Redundancy

Long-form video understanding remains challenging for Video Large Language Models (VideoLLMs), as the dense frame sampling introduces massive visual tokens while sparse sampling risks missing critical temporal evidence and leading to LLM hallucination. Existing training-free token reduction methods either treat videos equally as static images or rely on segment-level merging heuristics, which weaken fine-grained spatiotemporal modeling and introduce additional overhead. In this paper, we propose EchoPrune, a lightweight and training-free token pruning method that improves temporal resolution under a fixed LLM-side visual token budget. Our core idea is to interpret redundant video tokens as temporal echoes: if a token is well reconstructed from the previous frame, it is merely a temporally redundant echo; otherwise, it may capture new events, motion, or query-relevant visual evidence. Based on this insight, EchoPrune scores visual tokens by (i) query-guided crossmodal relevance and (ii) temporal reconstruction error, measured by correspondence matching and echo matching across consecutive frames. The selected tokens preserve task-relevant cues and temporal novelty while suppressing predictable redundancy, allowing VideoLLMs to observe more frames without increasing the decoding budget. Extensive experiments on LLaVA-OV, Qwen2.5VL, and Qwen3VL across six video understanding benchmarks show that EchoPrune enables VideoLLMs to process up to 20x frames under the same token budget, yielding improved performance (+8.6%) and inference speedup (5.6x for prefilling) on Qwen2.5VL-7B.

Diffusion Transformers (DiTs) have achieved state-of-the-art video generation quality, but they incur immense computational cost because standard inference applies the same number of denoising steps uniformly to every token in the sequence. It is well known that human vision ignores vast amounts of redundant motion. Why, then, do our densest models treat every spatiotemporal token with equal priority? In this paper, we introduce Heterogeneous Step Allocation (HSA), a training-free inference algorithm that assigns varying step budgets to different spatiotemporal tokens based on their velocity dynamics. To resolve the resulting sequence-length mismatch without sacrificing global context, HSA introduces a KV-cache synchronization mechanism that allows active tokens to attend to the full sequence while entirely bypassing inactive tokens. Furthermore, we derive a cached Euler update that advances the latent states of skipped tokens in a single operation without additional model evaluations. We evaluate HSA on the Wan-2 and LTX-2 models for both text-to-video (T2V) and image-to-video (I2V) generation. Our results demonstrate that HSA significantly outperforms previous state-of-the-art caching methods and the vanilla Flow Matching baseline, especially at aggressive acceleration regimes (e.g., 50% and 25% runtimes). Crucially, HSA achieves a superior quality-runtime Pareto frontier without the need for expensive offline profiling, robustly preserving structural integrity and generation quality even under tight computational budgets. Project page: https://ernestchu.github.io/hsa

Diffusion language models (DLMs) have recently emerged as competitive alternatives to autoregressive (AR) language models, yet differences in their activation dynamics remain poorly understood. We characterize these dynamics in LLaDA-8B and identify a striking layer-collapse property: a few early layers exhibit highly similar, collapsed activation patterns dominated by a single large super-outlier persisting over a long token range. Despite its apparent redundancy, this outlier is critical: pruning it causes outputs to degrade into repetitive random token loops. Paradoxically, layers in LLaDA contain more redundant representations overall, with redundancy most pronounced in earlier layers -- the reverse of AR models, where deeper layers grow redundant due to undertraining. Our analysis indicates that layer collapse in DLMs is not driven by undertraining but by overtraining: a dominant outlier becomes an indispensable information carrier while remaining representations collapse into redundant structure. These findings have strong practical implications, verified through controlled pre-training experiments. DLMs are surprisingly robust to compression: LLaDA under 3-bit GPTQ quantization drops only -1.8% on GSM8K, whereas Llama-3.1-8B drops -64.7%. Optimal sparsity allocation also reverses between families: at 50% average sparsity, allocating more to early layers in LLaDA yields +8.4% over the reverse strategy, while the same allocation costs Llama -8.4%. Our findings reveal that the DLM training objective fundamentally reshapes layer dynamics relative to AR models, with direct consequences for compression and deployment. Code: github.com/Conzel/super-outlier-dlm.

Current vision language models face hallucination and robustness issues against ambiguous or corrupted modalities. We hypothesize that these issues can be addressed by exploiting the shared information between modalities to compensate for the impaired one. To this end, we analyze multimodal interactions -- redundant (shared), unique (exclusive), and synergistic (emergent) task-relevant information provided by the modalities -- to determine their impacts on model reliability. Specifically, amplifying redundant interactions would increase this exploitable shared information to resolve these issues; yet, modern instruction datasets often eliminate redundancies to prioritize visual grounding. We bridge this gap through a self-captioning workflow featuring a \textsc{Multimodal Interaction Gate}: a mechanism to convert unique interactions into redundant interactions. Our findings suggest that increasing redundancy can reduce visual induced errors by 38.3% and improve consistency by 16.8%.

Production LLM workloads increasingly serve discriminative tasks, such as classification, recommendation, and verification, whose answers are read from the logits of a single prefill pass with no autoregressive decoding. Serving these prefill-only workloads on mixture-of-experts (MoE) models is bottlenecked not by compute but by the distributed execution required to fit the model: existing parallel strategies (tensor, expert, and pipeline parallelism) trade memory pressure for redundant computation, communication, and synchronization, severely degrading MoE prefill serving efficiency. We observe that these overheads stem from coupling expert placement with synchronous activation routing -- a design inherited from the decoding era. The long, compute-bound forward passes of large-batch prefill open a per-layer window wide enough to stream expert weights in the background, replacing per-layer activation AllToAll with asynchronous weight AllGather fully overlapped with computation. We propose MoE-Prefill, a prefill-only serving system whose backend, AsyncEP (Asynchronous Expert Parallelism), gathers experts by weight rather than routing them by activation, and whose frontend co-enforces a physically-derived saturation threshold through prefix-aware routing and true-FLOPs load tracking. On Qwen3-235B-A22B across four hardware/precision configurations, MoE-Prefill delivers 1.35-1.37x throughput over the strongest distributed baseline on real-world workloads and up to 1.59x on long-context synthetic workloads, sustaining 29.8-36.2% per-GPU model FLOPs utilization.

Models that are indistinguishable on in-distribution data can behave very differently under distribution shift. We introduce Perturb-and-Correct (P&C), a post-hoc method for constructing epistemically diverse predictors from a single pretrained network. P&C applies random hidden layer perturbations with a least-squares correction in the subsequent affine layer, producing predictors that agree on calibration data while remaining free to disagree away from it. We analyze this mechanism through the post-correction residual and its first-order sensitivity: the residual is controlled near the calibration distribution by a leverage term, while corrected sensitivity grows as inputs deviate from the calibration geometry. Empirically, P&C achieves a strong ID/OOD tradeoff across MuJoCo dynamics prediction and CIFAR-10 OOD detection, matching or outperforming standard post-hoc baselines while requiring only a single pretrained model. Our findings highlight the potential in further exploiting overparameterization as a strength of deep learning models.

AI agents that interact with graphical user interfaces (GUIs) require effective observation representations for reliable grounding. The accessibility tree is a commonly used text-based format that encodes UI element attributes, but it suffers from redundancy and lacks structural information such as spatial relationships among elements. We propose A11y-Compressor, a framework that transforms linearized accessibility trees into compact and structured representations. Our implementation, Compressed-a11y, applies a lightweight and structured transformation pipeline with modal detection, redundancy reduction, and semantic structuring. Experiments on the OSWorld benchmark show that Compressed-a11y reduces input tokens to 22% of the original while improving task success rates by 5.1 percentage points on average.

Electroencephalogram (EEG) signals are vital for automated seizure detection, but their inherent noise makes robust representation learning challenging. Existing graph construction methods, whether correlation-based or learning-based, often generate redundant or irrelevant edges due to the noisy nature of EEG data. This significantly impairs the quality of graph representation and limits downstream task performance. Motivated by the remarkable reasoning and contextual understanding capabilities of large language models (LLMs), we explore the idea of using LLMs as graph edge refiners. Specifically, we propose a two-stage framework: we first verify that LLM-based edge refinement can effectively identify and remove redundant connections, leading to significant improvements in seizure detection accuracy and more meaningful graph structures. Building on this insight, we further develop a robust solution where the initial graph is constructed using a Transformer-based edge predictor and multilayer perceptron, assigning probability scores to potential edges and applying a threshold to determine their existence. The LLM then acts as an edge set refiner, making informed decisions based on both textual and statistical features of node pairs to validate the remaining connections. Extensive experiments on TUSZ dataset demonstrate that our LLM-refined graph learning framework not only enhances task performance but also yields cleaner and more interpretable graph representations.

Depth pruning improves the inference efficiency of large language models by removing Transformer blocks. Prior work typically treats layer redundancy as an inherent structural property of pretrained networks, emphasizing importance criteria and search algorithms to identify removable layers. In this study, we empirically investigate depth pruning from a functional perspective. Evaluating representative LLM families across diverse calibration configurations and multiple search algorithms, we show that different configurations produce different pruning patterns. Furthermore, under a fixed calibration configuration, complex search algorithms yield marginal performance improvements over simple one-shot methods, converging to similar pruned subsets. Overall, our results suggest that the calibration configuration plays a substantially larger role than the choice of search algorithm in shaping pruning patterns and calibration perplexity, while contributing comparably to variance in downstream reasoning accuracy. This indicates that future pruning efforts may benefit from prioritizing the calibration configuration over search complexity.

Standard Retrieval-Augmented Generation (RAG) chunking methods often create excessive redundancy, increasing storage costs and slowing retrieval. This study explores chunk filtering strategies, such as semantic, topic-based, and named-entity-based methods in order to reduce the indexed corpus while preserving retrieval quality. Experiments are conducted on multiple corpora. Retrieval performance is evaluated using a token-based framework based on precision, recall, and intersection-over-union metrics. Results indicate that entity-based filtering can reduce vector index size by approximately 25% to 36% while maintaining high retrieval quality close to the baseline. These findings suggest that redundancy introduced during chunking can be effectively reduced through lightweight filtering, improving the efficiency of retrieval-oriented components in RAG pipelines.

Reasoning-capable large language models solve hard problems by emitting long chains of thought, paying heavily in latency, GPU time, and energy. Casual inspection of their traces reveals extensive reformulation, verification, and circular self-reflection, yet how much of this deliberation is actually necessary has never been measured at scale or explained from first principles. This paper closes both gaps. We formalise reasoning redundancy directly in terms of the reasoning model itself: the redundancy of a correct trace is the largest fraction of its trailing segmented steps that can be truncated while $π$, forced to terminate thinking and emit a final answer, still produces the correct answer. A large-scale quantification across four frontier reasoning models and two mathematical benchmarks shows that step-level redundancy is consistently high -- between 61% and 93% across the 8 (model, benchmark) conditions we study, with the median critical prefix equal to a single segmented step in six of the eight conditions -- that the finding is robust to the choice of judge family, and that although $ρ$ decreases with problem difficulty on MATH-500, all four models remain substantially redundant ($ρ\in [46\%, 85\%]$) even on the hardest Level-5 problems. We then prove that this redundancy is a structural consequence of length-agnostic outcome rewards, not a model-specific artefact: under any such reward, no finite expected stopping time is optimal. The result holds regardless of RL algorithm, base model, data distribution, or whether the policy is obtained via RL or distillation; over-thinking is therefore not a bug to be patched in individual models but a structural property of how current reasoning models are trained. Code: https://github.com/zhiyuanZhai20/how-much-thinking-is-enough

Existing QA benchmarks typically assume distinct documents with minimal overlap, yet real-world retrieval-augmented generation (RAG) systems operate on corpora such as financial reports, legal codes, and patents, where information is highly redundant and documents exhibit strong inter-document similarity. This mismatch undermines evaluation validity: retrievers can be unfairly undervalued even when they retrieve documents that provide sufficient evidence, because redundancy across documents is not accounted for in evaluation. On the other hand, retrievers that perform well on standard benchmarks often generalize poorly to real-world corpora with highly similar and redundant documents. We present RARE (Redundancy-Aware Retrieval Evaluation), a framework for constructing realistic benchmarks by (i) decomposing documents into atomic facts to enable precise redundancy tracking and (ii) enhancing LLM-based data generation with CRRF. RAG benchmark data usually requires multiple quality criteria, but LLMs often yield trivial outputs. CRRF scores criteria separately and fuses decisions by rank, improving the reliability of generated data. Applying RARE to Finance, Legal, and Patent corpora, we introduce RedQA, where a strong retriever baseline drops from 66.4% PerfRecall@10 on 4-hop General-Wiki to 5.0-27.9% PerfRecall@10 at 4-hop depth, revealing robustness gaps that current benchmarks fail to capture. RARE enables practitioners to build domain-specific RAG evaluations that faithfully reflect real-world deployment conditions.

Human perception generalizes well across different domains, but most vision models struggle beyond their training data. This gap motivates multi-dataset learning, where a single model is trained on diverse datasets to improve robustness under domain shifts. However, unified training remains challenging due to inconsistencies in data distributions and label semantics. Mixture-of-Experts (MoE) models provide a scalable solution by routing inputs to specialized subnetworks (experts). Yet, existing MoEs often fail to specialize effectively, as their load-balancing mechanisms enforce uniform input distribution across experts. This fairness conflicts with domain-aware routing, causing experts to learn redundant representations, and reducing performance especially on rare or out-of-distribution domains. We propose GEM (Global Expert Mapping), a planner-compiler framework that replaces the learned router with a global scheduler. Our planner, based on linear programming relaxation, computes a fractional assignment of datasets to experts, while the compiler applies hierarchical rounding to convert this soft plan into a deterministic, capacity-aware mapping. Unlike prior MoEs, GEM avoids balancing loss, resolves the conflict between fairness and specialization, and produces interpretable routing. Experiments show that GEM-DINO achieves state-of-the-art performance on the UODB benchmark, with notable gains on underrepresented datasets and solves task interference in few-shot adaptation scenarios.

Prefilling computational costs pose a significant bottleneck for Large Language Models (LLMs) and Large Multimodal Models (LMMs) in long-context settings. While token pruning reduces sequence length, prior methods rely on heuristics that break compatibility with hardware-efficient kernels like FlashAttention. In this work, we observe that tokens evolve toward \textit{semantic fixing points}, making further processing redundant. To this end, we introduce Delta Attention Selective Halting (DASH), a training-free policy that monitors the layer-wise update dynamics of the self-attention mechanism to selectively halt stabilized tokens. Extensive evaluation confirms that DASH generalizes across language and vision benchmarks, delivering significant prefill speedups while preserving model accuracy and hardware efficiency. Code will be released at https://github.com/verach3n/DASH.git.

Point-based Neural Networks (PNNs) have become a key approach for point cloud processing. However, a core operation in these models, Farthest Point Sampling (FPS), often introduces significant inference latency, especially for large-scale processing. Despite existing CUDA- and hardware-level optimizations, FPS remains a major bottleneck due to exhaustive computations across multiple network layers in PNNs, which hinders scalability. Through systematic analysis, we identify three substantial redundancies in FPS, including unnecessary full-cloud computations, redundant late-stage iterations, and predictable inter-layer outputs that make later FPS computations avoidable. To address these, we propose \textbf{\textit{FlashFPS}}, a hardware-agnostic, plug-and-play framework for FPS acceleration, composed of \textit{FPS-Prune} and \textit{FPS-Cache}. \textit{FPS-Prune} introduces candidate pruning and iteration pruning to reduce redundant computations in FPS while preserving sampling quality, and \textit{FPS-Cache} eliminates layer-wise redundancy via cache-and-reuse. Integrated into existing CUDA libraries and state-of-the-art PNN accelerators, \textit{FlashFPS} achieves 5.16$\times$ speedup over the standard CUDA baseline on GPU and 2.69$\times$ on PNN accelerators, with negligible accuracy loss, enabling efficient and scalable PNN inference. Codes are released at https://github.com/Yuzhe-Fu/FlashFPS.

Long Chain-of-Thought (CoT) reasoning is pivotal for the success of recent reasoning models but suffers from high computational overhead and latency. While prior works attempt to compress CoT via external compressor, they often fail to align with the model's internal reasoning dynamics, resulting in the loss of critical logical steps. This paper presents \textbf{C}ompressing \textbf{R}edundancy in Chain-of-Thought via \textbf{I}ntrinsic \textbf{S}aliency \textbf{P}runing (\textbf{CRISP}), a framework that compresses CoT by exploiting the model's intrinsic saliency. Our analysis reveals a distinct phenomenon: the reasoning termination token \texttt{[object Object]} acts as an information anchor, where its attention pattern effectively demarcates essential reasoning from redundancy. Based on this finding, we design a policy that utilizes these intrinsic attention signals to guide atomic compression operations. In contrast to coarse-grained pruning strategies, CRISP strategically distills the reasoning chain to maximize information density while preserving logical coherence. Empirical results across various backbone models and mathematical datasets demonstrate that CRISP achieves a 50-60% reduction in token count without compromising accuracy, effectively mitigating the efficiency bottleneck of long-context reasoning. We open-source our implementation to facilitate further research in efficient reasoning.

Multimodal large language models (MLLMs) have recently demonstrated strong capabilities in understanding and generating responses from diverse visual inputs, including high-resolution images and long video sequences. As these models scale to richer visual representations, inference increasingly relies on storing large numbers of vision tokens in the key-value (KV) cache, making memory consumption a central bottleneck. Existing methods address this issue by identifying redundancy in vision tokens and compressing the cache, but such compression is typically applied only after all inputs are processed, resulting in high peak memory usage during the prefill stage. In this work, we show that MLLMs exhibit inherent structural regularities and representational redundancy that can be exploited to control memory growth throughout inference. Based on this insight, we propose a sequential input-compression mechanism that enforces a fixed memory budget by performing structure-aware key-value cache compression during the prefill process. This approach substantially reduces peak memory usage while maintaining generative performance with only minimal degradation, enabling more practical and memory-efficient multimodal inference.

Speech Large Language Models route speech encoder representations into an LLM decoder that typically accounts for over 90% of total parameters. We study how much of this decoder capacity is actually needed for speech tasks. Across two LLM families and three scales (1-8B), we show that decoder redundancy is largely inherited from the pretrained LLM: text and speech inputs yield similar redundant blocks. We then measure excess capacity by pruning decoder layers and analysing post-pruning healing to increase robustness. Our findings show that 7-8B models retain good ASR performance with only 60% of decoder layers, and the same trend extends to smaller scales with reduced pruning tolerance. We then generalise to speech translation, and show that the same blocks of layers are redundant across speech encoders, tasks and languages, indicating that a more global redundancy structure exists, enabling a single pruned and multi-tasks SpeechLLM backbone to be deployed.

Multimodal Transformers often produce predictions without clarifying how different modalities jointly support a decision. Most existing multimodal explainable AI (MXAI) methods extend unimodal saliency to multimodal backbones, highlighting important tokens or patches within each modality, but they rarely pinpoint which cross-modal feature pairs provide complementary evidence (synergy) or serve as reliable backups (redundancy). We present Feature-level I2MoE (FL-I2MoE), a structured Mixture-of-Experts layer that operates directly on token/patch sequences from frozen pretrained encoders and explicitly separates unique, synergistic, and redundant evidence at the feature level. We further develop an expert-wise explanation pipeline that combines attribution with top-K% masking to assess faithfulness, and we introduce Monte Carlo interaction probes to quantify pairwise behavior: the Shapley Interaction Index (SII) to score synergistic pairs and a redundancy-gap score to capture substitutable (redundant) pairs. Across three benchmarks (MMIMDb, ENRICO, and MMHS150K), FL-I2MoE yields more interactionspecific and concentrated importance patterns than a dense Transformer with the same encoders. Finally, pair-level masking shows that removing pairs ranked by SII or redundancy-gap degrades performance more than masking randomly chosen pairs under the same budget, supporting that the identified interactions are causally relevant. Code is available at https://github.com/dut0817/FL-I2MoE.

The effectiveness of multimodal instruction tuning depends not only on dataset scale, but critically on whether training samples genuinely require visual reasoning. However, existing instruction datasets often contain a substantial portion of visually redundant samples (solvable from text alone), as well as multimodally misaligned supervision that can degrade learning. To address this, we propose VisNec (Visual Necessity Score), a principled data selection framework that measures the marginal contribution of visual input during instruction tuning. By comparing predictive loss with and without visual context, VisNec identifies whether a training instance is vision-critical, redundant, or misaligned. To preserve task diversity, we combine VisNec with semantic clustering and select high-necessity samples within each cluster. Across 10 downstream benchmarks, training on only 15% of the LLaVA-665K dataset selected by VisNec achieves 100.2% of full-data performance. On the smaller Vision-Flan-186K dataset, our selection not only further reduces data size but also surpasses full-data training by 15.8%. These results demonstrate that measuring and leveraging visual necessity provides an effective solution for both efficient and robust multimodal instruction tuning. Codes and selected subsets will be released upon acceptance.

Time Series Foundation Models (TSFMs) leverage extensive pretraining to accurately predict unseen time series during inference, without the need for task-specific fine-tuning. Through large-scale evaluations on standard benchmarks, we find that leading transformer-based TSFMs exhibit redundant components in their intermediate layers. We introduce a set of tools for mechanistic interpretability of TSFMs, including ablations of specific components and direct logit attribution on the residual stream. Our findings are consistent across several leading TSFMs with diverse architectures, and across a diverse set of real-world and synthetic time-series datasets. We discover that all models in our study are robust to ablations of entire layers. Furthermore, we develop a theoretical framework framing transformers as kernel regressors, motivating a purely intrinsic strategy for ablating heads based on the stable rank of the per-head projection matrices. Using this approach, we uncover the specific heads responsible for degenerate phenomena widely observed in TSFMs, such as parroting of motifs from the context and seasonality bias. Our study sheds light on the universal properties of this emerging class of architectures for continuous-time sequence modeling.

This paper presents a large-scale budget-aware benchmark of multi-objective soft-computing metaheuristics for a bi-objective redundancy allocation problem in repairable k-out-of-n systems. The problem combines cost minimization and steady-state availability maximization under weight constraints, with binary subsystem-level decisions determining both the number of redundant components and the redundancy strategy. Four strategies are considered: cold standby, warm standby, hot standby, and a mixed active-warm standby strategy. Subsystem availability is evaluated using continuous-time Markov chains, while the optimization task is treated as a constrained mixed-integer multi-objective problem. The study compares 65 metaheuristic algorithms from multiple algorithmic families under two initialization settings, with and without Scaled Binomial Initialization (SBI), across six case studies of increasing structural and dimensional complexity and four weight limits per case. Performance is assessed using hypervolume, budget-dependent convergence profiles, and non-parametric statistical comparisons. The results show that algorithm rankings are strongly budget-dependent, indicating that a single final-budget ranking can be misleading. SBI provides a substantial early advantage and can change the relative performance of competing methods, especially for larger instances. The best-performing algorithms vary across budget regimes: NNIA-SBI and CMOPSO-SBI are competitive under tight budgets, whereas NSGA-II+ARSBX-SBI performs robustly for medium and large budgets. From a system design perspective, Pareto-optimal solutions are dominated by hot standby and mixed redundancy strategies, while cold and warm standby rarely appear. The benchmark highlights the importance of initialization, computational budget, and problem complexity when selecting soft-computing optimizers for practical redundancy allocation.

Associative memory, traditionally modeled by Hopfield networks, enables the retrieval of previously stored patterns from partial or noisy cues. Yet, the local computational principles which are required to enable this function remain incompletely understood. To formally characterize the local information processing in such systems, we employ a recent extension of information theory -- Partial Information Decomposition (PID). PID decomposes the contribution of different inputs to an output into unique information from each input, redundant information across inputs, and synergistic information that emerges from combining different inputs. Applying this framework to individual neurons in classical Hopfield networks we find that below the memory capacity, the information in a neuron's activity is characterized by high redundancy between the external pattern input and the internal recurrent input, while synergy and unique information are close to zero until the memory capacity is surpassed and performance drops steeply. Inspired by this observation, we use redundancy maximization at each neuron as an information-theoretic learning goal. This dramatically increases the network's memory capacity to 1.59, a more than tenfold improvement over the 0.14 capacity of classical Hopfield networks, and also outperforming recent state-of-the-art implementations of Hopfield networks. Overall, this work establishes redundancy maximization as a new design principle for associative memories and opens pathways for new associative memory models based on information-theoretic goals.

Large reasoning models (LRMs) often exhibit overthinking, producing verbose Chain-of-Thought (CoT) traces that increase inference cost and obscure the underlying reasoning process. Existing CoT compression methods mainly rely on global length rewards, which conflate necessary intermediate reasoning with redundant text and may therefore compromise reasoning fidelity. This paper revisits overthinking from a semantic-efficiency perspective and decomposes CoT redundancy into two distinct forms: internal redundancy, defined as informational stagnation before the first correct answer, and external redundancy, defined as superfluous continuation after the first correct answer. Based on this decomposition, we propose a dual-penalty reinforcement learning framework that separately optimizes reasoning progress and termination behavior. Specifically, a sliding-window semantic similarity metric penalizes low-progress reasoning segments, while a normalized external-redundancy metric discourages post-answer continuation. Experiments on GSM8K, MATH500, and AIME24 across different model scales show that our method reduces average reasoning length by 41.3% on the 1.5B model and 40.1% on the 7B model, while preserving competitive accuracy and achieving the best overall accuracy-efficiency score among evaluated baselines. The learned compression behavior further transfers to out-of-domain reasoning tasks, including GPQA and LiveCodeBench. More importantly, our analysis reveals a clear asymmetry between the two redundancy types: external redundancy can be largely removed with little performance loss, whereas internal redundancy compression follows a sensitive accuracy-efficiency trade-off. These results suggest that effective CoT compression should optimize semantic efficiency rather than sequence length alone, offering a principled route toward more concise, efficient, and interpretable LRMs.

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.