MXSens: Sensitivity-Aware Mixed-Precision Quantization for Efficient LLM Inference

4-bit quantization enables efficient LLM inference, but suffers from significant accuracy degradation due to outliers. Prior work addresses this problem via data rotation or mixed-precision integer quantization, but often relies on software-managed scaling and frequent dequantization, incurring substantial overhead. Microscaling formats, such as MXINT, eliminate these inefficiencies by encoding scales in hardware, yet remain incompatible with rotation-based methods. Our analysis reveals that outliers vary in severity, from rare extremes to frequent mild deviations, and that quantization sensitivity is unevenly distributed across layers and columns. These insights motivate a fine-grained, sensitivity-guided approach. We introduce MXSens, a training-free method that assigns mixed mantissa bitwidths (4/6/8) based on column- and layer-wise sensitivity, naturally leveraging the block-wise structure of MXINT. MXSens outperforms state-of-the-art quantization methods across a range of models and tasks. Under the W4A4KV4 setting, MXSens achieves perplexities of 3.77 and 7.63 on LLaMA-2-70B and LLaMA-3-8B, respectively, substantially improving over existing baselines on WikiText-2. Our work establishes a new balance between accuracy and resource efficiency for LLM quantization.

Stabilizing Extrapolation in Looped Transformers via Learned Stochastic Stopping

Looped Transformers, which repeatedly apply a shared transformer block, are an architecturally natural fit for variable-length algorithmic tasks. Although they can exhibit strong length generalization beyond the length of training sequences, this behavior is brittle, yielding high out-of-distribution (OOD) variance, even across well-performing in-distribution solutions. We trace this variance to the spurious correlation in simple algorithmic tasks between sequence length and number of loops. Introducing stochasticity into the number of loops during training sharply reduces OOD variance and stabilizes predictions across inference-time loop counts. To improve upon heuristic randomization schemes, we further analyze RL-Halting as a learned stochastic schedule and find that it generally improves the accuracy-stability trade-off. Across binary addition, Dyck-1, Unique Set, and Copy, learned stochastic stopping often improves this trade-off but can also stabilize a suboptimal computation. Our work suggests that "when to stop" should be treated as a training-time design choice, not merely an inference-time computation-allocation rule.

Improving Neural Network Training by Decoupling the Magnitude and Direction of Weight Vectors

Modern neural network training relies on optimizers such as Adam and Muon which act on each weight matrix as a single object. Yet every weight matrix carries two distinct quantities -- a \emph{magnitude} and a \emph{direction} -- and all optimizers stepping in the matrix as a whole couple their dynamics: the directional change from an update depends on the current magnitude, while the magnitude drifts as a byproduct of learning the direction, so neither is governed directly by the learning rate. Typical training therefore leans on surrounding recipes such as weight decay and warmup to keep learning stable at scale, though these regulate the coupling only indirectly; other recent methods instead constrain the weight to a fixed-norm sphere, but add no learnable magnitude, leaving scale control to normalization layers alone. We propose \emph{Magnitude--Direction (MD) Decoupling}, an optimizer modification that factorizes each weight into a fixed-norm direction on a hypersphere and learnable per-row and per-column magnitude gains, updated at separate learning rates, all while the model still sees a single fused weight tensor. The method is agnostic to the base optimizer and removes the need for weight decay and warmup. Across both Adam and Muon, MD Decoupling improves on well-tuned baselines, transfers the optimal LR across model width without retuning, and continues to help at scale on large Mixture-of-Experts (MoE) models. Treating magnitude and direction as separately controlled quantities thus yields more predictable training dynamics and a simple, broadly applicable improvement to modern optimizers.

Tying the Loop -- Tied Expert Layers in Mixture-of-Experts Language Models

Mixture-of-Experts (MoE) architectures efficiently scale Large Language Models (LLMs) by activating only a small fraction of their experts per token, yet the full parameter count - dominated by the expert parameters - must be held in training and inference memory. To address this, we introduce Expert Tying, an architectural modification that shares expert parameters across consecutive transformer layers while preserving independent, layer-wise routing and attention. We evaluate this approach across common, state-of-the-art architectures, including OLMoE, Qwen3, and DeepSeek-style MoEs. Our pretraining experiments demonstrate that tying experts can reduce memory footprint by almost 2x at virtually no degradation in perplexity or downstream quality. By exploiting the parameter redundancy inherent in MoE pathways, our method provides a highly favorable compute-to-memory trade-off, advancing efficient training and scaling of next-generation LLMs.

Local MixVR: Breaking the Communication-Sample Dependence in Distributed Learning

Communication overhead is a crucial bottleneck in scalable distributed learning. While existing methods aim to efficiently utilize data points, such as Local SGD, Minibatch SGD, and their accelerated variants, they still exhibit communication-round complexity that scales with the total number of samples $N$. In this paper, we introduce Local MixVR, a distributed framework that integrates local updates with variance-reduction techniques to mitigate local noise. We show that Local MixVR is the first distributed method to eliminate the dependence of communication complexity on $N$, achieving a complexity that scales only with the number of workers $M$. In common regimes where $M<O\left(N^{1/4}\right)$, Local MixVR outperforms the state-of-the-art Minibatch Accelerated SGD baseline, bridging a long-standing gap in distributed optimization and establishing a new paradigm for communication-efficient training.

Apertus LLM Family Expansion via Distillation and Quantization

The wide adoption of LLMs has led to their use in great variety of applications and scenarios, such as chatbot assistants and data annotation, creating the need for the models to satisfy certain budget and hardware constraints. This has led to the trend of LLMs being released in batches consisting of similar models of various sizes for the family of models to adhere to as wide of a range of constraints as possible. In this paper, we validate distillation and quantization as a cost-effective way to expand model families to new sizes and hardware formats. Based on the open-recipe Apertus 8B LLM, we produce Apertus-v1.1 - a distilled family of models with up to 4B parameters trained on 1.7T permissive license tokens. We demonstrate cost-efficiency and strong accuracy performance of our approach for covering large ranges of hardware and systems requirements.

Toward Cross-Lingual Quality Classifiers for Multilingual Pretraining Data Selection

As Large Language Models (LLMs) scale, data curation has shifted from maximizing volume to optimizing the signal-to-noise ratio by performing quality filtering. However, for many languages, native high quality data is insufficient to train robust quality classifiers. This work investigates the idea that quality markers in embedding space may show cross-lingual consistency, which would allow high-resource languages to subsidize the filtering of low-resource ones. We evaluate various filtering strategies, including cross-lingual transfer, third quartile sampling (Q3), and retention rate tuning. Our results demonstrate that massive multilingual pooling frequently outperforms monolingual baselines in both rank stability and aggregate accuracy for a 1B model trained on 103B tokens, delivering gains for high resource languages (1.2% increase in aggregate normalized accuracy for French) and matching or exceeding monolingual baselines for low-resource languages. However, we find that scale alone does not guarantee stability. Furthermore, for high-resource languages like French, we show that refining the decision boundary through third quartile sampling (Q3) or tuning the retention rate is necessary to fully leverage the multilingual signal.

Weight Decay may matter more than muP for Learning Rate Transfer in Practice

Transferring the optimal learning rate from small to large neural networks can enable efficient training at scales where hyperparameter tuning is otherwise prohibitively expensive. To this end, the Maximal Update Parameterization (muP) proposes a learning rate scaling designed to keep the update dynamics of internal representations stable across different model widths. However, the scaling rules of muP rely on strong assumptions, particularly about the geometric alignment of a layer's inputs with both its weights and gradient updates. In this large-scale empirical investigation, we show that these assumptions hold only briefly at the start of training in the practical setups where learning rate transfer is most valuable, such as LLM training. For the remainder of training it is weight decay rather than muP that correctly stabilizes the update dynamics of internal representations across widths, facilitating learning rate transfer. This suggests muP's scaling primarily acts as a form of implicit learning rate warmup, allowing us to largely replace it with modified warmup schedules. Together these findings fundamentally challenge prevailing beliefs about learning rate transfer and can explain empirical practice such as why muP requires the independent weight decay variant for successful transfer.

Apertus: Democratizing Open and Compliant LLMs for Global Language Environments

We present Apertus, a fully open suite of large language models (LLMs) designed to address two systemic shortcomings in today's open model ecosystem: data compliance and multilingual representation. Unlike many prior models that release weights without reproducible data pipelines or regard for content-owner rights, Apertus models are pretrained exclusively on openly available data, retroactively respecting robots.txt exclusions and filtering for non-permissive, toxic, and personally identifiable content. To mitigate risks of memorization, we adopt the Goldfish objective during pretraining, strongly suppressing verbatim recall of data while retaining downstream task performance. The Apertus models also expand multilingual coverage, training on 15T tokens from over 1800 languages, with ~40% of pretraining data allocated to non-English content. Released at 8B and 70B scales, Apertus approaches state-of-the-art results among fully open models on multilingual benchmarks, rivalling or surpassing open-weight counterparts. Beyond model weights, we release all scientific artifacts from our development cycle with a permissive license, including data preparation scripts, checkpoints, evaluation suites, and training code, enabling transparent audit and extension.

TiMoE: Time-Aware Mixture of Language Experts

Large language models (LLMs) are typically trained on fixed snapshots of the web, which means that their knowledge becomes stale and their predictions risk temporal leakage: relying on information that lies in the future relative to a query. We tackle this problem by pre-training from scratch a set of GPT-style experts on disjoint two-year slices of a 2013-2024 corpus and combining them through TiMoE, a Time-aware Mixture of Language Experts. At inference time, TiMoE masks all experts whose training window ends after the query timestamp and merges the remaining log-probabilities in a shared space, guaranteeing strict causal validity while retaining the breadth of multi-period knowledge. We also release TSQA, a 10k-question benchmark whose alternatives are explicitly labelled as past, future or irrelevant, allowing fine-grained measurement of temporal hallucinations. Experiments on eight standard NLP tasks plus TSQA show that a co-adapted TiMoE variant matches or exceeds the best single-period expert and cuts future-knowledge errors by up to 15%. Our results demonstrate that modular, time-segmented pre-training paired with causal routing is a simple yet effective path toward LLMs that stay chronologically grounded without sacrificing general performance much. We open source our code at TiMoE (Github): https://github.com/epfml/TiMoE