In fine-grained sparse Mixture-of-Experts (MoE) models, a large pool of specialized experts replaces a small homogeneous set, shifting performance and throughput to be governed by inference-time expert activation. Yet most existing optimization recipes implicitly assume a fixed activation budget (e.g., a constant Top-k per layer), whose behavior in fine-grained MoEs is poorly understood. We first characterize runtime skipping strategies, quantifying the accuracy–efficiency trade-off of (i) uniform fixed activation and (ii) static layer-wise Top-k allocation found by search. Our analysis reveals that static skipping can already provide substantial throughput gains, but optimal static schedules vary significantly across models and routing mechanisms. We therefore introduce Adaptive Skipping with Entropy-Penalized Thresholding (ASET), a training-free policy that adapts token-level activation using router confidence and entropy while remaining within the model’s original budget. Across the fine-grained MoEs we study, static skipping policies yield 10–78% throughput gains with minimal performance degradation, including ≥10% improvement on DeepSeek-V3 without measurable loss. On the OLMoE testbed, ASET yields a Pareto frontier between average activation and task quality. Overall, these results identify expert skipping as a practical lever for faster fine-grained MoE inference, with adaptive activation helping when fixed budgets are too rigid.

Post-training quantization facilitates efficient LLM deployment, yet it disproportionately erodes reasoning and mathematical performance. To recover fidelity without costly fine-tuning, we identify a critical vulnerability in existing output-space methods: the tendency to overfit the limited calibration manifold, which necessitates structural regularization for unseen data. We propose **ROCOB** (**R**esilient **O**utput **C**ompensation with **O**ptimal **B**udgeting), a training-free framework based on *Dual-Manifold Alignment*. ROCOB couples a Principal Activation Component (PAC), which minimizes error on the calibration manifold, with an Orthogonal Geometric Component (OGC), which recovers residual weight-space structure without fitting a second calibration objective. To compare methods fairly under the same added-parameter budget, ROCOB uses Global Spectral Energy Maximization (GSEM) to allocate ranks according to recovered spectral energy and module cost. Across Qwen3 and Llama-3 models, quantizers, and reasoning benchmarks, ROCOB improves average post-quantization recovery in most evaluated model–quantizer settings while preserving a training-free construction pipeline.

Large Language Models (LLMs) have demonstrated remarkable capabilities across diverse natural language processing tasks. However, their extensive memory requirements, particularly due to KV cache growth during long-text understanding and generation, present significant challenges for deployment in resource-constrained environments. Quantization has emerged as a promising solution to reduce memory consumption while preserving historical information. We propose XQuant, a training-free and plug-and-play framework that achieves ultra-low equivalent bit-width KV cache quantization. XQuant introduces two key innovations: a computationally negligible data-free calibration method and cross-layer KV cache compression, enabling quantization to sub-1.4 bits. Extensive experiments on TruthfulQA and LongBench demonstrate that XQuant outperforms state-of-the-art methods (e.g., KIVI-2bit and AsymKV-1.5bit) by achieving lower bit-width while maintaining superior performance, establishing a better trade-off between memory efficiency and model accuracy. The source code is available at https://github.com/brinenick511/XQuant.