is available at https://github.com/galilai-group/llm-jepa#stp.

Hai Huang, Yann LeCun, Randall Balestriero

Atlassian · NYU · Brown

Efficient Language Model Training LLM Training Joint-Embedding Predictive Architecture Representation Geometry in Language Models Language Model Scaling Laws

Feb 25, 2026

A Diversity Diet for a Healthier Model: A Case Study of French ModernBERT

Diversity has been gaining interest in the NLP community in recent years. At the same time, state-of-the-art transformer models such as ModernBERT use very large pre-training datasets, which are driven by size rather than by diversity. This summons to investigate the impact of diversity on pre-training. We do so in this study, with the express intent of reducing pre-training dataset size, while retaining at least comparable performance. We compare diversity-driven sampling algorithms, and we use the best one to pre-train several ModernBERT models on French with a fixed compute budget. We fine-tune and evaluate them on a variety of French benchmarks. We compare them with models pre-trained on randomly sampled data of commensurate size, with the same compute budget. We find that both random and diversity-driven sampling may reduce the pre-training dataset by up to 94% and the pre-training time by up to 73% while maintaining performance. Moreover, in some tasks, the inherent quality of models, estimated via head-only fine-tuning, is up to 10 points higher with diversity sampling than with random sampling.

Louis Estève, Christophe Servan, Thomas Lavergne +1

Université Paris-Saclay, CNRS, LISN

Language Model Pretraining Diversity-Aware Sampling Efficient Language Model Training Language Modeling

Feb 25, 2026

Efficient Scaling of LLM Training with Flexible Context Parallelism

Scaling long-context capabilities is crucial for Large Language Models (LLMs). However, real-world data contain a large number of sequences with heterogeneous lengths. Existing training libraries for LLMs rely on static parallelism strategies, which suffer from severe load imbalance, redundant communication, and suboptimal hardware utilization under data heterogeneity. In this work, we propose Flexible Context Parallelism (FCP), an efficient parallelism strategy that adaptively reconfigures communication groups and context parallelism degrees during LLM training. We generalize more flexible non-power-of-two parallelism degrees and develop a polynomial-time algorithm to generate near-optimal parallelism strategies with only millisecond-level overhead per training batch. FCP is able to maintain high hardware efficiency even under extreme data heterogeneity. Experimental results demonstrate that FCP significantly outperforms Megatron-LM and DeepSpeed in both LLM and MLLM training, achieving up to 1.46x speedup in average throughput while maintaining near-linear scaling efficiency across large-scale clusters. For extremely unbalanced batches, FCP even achieves 2.24x speedup.

Yifan Niu, Han Xiao, Dongyi Liu +2

The Hong Kong University of Science and Technology (Guangzhou) · Huawei Technologies Co., Ltd. · The Hong Kong University of Science and Technology

Context Parallelism Efficient Language Model Training LLM Training Distributed Training Communication-Efficient Distributed Training

Feb 19, 2026

Entropy-Based Data Selection for Language Models

Modern language models (LMs) increasingly require two critical resources: computational resources and data resources. Data selection techniques can effectively reduce the amount of training data required for fine-tuning LMs. However, their effectiveness is closely related to computational resources, which always require a high compute budget. Owing to the resource limitations in practical fine-tuning scenario, we systematically reveal the relationship between data selection and uncertainty estimation of selected data. Although large language models (LLMs) exhibit exceptional capabilities in language understanding and generation, which provide new ways to alleviate data scarcity, evaluating data usability remains a challenging task. This makes efficient data selection indispensable. To mitigate these issues, we propose Entropy-Based Unsupervised Data Selection (EUDS) framework. Empirical experiments on sentiment analysis (SA), topic classification (Topic-CLS), and question answering (Q&A) tasks validate its effectiveness. EUDS establishes a computationally efficient data-filtering mechanism. Theoretical analysis and experimental results confirm the effectiveness of our approach. EUDS significantly reduces computational costs and improves training time efficiency with less data requirement. This provides an innovative solution for the efficient fine-tuning of LMs in the compute-constrained scenarios.

Hongming Li, Yang Liu, Chao Huang

School of Computer and Communication Engineering, University of Science and Technology Beijing, Beijing 100083, China · National Key Laboratory of General Artificial Intelligence, Beijing Institute for General Artificial Intelligence, Beijing 100086, China · Shunde Innovation School, University of Science and Technology Beijing, Foshan 528399, China

Data Selection for Language Model Fine-Tuning Training Data Selection Efficient Language Model Training LLM Fine-Tuning

Feb 19, 2026

Predictive Batch Scheduling: Accelerating Language Model Training Through Loss-Aware Sample Prioritization

We introduce Predictive Batch Scheduling (PBS), a novel training optimization technique that accelerates language model convergence by dynamically prioritizing high-loss samples during batch construction. Unlike curriculum learning approaches that require predefined difficulty metrics or hard example mining methods that demand expensive per-sample loss tracking, PBS employs a lightweight linear predictor trained online to estimate sample difficulty from static token-level features. Our predictor achieves 0.44 correlation with actual loss using only four simple features: token frequency, sequence length, vocabulary diversity, and rare token ratio. Experiments on a 130M parameter transformer demonstrate that PBS achieves 6-13% faster convergence measured by evaluation loss across training checkpoints, with the predictor's correlation improving from 0.14 to 0.44 over 10,000 training steps. These results validate that token frequency statistics encode meaningful information about sample difficulty, enabling effective curriculum learning with negligible computational overhead.

Sumedh Rasal

Georgia Institute of Technology

Efficient Language Model Training Adaptive Sampling Curriculum Learning Language Model Pretraining

Feb 16, 2026

ÜberWeb: Insights from Multilingual Curation for a 20-Trillion-Token Dataset

Multilinguality is a core capability for modern foundation models, yet training high-quality multilingual models remains challenging due to uneven data availability across languages. A further challenge is the performance interference that can arise from joint multilingual training, commonly referred to as the "curse of multilinguality". We study multilingual data curation across thirteen languages and find that many reported regressions are not inherent to multilingual scaling but instead stem from correctable deficiencies in data quality and composition rather than fundamental capacity limits. In controlled bilingual experiments, improving data quality for any single language benefits others: curating English improves non-English performance in 12 of 13 languages, while curating non-English yields reciprocal improvements in English. Bespoke per-language curation produces substantially larger within-language improvements. Extending these findings to large-scale general-purpose training mixtures, we show that curated multilingual allocations comprising under 8% of total tokens remain remarkably effective. We operationalize this approach within an effort that produced a 20T-token pretraining corpus derived entirely from public sources. Models with 3B and 8B parameters trained on a 1T-token random subset achieve competitive multilingual accuracy with 4-10x fewer training FLOPs than strong public baselines, establishing a new Pareto frontier in multilingual performance versus compute. Moreover, these benefits extend to frontier model scale: the 20T-token corpus served as part of the pretraining dataset for Trinity Large (400B/A13B), which exhibits strong multilingual performance relative to its training FLOPs. These results show that targeted, per-language data curation mitigates multilingual interference and enables compute-efficient multilingual scaling.

DatologyAI, :, Aldo Gael Carranza +32

DatologyAI

Multilingual Language Models Training Data Curation Language Model Pretraining Efficient Language Model Training

Feb 12, 2026

Stabilizing Native Low-Rank LLM Pretraining

Foundation models have achieved remarkable success, yet their growing parameter counts pose significant computational and memory challenges. Low-rank factorization offers a promising route to reduce training and inference costs, but the community lacks a stable recipe for training models from scratch using exclusively low-rank weights while matching the performance of the dense model. We demonstrate that Large Language Models (LLMs) can be trained from scratch using exclusively low-rank factorized weights for all non-embedding matrices without auxiliary "full-rank" guidance required by prior methods. While native low-rank training often suffers from instability and loss spikes, we identify uncontrolled growth in the spectral norm (largest singular value) of the weight matrix update as the dominant factor. To address this, we introduce Spectron: Spectral renormalization with orthogonalization, which dynamically bounds the resultant weight updates based on the current spectral norms of the factors. Our method enables stable, end-to-end factorized training with negligible overhead. Finally, we establish compute-optimal scaling laws for natively low-rank transformers, demonstrating predictable power-law behavior and improved inference efficiency relative to dense models.

Paul Janson, Edouard Oyallon, Eugene Belilovsky

Concordia University, Montreal, Canada · Mila Quebec AI Institute, Montreal, Canada · Sorbonne University, CNRS, Paris, France

Language Model Pretraining Low-Rank Matrix Decomposition Language Model Scaling Laws Efficient Language Model Training Low-Rank Compression Low-Rank Approximation

Feb 6, 2026

Revisiting the Shape Convention of Transformer Language Models

The architectural shape of dense Transformers has remained remarkably stable: narrow-wide-narrow feed-forward networks (FFNs) consume most non-embedding parameters. Motivated by theoretical and empirical evidences that residual wide-narrow-wide (hourglass) MLPs remain expressive despite bottlenecks, we revisit whether this architectural convention is necessary for dense language models. We study Hourglass Transformers, which replace the conventional FFN with residual stacks of hourglass sub-MLPs and use hourglass attention to decouple residual-stream width from attention width. This exposes a practical depth-width trade-off: compressing the FFN intermediate dimension allows wider hidden states and fewer layers at matched parameter budgets. Across model scales from 113M to 8B parameters, Hourglass Transformers achieve language-modeling and downstream performance comparable to conventional Transformers, while improving training compute efficiency by

8.7\%

at matched average downstream accuracy across the 906M, 3B, and 8B scales. After long-context extension, the 8B Hourglass model also outperforms its matched conventional baseline across 4k-64k context lengths. At 64k context, the reduced attention layer count lowers both computation and KV-cache requirements, yielding up to

1.93\times

faster token decoding and

50\%

lower KV-cache memory at the 1B scale. These results identify hourglass structures as a practical architecture-efficiency alternative for compute- and latency-conscious Transformer design.

Feng-Ting Liao, Guan-Ting Yi, Tzu-Quan Lin +2

MediaTek Research · National Taiwan University

Efficient Transformer Inference Transformer FFNs Transformer Efficient Language Model Training

Jan 21, 2026

Variance-Adaptive Muon: Pre-Orthogonalization Variance Modulation for Efficient Language Model Pretraining

Optimizer design plays a central role in efficient language model pretraining, directly affecting optimization dynamics, convergence speed, and compute cost under fixed training budgets. Muon has emerged as a strong optimizer by orthogonalizing momentum updates, yielding a matrix-valued analogue of sign-based normalization. However, unlike Adam-style methods, Muon does not explicitly incorporate gradient-variance information into its updates. Motivated by Adam's variance-adaptive interpretation, we propose Muon-NSR and Muon-VS, two variance-adaptive Muon variants for language model pretraining. Muon-NSR applies noise-to-signal ratio (NSR) modulation before Newton--Schulz orthogonalization, whereas Muon-VS uses variance scaling (VS) without introducing any additional hyperparameters beyond those of Muon. Both methods preserve Muon's spectral normalization structure while requiring only one additional variance buffer. Experiments on Llama-style and GPT-2 pretraining across model scales from 125M to 1.2B parameters show that our methods improve over well-tuned Muon baselines and remain competitive with representative adaptive Muon-family baselines. On Llama-1.2B, Muon-VS achieves a 1.33

\times

step-to-target speedup over a well-tuned Muon baseline, with Muon's final validation loss as the target. These results indicate that variance-adaptive modulation is a simple and effective mechanism for improving Muon-style optimizers in language model pretraining.

Jingru Li, Yibo Fan, Huan Li

College of Artificial Intelligence, Nankai University

Stochastic Optimization Language Model Pretraining Muon Optimizer Efficient Language Model Training