arXiv:2608. 20061v1 Announce Type: cross Abstract: Mixture-of-Experts (MoE) architectures significantly expand model capacity without a proportional increase in computational cost.
arXiv:2608. 10605v1 Announce Type: cross Abstract: In large-scale pretraining, the algorithm, architecture, and systems decisions are conventionally made in disconnected stages.
arXiv:2608. 03457v1 Announce Type: new Abstract: Diffusion language models (dLLMs) offer an alternative to autoregressive (AR) language modeling, yet the scaling behavior of Mixture-of-Experts (MoE) dLLMs remains poorly understood.
The paper investigates why speculative decoding (SD) performs better than expected on sparse mixture-of-experts (MoE) models. By evaluating 10 large MoE models, the authors find that recent fine‑grained models achieve higher speedups at low batch sizes, contrary to prior predictions of a concave speedup‑batch relationship. They attribute this to lower verification costs driven by temporal overlap of draft tokens and GPU execution dynamics, and develop a maximum‑entropy IPF framework that accurately predicts unique expert counts from routing overlap, closely matching real‑model behavior.
arXiv:2606. 01155v1 Announce Type: cross Abstract: Scaling laws for dense LLMs under infinite data are well explored, but how sparsity interacts with limited data is not.
In large-scale pretraining, the algorithm, architecture, and systems decisions are conventionally made in disconnected stages. A scaling law stage selects an architecture and training recipe, optimizing loss under compute constraints, and a separate systems stage then optimizes the implementation for hardware efficiency.
The paper investigates how data repetition affects Mixture-of-Experts (MoE) language models compared to dense Transformers. Across models from 80 M to 1 B active parameters, MoEs degrade more quickly as data is repeated, with performance dropping significantly beyond 4× repetition and overtaking dense models only when strong regularization is applied. The study also identifies routing stabilization and expert specialization as key factors in MoE overfitting, and explores regularization techniques that can partially mitigate this issue.
arXiv:2609.40316v1 Announce Type: cross Abstract: Looped transformers and Mixture-of-Experts (MoE) offer complementary routes to efficient scaling: recurrence increases computational depth at fixed p...
arXiv:2512. 13996v3 Announce Type: replace Abstract: Sparse Mixture-of-Experts architectures are essential for scaling model capacity efficiently, yet the standard Top-$k$ routing imposes a rigid sparsity pattern that ignores the intrinsic variance in token difficulty and layer-specific computational needs.
The paper introduces S‑SALAAD, a framework that jointly learns model weights, low‑rank, and structured sparse components during training. It supports hardware‑compatible sparsity patterns and couples structure learning with budget‑aware compression and dedicated CPU/GPU kernels, allowing efficient inference from a single checkpoint across multiple memory budgets. Experiments on Llama2 and Qwen3 demonstrate improved perplexity, memory usage, and latency on various hardware platforms compared to baselines.
arXiv:2609.36222v1 Announce Type: new Abstract: Large language models are increasingly expensive to serve. In large-scale serving systems, autoregressive decoding is often bottlenecked by transferrin...
arXiv:2609.25655v1 Announce Type: new Abstract: As large language models (LLMs) scale rapidly, dense full-parameter adaptation becomes increasingly expensive, motivating sparse and modular architectu...