A fourth result ties the three together: leaving the routers in FP16 lowers drift by 20% while raising loss, so routing fidelity and output quality are separable objectives.
Abstract
Mixture-of-Experts (MoE) models route each token to a few of many expert networks, and that routing is data-dependent in a way standard inference optimizations do not expect. This paper measures what three of them actually deliver on OLMoE-1B-7B, DeepSeek-V2-Lite, and Qwen3-30B-A3B. Fused Triton kernels reach 5.6x to 9.0x in isolation but 0.999x end to end against a measured 1.07x ceiling, because the model spends its time waiting on roughly a thousand kernel launches per forward pass rather than on the arithmetic those kernels improve. INT4 quantization changes on average 0.53 of the eight selected experts per token position, yet replaying exactly those changed routes through full-precision weights reproduces only 2.7% of the quality loss, which makes the experts substitutable rather than specialized. Removing all 23 graph breaks from PyTorch's compiler, the step prior work treats as the structural fix, makes the model three times slower. A fourth result ties the three together: leaving the routers in FP16 lowers drift by 20% while raising loss, so routing fidelity and output quality are separable objectives. Every number recomputes from committed per-token route dumps.
Expert pruning reduces the memory and serving cost of Mixture-of-Experts (MoE) models by removing low-importance experts identified by the router, assuming router probabilities provide a reliable importance signal. We observe that this assumption breaks down under over-dispersed routing, a regime associated with aggressive load-balancing during training, in which tokens are distributed nearly uniformly across experts and importance signals collapse. In this regime, perplexity does not predict downstream task accuracy: on gpt-oss-20B, the lowest-perplexity pruning configuration yields the worst mathematical reasoning, while the highest-perplexity configuration preserves it. This does not occur under standard routing (e.g., Mixtral-8x7B-Instruct), where perplexity and accuracy degrade together. Pruning under over-dispersed routing also exposes a capability trade-off in which no single scoring metric dominates: activation-aware scoring preserves mathematical reasoning but severely degrades knowledge-intensive science (an 18-point gap on GPQA), whereas frequency-based scoring exhibits the reverse. We propose Minimax Expert Score Allocation (MESA), a domain-aware method that iteratively boosts importance scores for experts serving whichever domain is currently worst-affected, minimizing worst-case domain degradation rather than average accuracy. At 25% expert pruning MESA achieves the smallest worst-case degradation across domains, outperforming activation-aware baselines on 7 of 11 benchmarks at a correspondingly reduced memory footprint, and it generalizes to gpt-oss-120B, Gemma-4-26B-A4B, and OLMoE-1B-7B. Our results indicate that over-dispersed routing is a qualitatively distinct pruning regime in which standard assumptions fail, and that recognizing it is a prerequisite for principled expert pruning of load-balanced MoE models.
Berkcan Kapusuzoglu, Connor Pryor, Sangwoo Cho et al.· 0 citations
ExactMoE, an inference design that applies symmetric group-128 four-bit weight quantization only to routed experts, stores those experts in kernel-native MARLIN form in pinned host memory, and executes all selected experts through a configurable GPU-resident slot cache and fused grouped MoE kernels, identifies a practical memory-transfer-throughput frontier for complete-expert MoE inference.
Sparse Mixture-of-Experts (MoE) routers commonly use the same scores both to select experts and to weight their already-computed outputs. We study whether these two roles, dispatch and aggregation, should be coupled. On pretrained OLMoE-1B-7B, we keep selected Top-8 expert IDs, expert computation, and total selected router mass fixed and change only within-set aggregation. A structured oracle improves full-horizon cross-entropy by 0.0160 +/- 0.0039 across three seeds; the router's top-scored expert is the counterfactual-best vertex only 17.2% of the time, with router-utility Spearman 0.030. We therefore train Fixed-Dispatch Adaptive Aggregation (FDAA), a 301K-parameter post-compute head optimized directly with the language-modeling objective while freezing the backbone, router, and experts. On OLMoE, FDAA improves fresh WikiText-103 test by Delta CE = -0.1523 +/- 0.0031 across three seeds, and mixed-domain training gives robust gains on WikiText-103, C4, and held-out Penn Treebank under frozen confirmatory evaluation. We also replicate the fixed-dispatch audit on DeepSeek-V2-Lite, which uses Top-6 routed experts plus shared experts. Best-vertex headroom remains significant on WikiText and C4, while router Top1 identifies the best selected expert in only 12.5% and 16.7% of audited examples. In a one-seed mixed-domain replication, FDAA improves locked WikiText and PTB, while C4 is statistically neutral. These results support a cross-architecture distinction between expert selection and expert commitment.
Mixture-of-Experts (MoE) models decouple total parameters from per-token compute, but deployment still requires storing every expert. Recent theory shows that pruning experts with the smallest router-norm changes during fine-tuning can preserve accuracy, but assumes full fine-tuning. We test whether lightweight adaptation can recover this signal. We briefly fine-tune with a parameter-efficient adapter, rank experts by the induced $\ell_2$ router change, and prune the least-changed experts in one shot. On Mixtral-8$\times$7B-Instruct (44.83% MMLU-Pro), router-only LoRA trains 0.002% of parameters and outperforms all-module LoRA at matched rank with half the experts removed (27.54% vs. 24.42%); signal quality declines as adaptation spreads to attention and expert weights. Accuracy improves monotonically with LoRA rank, reaching 28.76%. IA3, which leaves router weights frozen, matches direct router adaptation, whereas unconstrained additive adapters degrade the signal. Router-guided MMLU-Pro accuracy decays quasi-linearly rather than collapsing, remains nearly 1.8 times that of magnitude-based or random pruning at maximal compression, and reduces memory by 49% and per-token latency by 37%. At 25% compression, retention is competitive with methods using full activation statistics. The criterion also transfers to Qwen1.5-MoE fine-tuned for mathematics, retaining 49.7% mean accuracy over eleven benchmarks with half the experts removed while random pruning falls to single digits. Router sensitivity under lightweight fine-tuning therefore makes provably motivated expert pruning practical at scale.
Ali Janati, Kaoutar El Maghraoui, Xinyi Luo et al.· 0 citations
The design of transformer-based Large Language Models (LLMs) is being radically changed through new architectures that are able to overcome scalability limitations of previous designs, including Mixture-of-Experts (MoE), Multi-Head Latent Attention (MLA), and Multi-Token Prediction (MTP). As an open-weighted model released at the end of 2024, which has both state of the art architectural transparency and production scale efficiency, DeepSeeek-V3 represents the ultimate testing ground for investigating these modern technologies. This paper provides a comprehensive analysis of the architectural structure of DeepSeek-V3 based upon information from the DeepSeek-V3 Technical Report, industry benchmarking data and independent latency testing, to demonstrate how various techniques can be used to optimize training while still providing competitive performance in code generation and mathematical reasoning. In addition, latency testing conducted on a Distilled version of DeepSeek-V3, with approximately 14 billion parameters, running on a T4 GPU, reveals that although significant improvements have been made in optimizing latency there remains substantial barriers to deploying these models. Through this context, this research will serve as a reference document for practitioners and researchers who wish to understand current trends and challenges in increasing accessibility to high performance AI models.
Yassine Zouhdi, B. Hdioud· EPJ Web of Conferences· 0 citations
Modern fine-grained Mixture-of-Experts (MoE) models route each token to a small number of experts and renormalize their router probabilities. We show that this renormalization implicitly calibrates expert output gain to the training top-$k$: reducing $k$ at inference changes not only which experts are used but also the strength of the expert branch. We separate these effects by activating the top $k_1$ experts while normalizing by the probability mass of the top $k_2$ experts, introducing one integer with no parameters, training, or measurable compute overhead. On Qwen3.6-35B-A3B, reducing from 8 to 4 experts causes a 4.65-point MMLU drop under standard renormalization but only 0.35 points with $k_2=16$, while halving routed-expert compute. The result replicates on the $11\times$ larger Qwen3.5-397B-A17B, where reducing from 10 to 5 experts loses only 0.55 points with an appropriate reference set. Removing renormalization entirely is catastrophic, showing that preserving a suitable reference mass is crucial. We further find that perplexity and downstream accuracy favor different $k_2$, cautioning against selecting MoE compression settings using unlabeled text alone. Analyses also show that expert identity matters substantially more than expert weighting, while balanced and domain-specialized routing leaves limited room for expert pruning.
Xing Chen, Hengshuai Yao· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
