Overview

Token pruning studies how to shorten sequences processed by transformers while retaining information needed for the task, from visual and audio inputs to long-context and diffusion language models. Methods rank tokens using attention, query relevance, entropy, representation changes across layers, or temporal and spatial structure; they then drop, merge, cache, or selectively recompute tokens, often with adaptive budgets. Work evaluates these approaches on image and video understanding, speech, generation, reasoning, and embodied control, using measures such as accuracy, fidelity, latency, throughput, and memory. The central tension is that aggressive pruning can discard evidence needed later, while scoring and selection overhead may reduce real speedups; reliably predicting which tokens remain important across tasks and layers is still an open challenge.

Latest papers

57 papers6 papers in the last four weeks, down 14% on the four weeks before. 0.0% of all new papers.

Nov 5, 2026

Training-Free Predictive Spatio-Temporal Token Pruning for Few-Step Autoregressive Video Diffusion

Few-step autoregressive (AR) video diffusion models generate videos chunk by chunk and have made efficient video generation practical. However, further accelerating these models toward real-time generation remains challenging, as few-step distillation leaves little cross-step redundancy for existing training-free caching methods to exploit. We revisit this regime and identify two token-level redundancy patterns: cross-frame redundancy from static content shared across adjacent frames, and cross-step redundancy from tokens that converge early during denoising. Crucially, we find that early clean estimates along the model's own denoising trajectory provide an effective online signal for locating both patterns, requiring no retraining, auxiliary predictor, or reference video. Based on these findings, we propose PreSTP (Predictive Spatio-Temporal Token Pruning), a training-free framework that constructs a unified token-wise redundancy mask from cross-frame similarity and cross-step convergence. Tokens marked as redundant skip subsequent computation under a consistent last-computed-value restoration rule, and a content-adaptive mask reuse strategy further reduces dense probing cost. On 5-second generation, PreSTP achieves 1.52x-3.20x speedups and a better efficiency-fidelity trade-off than coarse-grained caching baselines such as TeaCache and FlowCache. On 30-second generation, it reaches up to 1.96x speedup while improving VBench-Long and suppressing long-horizon drift. Notably, it continues to reduce inference cost even in the extreme one-step regime, where cross-step redundancy vanishes entirely and coarse step-wise caching is no longer applicable.
Nov 3, 2026

Exploiting Biases in Attention for Transformers

Attention is known to be sparse in vision transformers. In particular, we analyze the unequal relevance of attention-matrix columns to the final computation and show that this phenomenon becomes increasingly pronounced with network depth. This can be interpreted as saying that tokens have an absolute importance for the attention process, independently of queries. Inspired by this, we propose a reformulation of the attention formula that makes this phenomenon explicit by adding an input-dependent bias to the query-key dot product. We show that this new term gives each token a scalar value indicating its anticipated influence on other tokens during the attention process, and can thus serve as a proxy for pruning attention maps. Across multiple transformer model sizes, this allows us to substantially reduce the computational cost of attention, in terms of both FLOPs and inference latency, while retaining most of the original classification accuracy. We publicly released the code and models at https://github.com/EIDOSLAB/AttentionWithBias.
Oct 27, 2026

3D-SceneQ: Empowering 3D LLM With Query-Guided Adaptive Pruning and Multi-Modal Feature Enhancement

Current 3D scene understanding pipelines typically concatenate vast numbers of 3D object tokens with text tokens and feed the resulting sequence to a Large Language Model (LLM). However, existing 3D-LLMs for scene understanding encounter two major limitations: (i) excessive inclusion of task-irrelevant object data, introducing noise that reduces reasoning accuracy and increases hallucinations; and (ii) reliance solely on point cloud data, which inherently lacks the rich semantic information available in complementary 2D modalities, such as color, material properties, texture, and high-level contextual relationships. To address these challenges, we introduce 3D-SceneQ, a 3D LLM that unifies adaptive token pruning with multimodal semantic enrichment, markedly advancing scene understanding, reasoning, and grounding. Specifically, we propose a Query-Guided Adaptive Pruning (QGAP) module that adaptively selects task-relevant objects based on the intended meaning of user instructions, leveraging a learnable latent query to retain critical information while effectively suppressing irrelevant noise. In addition, we introduce a Multi-modal Object-level Feature Enhancement (MOFE) module, which integrates 2D feature embeddings from pre-trained models into 3D representation to enrich object-level semantic information. Evaluations show that 3D-SceneQ outperforms current state-of-the-art methods across multiple benchmarks (e.g., +5.7 CIDEr on Scan2Cap and +17.0 EM@1 on SQA3D), demonstrating its remarkable capabilities in various 3D language reasoning tasks.
Oct 19, 2026

TokenGate: Adapting LLMs for Fast and Memory-Efficient Prefill with Learned Token Pruning

Expert human readers process large inputs (e.g., code repositories and long documents) by skimming to closely read only what matters. On the other hand, language models process all input tokens through each model layer, resulting in significant processing costs for long-context inputs. We introduce TokenGate, a small, drop-in component that extends existing trained models to enable end-to-end learnable token pruning over long-context inputs for fast and memory-efficient prefill. To enable end-to-end learning despite the non-differentiable top-K operator for token pruning, we introduce token-gated attention, a mechanism inspired by Mixture-of-Experts that re-weights attention logits using token importance scores. Across diverse tasks, TokenGate achieves 1.41.4-11.3×11.3\times faster time-to-first token (TTFT) and 1.51.5-3.4×3.4\times lower peak memory at no accuracy degradation for prefill in transformers. Compared to the best-performing prior methods for token-pruning, TokenGate gives up to 6.4×6.4\times faster prefill and 2×2\times lower peak memory. TokenGate natively extends both standard transformers (e.g., Qwen3) and hybrid-attention models (e.g., Qwen3.5) with simple lightweight adaptation comparable to standard parameter-efficient fine-tuning. We make our code available at: https://github.com/tokengate-models/TokenGate.
Oct 19, 2026

AudioScorer: Learning Question-Aware Token Relevance for Large Audio Language Models

Large Audio Language Models (LALMs) reason over speech, environmental sounds, music, and other acoustic information. However, their audio encoders produce dense token sequences that make long-form audio costly to process. Existing audio pruning methods reduce this cost using heuristics such as attention, representation similarity, or temporal sampling, but do not explicitly learn which tokens are most useful for a given question. We propose AudioScorer, a lightweight model that scores audio tokens according to the user question. It is trained by distilling the ranking of audio tokens under the frozen LALM's answer-gradient saliency, together with downstream answer supervision. Compared with other pruning methods, AudioScorer achieves the best or tied-best result at 10% token retention on 7 of 10 semantic and acoustic tasks, including speech recognition, emotion recognition, and audio-scene classification.
Oct 15, 2026

SATOR: Backpropagation-Free 3D Test-Time Adaptation via Sample-Adaptive Token Routing

Test-time adaptation (TTA) mitigates the performance degradation of source-pretrained models on unlabeled test data under distribution shifts. Recent backpropagation-free token-purging methods address 3D point cloud distribution shifts by filtering unreliable tokens. However, their reliance on predefined purge-size candidates requires multiple forward passes per sample and increases inference cost. We propose SATOR (Sample-Adaptive TOken Routing), a source-free and backpropagation-free 3D TTA framework that eliminates this search by deriving both the partitioning threshold and routing decision directly from each sample. SATOR computes token abnormality scores from complementary attention and geometric relation graphs. It then derives a sample-specific partition into retained and pruned tokens via Otsu-based separability, without evaluating candidate sizes. Keep-ratio-calibrated routing then selects the pruned prediction only when its entropy, penalized by the keep ratio, is lower than that of the original prediction. Experiments on ModelNet40-C, ScanObjectNN-C, and ShapeNet-C demonstrate that SATOR achieves state-of-the-art accuracy on three 3D corruption benchmarks.
Oct 9, 2026

AudioTETRIS: Token Elimination via Transitional Importance Signals for Efficient Audio LLMs

Large Audio Language Models (LALMs) achieve strong performance on speech-text tasks, but standard audio encoders produce hundreds of tokens for even short audio clips. Because self-attention scales quadratically in sequence length, this inflates inference-time compute substantially. Current training-free token compression methods prune or merge tokens based solely on coarse attention or feature similarity. However, treating audio tokens as generic, isotropic features smooths over spectral transitions, destroying fine-grained phonetic transitions essential for decoding speech. In contrast, speech signals exhibit highly structured temporal patterns where information density peaks at acoustic transitions. We propose AudioTETRIS, a training-free, two-stage framework that uses the audio encoder's representation geometry as an acoustic prior for token importance before applying attention-based compression. Instead of discarding encoder structure after the encoding stage, AudioTETRIS propagates this prior directly into the LLM decoder to compute a per-token fused score. This score drives progressive top-kk token selection across decoder layers without requiring additional training or auxiliary models. Evaluated on Qwen2-Audio-7B, Phi-4-Multimodal, and Audio Flamingo 3, AudioTETRIS retains near-uncompressed performance on several speech tasks using only 25% of audio tokens, matches or surpasses full-token accuracy at moderate compression rates, and consistently outperforms existing training-free baselines.
Oct 7, 2026

Sparse Planning in Visual World Models via Cost Gradients

Token-based world models enable fine-grained latent planning, but repeatedly processing large spatial token grids makes action search expensive. We introduce COSTGRAD, a training-free, goal-conditioned selector that ranks spatial tokens by the gradient norm of the planning cost with respect to each input token. By deriving importance from the downstream control objective, COSTGRAD targets tokens that matter for planning rather than merely for prediction. On AdaLN-conditioned predictors at 50%50\% sparsity, COSTGRAD matches or exceeds full-token planning on three of four continuous-control benchmarks, while giving a measured 2.6×2.6\times wall-clock speedup per environment planning step. Combining token sparsity with reduced CEM search increases this to a ∼5×\sim 5\times total speedup while still exceeding the full-token baseline. We also identify an architecture-dependent failure mode: in a matched AdaLN-vs-concat comparison, concat maintains comparable full-token performance but pure COSTGRAD loses its advantage over random selection. This difference tracks action-pathway drift: gradient-selected removal produces less drift than random removal on AdaLN, but more on concat. These results highlight selector-architecture compatibility as a design axis for sparse world-model planning. Project page and demos: https://ycxuyingchen.github.io/costgrad/
Oct 7, 2026

EntroPrefill: Renyi-Guided Context Pruning with Conditional Stability Guarantees for Retrieval-Augmented Generation

Mid-prefill pruning can reduce the sequence processed by deeper transformer layers, but attention concentration alone does not certify that discarded context is dispensable. We formulate EntroPrefill as a Renyi-guided proposal mechanism coupled to explicit constraints on discarded attention mass. Sink-isolated, regularized head pooling respects grouped-query attention while exposing a quantitative trade-off between specialization and worst-head coverage. We derive a mixture-to-head deletion envelope, a computable upper bound on feasible token removal, and a finite-sample observer guarantee that remains valid when the pruning layer is selected adaptively. We then establish a conditional transformer perturbation bound with explicit sufficient Lipschitz constants and a first-token decision-margin corollary. A counterexample shows why shallow observations alone cannot imply an unconditional future-output guarantee. The systems analysis distinguishes query-head unions, physical page allocation, and KV-transfer payload, and gives an arithmetic break-even condition for pruning. This manuscript is theoretical in scope: it defines the procedure, its assumptions, and its formal limits, but does not report measured acceleration or task-accuracy preservation. Experiments are reserved for subsequent validation of the assumptions, approximation tightness, and end-to-end resource trade-offs.
Oct 6, 2026

DIPrune: Task-Aware Token Pruning with Dual Importance for Efficient Multimodal Language Models

Recent training-free pruning approaches for Multimodal Large Language Models (MLLMs) effectively cut computational overhead by exploiting visual redundancy or text-vision attention. However, they frequently suffer from semantic degradation due to their task-agnostic design or unreliable attention estimates. Based on our empirical analysis, we have found that this issue arises because salient tokens in shallow layers persistently suppress emerging semantic ones through numerical inertia, leading to premature discarding of signals crucial for deep reasoning. To address the aforementioned issue, from the task-oriented aspects, we first reformulate training-free pruning as a minimization of the distortion in the final task loss and derive a tractable, token-wise upper bound to serve as a surrogate objective. Specifically, this formulation inherently reveals a previously neglected inter-layer term that accounts for gradients across layers. Accordingly, for the implementation, we propose DIPrune, a rank-based framework that employs a dual importance scoring mechanism to jointly optimize intra-layer static feature saliency and inter-layer dynamic semantic evolution. Extensive experiments on LLaVA and Qwen-VL demonstrate that DIPrune consistently achieves state-of-the-art results.
Oct 6, 2026

Later Is Better: Token Reduction for ViTs Under Distribution Shift

Training-free token reduction accelerates vision transformers by removing redundant tokens across layers, recovering most of the original accuracy at a fraction of the compute. These methods, however, are designed and evaluated primarily on clean data, and under real-world distribution shift their accuracy gap to the uncompressed model widens with the removal rate. We show that this gap is governed by the reduction schedule, the depth profile of removal, usually left fixed as an implementation detail. Concretely, we introduce a one-parameter late-concentrated power-law schedule that consistently improves out-of-distribution accuracy over flat at no extra inference cost. On ImageNet-C with DeiT-S, the late schedule closes 83% of that gap at a 26% compute reduction (+1.17pp), and 99% of it at a lighter 7% reduction (+0.26pp). The gain cannot be attributed to retaining more tokens or using extra compute: held to flat's compute, the late schedule removes more tokens in total and leaves fewer tokens at the end, yet still wins. Single-layer probes point to a mechanism: earlier reductions perturb features that pass through more remaining layers, front-loading reduction error in depth. The effect is broad, holding across five token-reduction methods (ToMe, EViT, ATS, ATC, PiToMe), nine backbones, all ImageNet-C corruption types, eight further shift suites, and two further modalities, video and vision-language QA. It is also specific to shift, still positive on clean and rising monotonically to ~4x that at the highest severity 5. The schedule keeps its gain under six test-time adaptation methods, and needs no per-input or per-domain tuning.
Sep 30, 2026

Sparse-WAM: Accelerating World Action Models via Action-Guided Sparse Imagination

World-action models (WAMs) leverage pretrained video models to improve generalization in robot control by jointly predicting future visual states and actions. This capability comes at a substantial inference cost, as dense future-frame tokens are repeatedly processed during denoising. Prior methods address this by token pruning that prioritizes visual fidelity to reduce denoising costs in video diffusion models. However, these methods do not use action relevance to determine which future-frame tokens to retain during joint denoising in WAMs. In this paper, we propose Sparse-WAM, a training-free framework for action-guided sparse imagination that selectively processes future-frame tokens to accelerate WAM inference. We observe substantial overlap in the spatial distribution of attention from action tokens to future-frame tokens (action-to-future attention) between consecutive denoising steps, despite continued updates to the future representations. Motivated by this, we develop Action-Guided Token Selection to retain frame-specific action-relevant regions together with cross-frame context. However, a naive implementation can incur attention-scoring and token-packing overhead that offsets the computational savings from pruning. We therefore introduce Pilot, an efficient engine that reduces sparse inference overhead through lightweight scoring and cross-step reuse of token selections. On LIBERO with FastWAM-Joint and RoboLab-120 with Cosmos 3 Edge, Sparse-WAM achieves inference speedups of approximately 2.0×2.0\times and 1.8×1.8\times, respectively, over dense eager inference on an NVIDIA RTX 4090, while largely preserving task performance.
Sep 30, 2026

Audio Token Attention Is Predictable Before the Language Model Runs

A large audio language model (LALM) turns a minute of speech into 750-1,500 tokens and prefills every one. Image-token pruning often cuts after the language model's first layers, where image tokens draw little attention. Audio tokens draw much more attention there, and their ranking is still far from final, so audio needs a ranking before the language model runs. Surprisingly, the attention an audio token will receive across the language model is already linearly predictable from its encoder output, before the language model runs. A linear map, fitted in closed form without labels, predicts this all-layer attention ranking at ρ≥.69ρ\geq .69 on eleven of thirteen LALMs. Our method, Triage, cuts audio tokens by this prediction and, on multiple choice, cuts again at layer 2, correcting the prediction with the attention observed there. Triage sets its compression without labels, under two budgets that limit how far its output may differ from the model's own full-audio output. At the conservative budget, its word error rate and accuracy stay within .04 of full audio. At the aggressive budget, Triage beats every baseline in all twelve transcription cases. On multiple choice, at 2.2-5x compression, it outperforms DART, the strongest baseline on average, by .043 in mean accuracy. Because it cuts before the language model, it raises the audio that fits in Qwen2.5-Omni-3B's context window from 21.8 to about 62 minutes. At its most compressive point, Triage lets one GPU serve 4x as many concurrent 5-minute streams of that model. Project page: https://audio-triage.github.io
Sep 8, 2026

FastE: Readout-Triggered Token Compression for LLM Embedding Inference

In this study, we identify depth-dependent prefix redundancy in final-readout LLM embedding models, notably across representative backbones including Qwen3-Embedding and Qwen3-VL-Embedding. We find that removing prefix states is substantially more damaging in shallow layers than at greater depth, showing that prefix states become increasingly compressible as the prefix and readout states propagate through the network. To this end, we introduce FastE, a training-free, plug-and-play method. FastE uses a shared fixed threshold on batch-mean readout-prefix alignment as a lightweight online heuristic for selecting when compression occurs, and ranks prefix states by the attention scores they receive from the readout position to determine which states are retained in subsequent layers. Our evaluations demonstrate FastE's ability to substantially reduce computational costs: on NarrativeQA with Qwen3-Embedding-0.6B, it reduces decoder-backbone FLOPs by 40.11% while retaining 99.53% of Full Forward nDCG@10. Across five text embedding benchmarks, two backbone scales, and three cross-modal retrieval tasks, the quality-efficiency trade-off is directly customizable through the maximum removal ratio without retraining. We believe FastE offers practical value for scalable embedding generation in retrieval, indexing, clustering, and multimodal representation systems.
Sep 7, 2026

Query-Aware Token Budgeting for Efficient Late-Interaction Visual Document Retrieval

Late-interaction visual document retrievers preserve fine-grained page evidence by storing many token embeddings per page, but the resulting storage and query-time interaction costs make large-scale deployment expensive. Pooling document tokens before indexing offers a natural remedy, yet static pooling must decide which visual evidence to preserve before the query is known. We study an alternative: a heavily compressed hot-path index generates candidates, after which query-aware token budgeting operates on the original token sets of the shortlisted pages. We formulate this stage-two selection as a budgeted MaxSim coverage problem, show that a clipped version is monotone submodular, and compare coverage-only, cluster-guided, token-wise, and marginal-gain policies. On ten ViDoRe tasks with ColModernVBERT, direct static pooling reduces macro normalized discounted cumulative gain at rank five from 0.6309 without compression to 0.4738 at a thirty-two-fold pool factor. Under the same candidate-generation regime and a pool-factor-eight-equivalent reranking budget, token top-k recovers 93.93 percent of the full-token score, while greedy marginal-gain selection recovers 98.39 percent. Held-out and leave-one-dataset-out evaluations yield positive greedy improvements over token top-k on every dataset. The latency analysis reveals two useful operating points: token top-k for interactive retrieval and the naive greedy implementation as a quality upper envelope. Together, these results show that late-interaction visual retrieval benefits from query-aware allocation rather than query-agnostic pooling alone.
Aug 25, 2026

MaST: Motion-aware Sparse Pipeline for Lightweight Object Tracking

Transformer-based object trackers are renowned for their strong performance, yet dense token processing often leads to prohibitive computational cost, limiting real-time deployment on edge devices. While recent works explore token pruning to reduce computation, they often stop short of an end-to-end sparse pipeline, as early-layer token scores can be noisy without a motion prior, and many trackers ultimately fall back to dense reshaping to feed the dense prediction head that partially negates the savings. We introduce Motion-aware Sparse Tracker (MaST), a sparse tracking framework that makes sparsity effective from tokens to boxes. First, MaST injects a lightweight motion prior to refine cross-attention-based importance scores, enabling earlier and more stable token reduction in the search region. Second, we introduce a natively sparse prediction head that operates directly on the retained unstructured tokens with a score-first, regress-once design, eliminating dense padding/reshaping and reducing redundant computation. Extensive experiments on multiple benchmarks demonstrate that MaST establishes new state of the art among lightweight trackers, where MaST-tiny attains 63.8 AUC on LaSOT and 80.1 SUC on TrackingNet, surpassing the prior best AsymTrack-S by +1.0 AUC and +2.2 SUC while running at 152 FPS on Jetson Nano, nearly twice as fast as AsymTrack-S at 88 FPS. Code is available at https://github.com/TsingWei/MaST.
Aug 11, 2026

Putting Registers to Work: Task Registers for Token Pruning in Vision Transformers

Token-pruning policies are usually designed for a single recognition pipeline, but pretrained Vision Transformers are reused across tasks with different spatial demands. We ask which parts of a pruning policy transfer across image classification, semantic segmentation, and object detection. For each pipeline, controlled probes freeze the no-pruning checkpoint and apply a series of parameter-free reduction criteria at one eligible layer at a time without retraining. The probes reveal three differences: segmentation and detection rank the criteria differently, classification is especially sensitive to attention-based pruning in the earliest layers, and the dense tasks prefer opposite recovery endpoints. These findings motivate Task-Adaptive Pruning (TAP). Existing register tokens serve as task-agnostic storage for feature artifacts. TAP instead introduces one task register per task and activates only the current one. Its evolving state ranks tokens, distributes an exact removal budget over depth, and sets the recovery scale for dense features. At a final keep rate of ρ=0.5ρ=0.5, our jointly adapted model, TAP-J, reaches 47.047.0 mIoU at 1.30×1.30\times encoder throughput on ADE20K and 53.753.7 box AP at 1.32×1.32\times encoder throughput on COCO while remaining competitive on ImageNet-1K.
Aug 9, 2026

Deferred Audio Pruning with Local Audio-Visual Dynamics for Omni-LLMs

Omni-modal LLMs jointly process audio, video, and text, but long multimodal sequences incur substantial prefill and KV-cache costs. Existing omni-modal compression methods primarily focus on pre-LLM token reduction, leaving modality-specific compression across the LLM boundary underexplored. We propose A-PACK, a two-stage framework that defers audio pruning until query-conditioned multimodal interactions emerge. Our analysis shows that audio exhibits higher task-relevant information density and representational diversity per token than video. We further find that local audio-visual dynamics provide a more effective cue for visual selection than token-wise matching. We therefore preserve audio and compress video with local dynamics before the LLM, then progressively prune low-relevance audio and visual tokens and their KV-cache entries inside the LLM. Across four benchmarks on Qwen2.5-Omni-7B/3B, A-PACK achieves the strongest average performance among the evaluated prior methods while reducing prefill FLOPs by up to 78% and improving decoding throughput by up to 2.21x.
Aug 3, 2026

DiffPrune: differentiable information throttling for token pruning in vision-language models

Visual token pruning reduces the computational cost of Vision-Language Models (VLMs) by removing redundant visual tokens. The key is to learn a score that measures whether a token is useful. Existing methods typically rely on Gumbel-Softmax to approximate discrete selection during training. Such selectors make the score depend on the behavior of a relaxed pruning operator, not directly on the consequence of information loss. In this paper, we propose DiffPrune, which gives token scores a direct meaning. During training, DiffPrune keeps all tokens and weakens each token's information according to its score. If weakening a token hurts the task, the scorer is pushed to protect it; if not, the token can receive a lower score. Because the loss is differentiated through this actual information-throttling path, the scorer avoids the unstable surrogate path of relaxed token selection. DiffPrune implements this idea with an Information Throttler, which injects variance-preserving noise into visual tokens, where high-score tokens remain close to their original representations, while low-score tokens carry less original information. At inference, the throttler is removed, and hard top-K pruning is applied using the learned scores. Across ten VLM benchmarks, DiffPrune retains 96.5% of full-model accuracy while accelerating LLM prefill by 2.85x, with only 0.69 ms inference overhead. Code will be publicly available.
Aug 2, 2026

When Replanning Becomes the Bottleneck: Budgeted Replanning for Embodied Agents

Embodied agents replan frequently to recover from execution drift, partial observability, and coordination hazards, but each LLM-based replanning call can consume an accumulated textual context that grows over time and across agents. Once this context becomes large, replanning latency develops heavy tails and can miss real-time deadlines even when task success remains high, a failure mode that is hard to detect from average latency or success alone. We present BRACE, a controller that formulates replanning as a budgeted control loop by deciding whether to replan, selecting a replanning mode, and allocating an explicit token budget and latency service-level objective (SLO) while accounting for optional efficiency modules. As a reusable component, we introduce E-RECAP, a cost-aware progressive token pruning method that predicts token utility and prunes replanning contexts across transformer layers while preserving critical head and tail tokens. Across Meta Habitat, RoboFactory, and AirSim, BRACE with E-RECAP reduces replanning-call token counts by 62-92% and SLO violation rates from 85.5-100.0% to 4.7-50.0% in settings where task success is already saturated. In a harder RoboFactory setting where open-loop, frozen-plan, and No BRACE all fail, BRACE + E-RECAP reaches 80.0% success with 4.6% SLO violations, demonstrating that tail-aware per-call budgeting is effective across embodied platforms.
Aug 1, 2026

Pruned BPE: Post-training Visibility Pruning and Token Reallocation for Byte Pair Encoding

Byte Pair Encoding (BPE) is widely used for subword tokenization, but standard BPE exposes every learned merge token to the downstream model, including tokens that mainly serve as intermediate construction units and rarely appear in the final encoded corpus. This paper proposes Pruned BPE, a post-training visibility-pruning and token-reallocation method that separates merge construction from model-visible vocabulary selection. After standard BPE training, tokens are evaluated by final exposure. Low-exposure tokens are retained as internal-only merge nodes, while their visible vocabulary slots are reassigned to better-exposed candidates learned through resumed training. During encoding, internal-only tokens are recursively expanded into visible descendants while the original BPE merge order is preserved. Experiments on two non-overlapping English- and Chinese-dominated corpora and their combination show that Pruned BPE consistently reduces encoded length relative to Standard BPE at the same training corpus, evaluation corpus, and model-visible vocabulary size. At a 40% exposure threshold, the reduction is approximately 0.27%--0.36% on same-corpus evaluations. In a vocabulary-only evaluation using a shared exact minimum-token dynamic-programming encoder, Pruned BPE retains an advantage of approximately 0.23%--0.31%, indicating that the improvement arises from a more efficient visible vocabulary. These gains represent a meaningful fraction of the approximately 1.5%--3.8% marginal reduction that would otherwise require adding another 2K Standard BPE tokens. Qualitative analysis shows that internal-only tokens include reusable English fragments, Chinese components, partial UTF-8 byte sequences, and structured-text fragments. The results indicate that post-training visibility pruning can improve BPE vocabulary efficiency without increasing the vocabulary exposed to the language model.
Jul 30, 2026

WIDE: Boosting Adaptive LLM Inference via Token-level Dynamic Width Pruning

Pruning is a promising approach for improving the efficiency of large language models (LLMs). Static structured pruning is hardware-friendly, but its input-agnostic computation allocation often degrades model quality at high sparsity. Dynamic pruning adapts computation to individual tokens, but coarse-grained layer-level routing limits allocation flexibility, and finer-grained decisions introduce irregular execution that can offset computational savings. To address these challenges, we present WIDE, the first end-to-end differentiable framework that combines token-wise structured dynamic width pruning with GPU kernel co-design for both prefill and decoding. WIDE uses differentiable routers to select GQA-aligned query-head groups and GPU-tile-aligned FFN-channel groups for each token. To execute these decisions efficiently, it reorders routing masks to cluster active tokens within each group, enabling hardware-agnostic block-level skipping and architecture-aware skipping of inactive loads and tensor-core operations. Experiments on the Llama3 and Qwen3 families demonstrate improved quality retention over representative static and dynamic pruning methods. At 50% target routing sparsity, WIDE improves Llama3.1-8B average zero-shot accuracy by up to 20.26 points over the strongest dynamic-depth baseline under calibration-only settings, while retaining 97.6%/88.2% of dense Qwen3-14B likelihood/generation performance after LoRA tuning. Separate latency benchmarks demonstrate end-to-end speedups of up to 1.68×1.68\times for prefill and 1.55×1.55\times for decoding over the dense baseline. These results demonstrate that pruning--kernel co-design pushes the frontier of token-wise dynamic structured pruning, combining improved quality retention with practical acceleration across both prefill and decoding. Our code is avaliable at https://github.com/EIT-NLP/LLM-Pruning/tree/main/WIDE.
Jul 30, 2026

Capturing Token Tendencies for Training-Free Token Pruning in Multimodal Large Language Models

While visual token pruning is essential for efficient Multimodal Large Language Models (MLLMs), existing training-free methods suffer from a critical limitation: they rely on static, instantaneous heuristics to perform irreversible filtering. This approach ignores the hierarchical nature of MLLMs, where token importance often evolves dynamically rather than remaining fixed across layers. Consequently, tokens essential for deep-layer reasoning are often prematurely discarded by shallow-layer estimates. To address this, we propose Trend-aware Pruning, a novel framework that elevates pruning from a local snapshot decision to a temporal trajectory modeling problem. Instead of relying on isolated scores, our method captures the momentum of attention flow. This enables a dynamic rectification mechanism that selectively reactivates "late-blooming" tokens, those initially undervalued but exhibiting rising semantic importance, thereby preventing the loss of critical visual cues. Extensive experiments demonstrate that our approach achieves a superior efficiency-performance trade-off across diverse multimodal tasks. Notably, it reduces visual tokens by over 77.8%, retaining only approximately 23 tokens in the final layer while maintaining competitive performance, offering a robust and reversible solution for high-efficiency multimodal inference.
Jul 26, 2026

Omni-Prune: Query-Aware Unified Token Pruning for Efficient Omnimodal Large Language Models

Omnimodal large language models (OmniLLMs) are rapidly extending multimodal reasoning to cover synchronized audio and video. However, the resulting audio-video token sequences are long, leading to high prefill latency and GPU memory usage at inference time. Existing token pruning methods, designed mainly for vision-only inputs, miss both the cross-modal links between audio and video and the user query that decides which content matters. To bridge this gap, we present Omni-Prune, a training-free, query-aware audio-visual token pruning framework that jointly removes redundancy from both modalities while keeping task-relevant cross-modal evidence. Specifically, Omni-Prune first splits the token sequence into adaptive time windows placed at audio saliency peaks, then scores audio and video tokens on a single scale that combines encoder attention with text-query relevance, and pairs related audio-video tokens so that they are kept together. Within each window, a final K-medoids step then selects a few representative tokens, adding diverse cues that score-based selection alone would miss. Extensive experiments demonstrate that Omni-Prune outperforms established baseline methods, delivering up to 3.25x prefill speedup and 1.3x memory reduction while retaining over 99% of full-model performance.
Jun 23, 2026

Spectral Evolution-Guided Token Pruning in Multimodal Large Language Models

Reducing visual token redundancy is critical for accelerating Multimodal Large Language Models (MLLMs) without degrading cross-modal reasoning performance. Existing token pruning methods typically rely on single-layer signals, such as attention scores or token similarities, which overlook the cross-layer transformation of visual representations and may exhibit positional bias in multimodal token sequences. To address this limitation, we propose a training-free token pruning framework based on Cross-Layer Spectral Evolution (CLSE). Instead of measuring token importance from single-layer feature magnitudes, CLSE quantifies how token representations evolve across Transformer layers in the frequency domain. This evolution reflects the transition from high-frequency structural details to low-frequency semantic abstractions. We observe that tokens with stronger spectral redistribution across layers are more likely to be semantically active and should therefore be preserved. By modeling cross-layer token dynamics, CLSE provides a stable importance criterion that mitigates positional bias. Extensive experiments on both image and video benchmarks demonstrate that CLSE achieves a superior trade-off between efficiency and accuracy under aggressive token reduction. Across multiple MLLMs, CLSE reduces FLOPs, KV cache memory, and latency while maintaining competitive or improved performance.
Jun 22, 2026

Keep The Essentials: Efficient Reference Conditioned Generation via Token Dropping

Reference-based diffusion models enable highly controllable image generation by leveraging elements from input images to guide prompt-driven synthesis. However, these models are computationally expensive in runtime, and their cost scales severely with the number of input references. While the efficiency of diffusion models has been extensively studied in the context of prompt-driven generation, it remains largely under-explored in the realm of reference-based models. This setting presents unique challenges not addressed by methods focusing solely on generation. In particular, the wasteful representation of references as dense token grids offers significant opportunities for improvement. In this work, we present Sparse Context, a method for constructing sparse reference representations by retaining only a reduced subset of reference tokens. We observe that even without modifying the model, dropping a significant portion of reference tokens at inference time largely preserves its generation capabilities. To fully realize this potential, we fine-tune the model with random token dropping at varying ratios, encouraging robustness to partial reference representations. Crucially, this training strategy decouples the model from any specific token selection rule, allowing flexible control at inference time. At inference time, instead of random dropping, we apply task-aware token selection strategies that prioritize the most informative regions of the reference images, adapting the token budget to the input and task requirements. Extensive experiments show our method achieves a 4x increase in inference speed for multi-reference generation and an 2x for single reference generation. Importantly, this efficiency is achieved without compromising visual quality across both spatially-aligned editing and subject-driven generation.
Jun 2, 2026

Entropy Gate: Entropy Quenching for Near-Lossless Token Compression in LLM Pipelines

LLM pipelines waste substantial token budgets on low-information content: repeated context, verbose responses, and redundant boilerplate. We introduce Entropy Gate, a token compression framework applying entropy quenching −- a thermodynamic process that progressively freezes out low-energy tokens while preserving semantic fidelity. Each token receives a multi-factor information energy E(t)E(t) combining statistical, structural, and positional components. An adaptive quenching schedule T(τ)=T0/(1+ατ)T(τ) = T_0 / (1 + ατ) removes tokens whose Boltzmann survival probability pi=exp⁡(−Ei/kT)p_i = \exp(-E_i / kT) falls below threshold, with a fidelity gate halting compression when energy-weighted similarity drops below θθ. We prove token selection by descending E(t)E(t) maximizes expected semantic preservation, that quenching produces nested survival sets, and that achievable compression approaches the information-theoretic limit CR→1−I(P;T)/H(P)\text{CR} \to 1 - I(P; T)/H(P). A Phase 1 heuristic achieves 40-60% compression across five prompt categories while maintaining SE>0.80S_E > 0.80, with energy-squared amplification E→E2E \to E^2 adding 10-25 percentage points. Context deduplication adds 50-70% savings on repeated blocks. Output-side quenching, motivated by findings that brevity improves accuracy, further reduces response overhead. Combined with external memory, reduction composes multiplicatively to 88-96% for agentic workloads. The framework is stateless, model-agnostic, and deploys as an OpenAI-compatible HTTP proxy.
May 28, 2026

AsymVLM: Asymmetric Token Pruning for Efficient Vision-Language Model Inference

Vision-Language Models (VLMs) process thousands of visual tokens per image alongside comparatively few text tokens, yet existing compression methods treat both modalities uniformly. We observe that the two modalities have fundamentally different properties: vision tokens are spatially redundant and dominate prefill, while text tokens are causally dependent and accumulate during decoding. Based on this asymmetry, we propose and empirically evaluate AsymVLM, which applies aggressive pruning to vision tokens before prefill using a learned importance scorer with per-sample adaptive budgeting, and temporal threshold-based eviction to text tokens only when they exceed a fixed budget. Our experiments indicate that AsymVLM achieves the highest FLOPs savings (up to 54%) among state-of-the-art methods while outperforming existing approaches by 2--3% on document and chart understanding tasks where visual information is spatially localized and query-specific, and maintaining competitive accuracy on holistic benchmarks. In text-dominated scenarios, our eviction strategy substantially outperforms standard LLM cache compression methods by adapting to the short-context nature of VLM.
May 27, 2026

Beyond Surrogate Gradients: Fully Differentiable Token Pruning for Vision-Language Models

Visual token pruning reduces the computational cost of Vision-Language Models (VLMs) by removing redundant visual tokens. Existing methods typically rely on Gumbel-Softmax to approximate discrete selection during training. However, the optimization is driven by surrogate gradients rather than the true selection process, leading to unreliable learning of token importance. In this paper, we propose DiffPrune, which reformulates pruning as continuous control of token information instead of discrete selection learning. Specifically, we introduce an Information Throttler that modulates each token using variance-preserving noise conditioned on importance scores, where higher scores induce less information suppression during training. This design directly operates on token representations, naturally providing a fully differentiable optimization path for learning token importance. At inference, tokens are removed via hard thresholding on the learned scores. Across ten VLM benchmarks, DiffPrune retains 96.5% of full-model accuracy while accelerating LLM prefill by 2.85x, with only 0.69 ms of inference overhead.
May 19, 2026

Stage-adaptive Token Selection for Efficient Omni-modal LLMs

The unified audio-visual understanding capability of omni-modal large language models (om-LLMs) comes at the high cost of processing tens of thousands of visual and audio tokens throughout the LLM. To reduce the cost, training-free methods for om-LLM token selection are being actively developed, operating either pre-LLM or inner-LLM. Recognizing that the non-textual token redundancy varies across stages and LLM layers, we propose in this paper a novel stage-adaptive token selection (SEATS) method that operates in both pre-LLM and inner-LLM stages. Given an overall token budget measured by the token retention rate (TRR), SEATS performs top-down token budget allocation in an input-dependent manner. Specifically, a larger TRR is used for pre-LLM token selection, whilst within the LLM, all tokens are retained in shallow layers, increasingly decreased TRRs are applied in middle layers, and the non-textual tokens are fully removed in late layers. Extensive experiments on three representative om-LLMs and six benchmarks verify the viability of the proposed method. By retaining only 10% of visual and audio tokens, SEATS preserves 94.9% of the full-token performance on Qwen2.5-Omni while achieving a 9.3x FLOPs reduction and a 4.9x prefill speedup.