arXiv subjects

Jeffrey Quesnelle

Publications and source records attributed to Jeffrey Quesnelle.

9 recordsLinked to original sources

arXiv2026

Efficient Pre-Training with Token Superposition

Pre-training of Large Language Models is often prohibitively expensive and inefficient at scale, requiring complex and invasive modifications in order to achieve high data throughput. In this work, we present Token-Superposition Training (TST), a simple drop-in method that significantly improves the data throughput per FLOPs during pre-training without modifying the parallelism, optimizer, tokenizer, data, or model architecture. TST is done in two phases: (i) A highly efficient superposition phase where we combine many contiguous tokens into one bag and train using a multi-hot cross-entropy (MCE) objective, and (ii) a recovery phase where we revert back to standard training. We extensively evaluate TST on the scale of 270M and 600M parameters and validate on 3B and a 10B A1B mixture of experts model, demonstrating that it is highly robust in different settings. Ultimately, TST consistently outperforms baseline loss and downstream evaluations, and under equal-loss settings, TST yields up to a 2.5x reduction in total pre-training time at the 10B A1B scale.

cs.CL↗

arXiv2026

Decoupling the Benefits of Subword Tokenization for Language Model Training via Byte-level Simulation

Subword tokenization is an essential part of modern large language models (LLMs), yet its specific contributions to training efficiency and model performance remain poorly understood. In this work, we decouple the effects of subword tokenization by isolating them within a controlled byte-level pretraining pipeline. We formulate and test hypotheses across various dimensions, including sample throughput, vocabulary scaling, and the linguistic prior of subword boundaries. By simulating these effects in a byte-level setting, we refine our understanding of why subword models outperform raw byte models and offer insights to improve the pretraining of future byte-level and subword models. Specifically, our experiments highlight the critical role of increased training throughput and the integration of subword boundaries as either explicit priors or inductive biases.

cs.CL↗

arXiv2026

Long Context Pre-Training with Lighthouse Attention

Training causal transformers at extreme sequence lengths is bottlenecked by the quadratic time and memory of scaled dot-product attention (SDPA). In this work, we propose Lighthouse Attention, a training-only symmetrical selection-based hierarchical attention algorithm that wraps around ordinary SDPA and can be easily removed towards the end of the training. Our hierarchical selection is also gradient-free, which exempts us from dealing with a complicated and potentially inefficient backward pass kernel. Our contribution is three-fold: (i) A subquadratic hierarchical pre- and post-processing step that does adaptive compression and decompression of the sequence. (ii) A symmetrical compression strategy that pools queries, keys and values at the same time, while preserving left-to-right causality, which greatly improves parallelism. (iii) A two stage training approach which we pre-train for the majority of the time with Lighthouse Attention and recover a full attention model at the end with a short training. We run preliminary small scale LLM pre-training experiments that show the effectiveness of our method compared to full attention training with all other settings matched, where we achieve a faster total training time and lower final loss after the recovery phase. Full code is available at: https://github.com/ighoshsubho/lighthouse-attention

cs.CL↗

arXiv2026

YaRN: Efficient Context Window Extension of Large Language Models

Rotary Position Embeddings (RoPE) have been shown to effectively encode positional information in transformer-based language models. However, these models fail to generalize past the sequence length they were trained on. We present YaRN (Yet another RoPE extensioN method), a compute-efficient method to extend the context window of such models, requiring 10x less tokens and 2.5x less training steps than previous methods. Using YaRN, we show that LLaMA models can effectively utilize and extrapolate to context lengths much longer than their original pre-training would allow, while also surpassing previous the state-of-the-art at context window extension. In addition, we demonstrate that YaRN exhibits the capability to extrapolate beyond the limited context of a fine-tuning dataset. Code is available at https://github.com/jquesnelle/yarn

cs.CL↗

arXiv2026

DeMo: Decoupled Momentum Optimization

Scaling neural network training increasingly depends on synchronous data-parallelism, yet full-precision gradient all-reduce imposes a severe communication bottleneck. We propose Decoupled Momentum Optimization (DeMo), a drop-in replacement for any momentum-based optimizers that significantly reduces the communication bandwidth while maintaining convergence. DeMo (i) decouples local momentum updates, (ii) applies a fast orthonormal transform (e.g., DCT) followed by top-k sparsification, and (iii) reuses the momentum buffer as error feedback via momentum subtraction. This design reduces per-step communication by up to two orders of magnitude with minimal computational overhead. Experiments on 300M and 1B-parameter DeMo language models show DeMo transmits up to 85x less data per GPU than AdamW-DDP while achieving comparable loss and accuracy. DeMo is topology-agnostic and enables training across multi-datacenter or Ethernet-based setups. Code is available at https://github.com/bloc97/DeMo

cs.LG↗

arXiv2025

CurvaDion: Curvature-Adaptive Distributed Orthonormalization

As language models scale to trillions of parameters, distributed training across many GPUs becomes essential, yet gradient synchronization over high-bandwidth, low-latency networks remains a critical bottleneck. While recent methods like Dion reduce per-step communication through low-rank updates, they synchronize at every step regardless of the optimization landscape. We observe that synchronization requirements vary dramatically throughout training: workers naturally compute similar gradients in flat regions, making frequent synchronization redundant, while high-curvature regions require coordination to prevent divergence. We introduce CurvaDion, which uses Relative Maximum Momentum Change (RMMC) to detect high-curvature regions requiring synchronization. RMMC leverages momentum dynamics which are already computed during optimization as a computationally tractable proxy for directional curvature, adding only $\mathcal{O}(d)$ operations per layer. We establish theoretical connections between RMMC and loss curvature and demonstrate that CurvaDion achieves 99\% communication reduction while matching baseline convergence across models from 160M to 1.3B parameters.

cs.LG↗

arXiv2025

Hermes 4 Technical Report

We present Hermes 4, a family of hybrid reasoning models that combine structured, multi-turn reasoning with broad instruction-following ability. We describe the challenges encountered during data curation, synthesis, training, and evaluation, and outline the solutions employed to address these challenges at scale. We comprehensively evaluate across mathematical reasoning, coding, knowledge, comprehension, and alignment benchmarks, and we report both quantitative performance and qualitative behavioral analysis. To support open research, all model weights are published publicly at https://huggingface.co/collections/NousResearch/hermes-4-collection-68a731bfd452e20816725728

cs.AI↗

arXiv2024

Hermes 3 Technical Report

Instruct (or "chat") tuned models have become the primary way in which most people interact with large language models. As opposed to "base" or "foundation" models, instruct-tuned models are optimized to respond to imperative statements. We present Hermes 3, a neutrally-aligned generalist instruct and tool use model with strong reasoning and creative abilities. Its largest version, Hermes 3 405B, achieves state of the art performance among open weight models on several public benchmarks.

cs.CL↗

arXiv2017

On the linkability of Zcash transactions

Zcash is a fork of Bitcoin with optional anonymity features. While transparent transactions are fully linkable, shielded transactions use zero-knowledge proofs to obscure the parties and amounts of the transactions. First, we observe various metrics regarding the usage of shielded addresses. Moreover, we show that most coins sent to shielded addresses are later sent back to transparent addresses. We then search for round-trip transactions, where the same, or nearly the same number of coins are sent from a transparent address, to a shielded address, and back again to a transparent address. We argue that such behavior exhibits high linkability, especially when they occur nearby temporally. Using this heuristic our analysis matched 31.5% of all coins sent to shielded addresses.

cs.CR↗