2026-10-02

Mixture-of-Experts (MoE) models enable parameter scaling with limited per-token computation by activating only a small subset of experts for each token, but deploying them still requires loading the complete expert pool into memory. Structured expert pruning can effectively reduce the memory usage by removing experts. However, existing pruning methods either use expert ranking criteria that are not well aligned with model performance or rely on effective expert subset searching that is computationally expensive. Moreover, these methods typically overlook the routing-behavior redundancy among the retained experts. In this paper, we propose MoE Pruning via Router Bias Learning and Expert Approximation (MoRA), a framework for structured MoE expert pruning. We introduce a learnable router bias for each expert and optimize these biases by minimizing the language-modeling loss and a routing-diversity regularizer. The learned router biases sharpen the routing probability distributions to identify experts critical to model performance while encouraging the selection of experts with diverse routing preferences. In addition, we introduce an expert approximation mechanism as a post-pruning enhancement. It leverages the remaining experts to approximate the outputs of pruned experts by affine transformation, further improving the performance of the pruned model. We evaluate MoRA on Qwen3-30B-A3B, DeepSeek-V2-Lite, and Moonlight-16B-A3B, removing 25\% and 50\% of the routed experts in each MoE layer. Extensive experiments on nine zero-shot benchmarks show that MoRA outperforms state-of-the-art pruning algorithms. Our code will be released.

Tabular foundation models (TFMs) achieve strong predictive performance through in-context learning, yet repeatedly conditioning on labeled data makes inference expensive. Knowledge distillation can reduce this cost by transferring their predictive ability to lightweight, dataset-specific students. However, the dependence of TFM predictions on both a labeled context and a query introduces two design questions: how to construct teacher supervision and whether expanding query coverage improves distillation. We examine these questions across two TFMs and both neural and tree-based students, and derive an effective distillation recipe. The recipe uses the full labeled training set as teacher context and trains students solely on teacher predictions for observed and synthetic queries. On TabArena, the resulting students outperform their supervised trained tuned-and-ensembled counterparts by 57-98 Elo points. Applied unchanged to TALENT, the same recipe improves matched default students on 236-258 of 300 datasets and reduces median primary error by 4.0-6.4%. The distilled students also achieve median inference speedups of 3.0-21.6 times over their teachers, offering a practical trade-off between predictive performance and repeated inference cost. Code is available at https://github.com/nums-ai/TFM_Distillation .

Wanda (Sun et al., 2024) prunes large language models by scoring weights independently within each linear projection, although queries and keys interact through dot products. We introduce QK-Wanda, which scores query and key weights by their individual deletion costs under an unmasked pre-RoPE reconstruction objective. It augments Wanda scores with information from the opposite projection (keys for query weights, and queries for key weights), allowing both projections to share a pruning budget. Its closed-form scores require no gradients or weight updates; full pruning takes 1.3% longer than Wanda on A100 and 3.1% longer on H200 with the calibration used in our main experiments. We evaluate QK-only pruning across 15 models from TinyLlama, Llama 2, Llama 3, and Qwen2.5, spanning 0.5B-72B parameters. Relative to Wanda, QK-Wanda reduces QK reconstruction error by an average of 60% at 50% sparsity and 45% at 80%. Downstream gains depend on the model. At 80% sparsity on Llama 2 70B, WikiText-2 and C4 perplexity decrease by 20.3% and 13.5%, while mean zero-shot accuracy rises by 5.94 percentage points. Qwen2.5-72B also improves, but Llama-3.1-70B has substantially higher perplexity despite lower reconstruction error. These results show both the promise of coupled pruning criteria and the limits of local reconstruction as a predictor of model quality.

KV cache pruning reduces long-context inference memory usage by evicting less important key-value pairs. KVzip estimates importance through context reconstruction: prompting a model to repeat the context chunk by chunk. This achieves strong compression quality at the cost of additional forward passes. Learned approximations reduce this cost but require model-specific training. We analyze how KVzip identifies important cached information and show how to approximate its reconstruction scores using information already computed during prefill. These findings motivate EchoPress, a training-free method that approximates reconstruction attention using queries and keys from standard prefill. For each request, it reconstructs only the first chunk to calibrate importance scores for the remaining context. Experiments on LongBench and RULER with Qwen3-8B and Llama-3.1-8B-Instruct show that EchoPress matches KVzip in task accuracy across eviction ratios from 50% to 90%, while reducing compression overhead by a factor of 1.7-19.6 and total prefill time by a factor of up to 2.9. Code is available at https://github.com/ljwljwljwljw/kvpress/tree/echo-press.

Scaling Transformers to long contexts is constrained by the quadratic cost of self-attention and the linear growth of key-value cache memory transfer. Sparse attention mitigates this by retrieving only relevant tokens, but current approaches either require large-scale training or, within the training-free regime, rely on semantically coarse heuristics or expensive clustering that is difficult to update efficiently during decoding. We introduce CommunityKV, a framework that formulates sparse attention as a community detection problem. CommunityKV constructs a token graph from the $QK^T$ scores already computed during standard prefill, and partitions the graph into communities to enable retrieval of semantically coherent token groups. A local update rule assigns newly generated tokens to communities in constant time, enabling sparse retrieval throughout streaming decoding without global re-partitioning. We evaluate CommunityKV on Qwen3 and Llama-3.1 models across three long-context benchmarks. With one graph per query head, CommunityKV delivers up to $1.25\times$ the end-to-end generation throughput of dense attention, while query-group graph aggregation yields up to $1.71\times$ with comparable accuracy.

Trellis-coded quantization enables high-dimensional compression of large language model (LLM) weights at ultra-low bit widths without the exponentially large codebooks required by conventional vector quantization. Practical deployment, however, presents two challenges: reconstructing compressed weights at sufficient parallel throughput to avoid making dequantization an inference bottleneck, and maintaining quantization accuracy without costly incoherence transformations. We address these challenges with two complementary techniques. First, we introduce an ultra-low-complexity trellis dequantizer that uses a structured, hardware-efficient state-to-value mapping while preserving diverse reconstruction choices for trellis search. Second, we reformulate discrete trellis path optimization with a curvature-aware objective that reflects model sensitivity directly in the original coordinate space. Together, these techniques enable high-quality ultra-low-bit trellis quantization with inexpensive, highly parallel runtime reconstruction and without relying on Hadamard-based incoherence processing.

The widespread adoption of Mixture-of-Experts (MoE) has created a growing need for deployment on heterogeneous platforms. However, it exposes a fundamental mismatch between the algorithmic demands of large-scale MoE and the disparate characteristics of hardware.Existing CPU-GPU hybrid inference systems fail to resolve this as they either encounter PCIe bandwidth bottlenecks when loading experts to GPUs, or rely heavily on CPU computation. Consequently, this leads to low resource utilization and inevitable violations of fixed latency budgets as parameters scale. In this paper, we identify and exploit Cross-Asymmetry--a structural alignment between the algorithmic workload skew of MoE routing and the physical disparity of heterogeneous hardware. To this end, we introduce RapidMoE, a residual offloading system for efficient large-scale MoE inference. We propose how RapidMoE leverages a residual-split framework to enable offloading paradigm shift from expert-level to bit-level, which unfolds across three key dimensions: (1) data representation, enabling compact and decoupled storage; (2) routing strategy, partitioning computation into dual paths aligned with hardware capabilities; (3) execution parallelism, scheduling a balanced storage-compute workload across devices. We further employ a novel Unified Multi-Level Importance Arbitration to adaptively adjust the critical expert set at runtime, ensuring the accuracy-latency Pareto frontier. These innovations exploit inherent cross-asymmetry, fundamentally breaking the algorithm-hardware misalignment. Experimental results show that RapidMoE achieves up to 3.5x speedup in decoding and 2.1x speedup in prefill compared to state-of-the-art (SOTA) offloading systems.