Tema

Recuperación y RAG

1216documentos

101últimos 7 días

SWHForest: A Structural Curation and Search Layer for Large-Scale Code Corpora

Pre-training code language models relies on massive code corpora usually streamed over the internet. However, researchers rarely need a multi-terabyte archive in its entirety; instead, their workflow centers on curated subsets: selecting well-documented files, filtering out repetitive boilerplate, auditing for benchmark contamination, or retrieving specific code patterns. Yet, because modern corpora are typically published as file pointers and basic attributes rather than source text, no general structural layer exists to select which files to download: curating a subset requires downloading and parsing millions of files first just to discard most of them. We present SWHForest, an open-source framework and public dataset that provides this missing structural layer for any pointer-based corpus without redistributing source code. A single pass over the archive parses each file into an AST and immediately discards the source text, keeping only its structural summary. Queried locally or exposed by archive providers as a server-side filtering API, SWHForest selects matching files first (via SQL or index intersection) and downloads only surviving candidates. Across the ten largest languages of Stack-Edu (52.1M unique files, 341M definitions), SWHForest answers curation queries in milliseconds while cutting downloads by 46-72%, reducing live structural searches from hours of per-language streaming to seconds or minutes.

arXiv AI Papershace 4 días

Flex-DDCM: Zero-Shot Task-Oriented Image Compression via Diffusion Trajectory Guidance

Recent zero-shot image compression methods exploit the powerful priors of pretrained diffusion models to produce perceptually pleasing reconstructions without compression-specific training. However, image compression serves not only human viewing but also machine analysis, while existing zero-shot codecs have limited flexibility in meeting the diverse reconstruction requirements of these applications. In this paper, we propose Flex-DDCM, a framework for zero-shot task-oriented image compression via diffusion trajectory guidance. Building on Turbo-DDCM, we adapt the coded diffusion trajectory through task-guided noise selection, embedding the effects of task guidance into the compressed representation. Guidance terms can be applied individually or combined, with timestep-dependent strengths controlling their influence throughout denoising. This design enables task adaptation without introducing additional decoding complexity. Experiments demonstrate flexible trade-offs among pixel fidelity, perceptual quality, and downstream task performance, with competitive performance against trained task-oriented codecs and faster decoding than Diff-ICMH.

arXiv AI Papershace 4 días

Decomposing Earth Embeddings with Sparse Autoencoders

Earth observation foundation models produce global metre-resolution embeddings that underpin work across agriculture, ecology, hydrology, and beyond. Yet these embeddings are opaque: each dimension entangles unrelated physical concepts, which makes them difficult to interpret and their robustness hard to assess. Here, we propose sparse autoencoders (SAEs) as a principled way to decompose Earth embeddings into sparse, approximately monosemantic features over an overcomplete dictionary. Because LLM judges of interpretations do not transfer to satellite pixels, and spatial autocorrelation matters, we introduce a tailored evaluation protocol. It scores any SAE on reconstruction at matched sparsity, alignment with reference concepts, and spatial scale, each against baselines that include an SAE whose decoder is a frozen random dictionary. On this protocol, SAEs reconstruct AlphaEarth and Tessera embeddings better than every baseline at matched sparsity. Their features align with far fewer reference concepts than raw embedding dimensions (on AlphaEarth, 88% align with at most one of 20 concept groups, most with none, against a median of 12 per raw dimension). They also carry more of their variance at fine spatial scales (23 to 32% within a 640 m patch for BatchTopK and Matryoshka, against 9%). An SAE with a frozen random decoder shares both properties, which therefore follow from sparsity, whereas only a learned dictionary reconstructs the embedding accurately: on AlphaEarth, trained SAEs leave eight times less variance unexplained. We test the sparse code on two applications, discovering land-cover sub-concepts and detecting out-of-distribution pixels by area-of-applicability estimation. Both indicate how sparse features could be used, but neither is yet reliable enough for practical use. Together, these results are a step toward Earth observation models whose features can be inspected.

arXiv AI Papershace 4 días

Search-by-Answer Bias in Retrieval Models

Dense retrievers are meant to select passages that match a query's semantic specification. We show they suffer from search-by-answer bias: they favor passages that mention answer objects the model knows (scores as likely), over passages that support valid alternative answers. We test this with controlled factual contrasts from LAMA/TREx. For each retriever, we measure how strongly the model scores the canonical object over an alternative object using a token-prediction head, and ask whether this score predicts passage rankings. Across eight retrievers, including DPR, Spider variants, Contriever, E5, and Qwen3-Embedding, larger token-prediction scores consistently predict higher retrieval scores for passages containing the corresponding object. This holds when passages differ only in the answer object, and even when the passage that mentions the high-scoring object expresses a wrong relation. Dense retrieval thus exhibits search-by-answer bias during passage selection, shaping which evidence is retrieved.

arXiv AI Papershace 4 días

The comparison of density estimation for transfer learning in Banach, Hilbert, Euclidean spaces

Domain adaptation or DA methods are indispensable when target labels or computational resources are scarce, yet the domain kernel density and subspace alignment approaches completely assume a Hilbertian geometry. They measure distributional discrepancy with symmetric L2 divergences over Euclidean feature spaces or reproducing kernel Hilbert spaces (RKHS). This preprint summarizes the manuscript Non-Hilbertian Domain Adaptation. Total Variation Alignment via L1 Density Regularization on Feature Manifolds, which proposes Banach Manifold Transfer Learning or BMTL. BMTL makes two coupled design decisions. The first one is that features are embedded on a geodesic manifold rather than treated as points of a flat Euclidean space, and the second one is that the density matching regularizer is formulated in an L1 Banach norm, whose minimization is equivalent, by Scheffe's theorem, to contracting the Total Variation or TV distance between domain densities, a bounded metric that weights all regions of the sample space uniformly. The L2 formulation, by contrast, squares deviations and therefore amplifies errors in the tails, over emphasizing density outliers that do not reflect meaningful distributional shift. Experiments on MNIST-USPS, COIL20, and Office+Caltech10 show consistent gains; the L1 variant beats an otherwise identical L2 ablation on every task, for example 92.44% vs. 89.80% on USPS->MNIST and surpasses the strongest baseline DMDA by +3.68% on average over twelve Office+Caltech transfers. This document situates the contribution with an explicit comparison of Euclidean, Hilbert, and Banach geometries.

arXiv AI Papershace 4 días

D3Tree: Learning to Prune Block-Diffusion Draft Trees for Efficient Verification

Block-diffusion drafters propose a whole block of tokens in one forward pass, and draft trees built from their position-wise marginals increase the expected accepted length of speculative decoding. Every tree node occupies a query row in target verification, so under concurrent serving low-yield branches compete with other requests for verification capacity. Existing tree methods select verified nodes by draft probability under a fixed or cost-selected per-request budget, and thus always retain a prefix of the drafter's ranking. We present D3Tree, which learns which nodes of a block-diffusion draft tree are worth verifying. A tree-size head with 2H+2 parameters predicts each candidate's conditional survival from the draft state, the candidate embedding, and its draft log-probability. Thresholding the product of these scores along parent chains yields an ancestor-closed subtree whose size adapts to each request and decoding round. The head is trained on a frozen drafter with conditional labels on the full candidate tree; an optional joint mode also steers the drafter toward the retained reference path. At serving time, D3Tree verifies whole trees under light load and, under heavy load, allocates one batch-wide node budget by predicted survival, sized to the target's matrix-multiplication tiles. Retained trees share one ancestor-masked target pass, while construction, pruning, and scheduling run on the GPU and overlap with the host. Across six vision-language verifiers and 17 text and multimodal benchmarks, D3Tree is fastest in 70 of 72 load settings, with 3.2-4.0x speedup over autoregressive decoding for one request and 1.3-1.8x at 64 concurrent requests, where fixed-budget trees fall below autoregressive speed. Under a time-varying arrival trace, it has the lowest mean end-to-end latency in every arrival phase.

arXiv AI Papershace 4 días

AOP-Smart-Finder: A High-Throughput Human-AI Hybrid Framework for AOP Discovery and Evidence Mapping from Large-Scale Literature

The rapid growth of scientific literature and the high heterogeneity of biological terminology and mechanistic descriptions make it difficult for researchers to integrate scattered research evidence, identify connections among biological events, and construct mechanistic networks. Here, we developed a high-throughput human-AI hybrid framework, AOP-Smart-Finder, for biological event discovery, relationship mining, and Adverse Outcome Pathway (AOP) construction and evidence mapping from large-scale literature. This framework integrates Large Language Models (LLMs), AOP-Smart retrieval-augmented generation (RAG) technology, and Python-based automation, supporting two analysis modes: one based on a single compound and the other based on a predefined set of biological events. In the di(2-ethylhexyl) phthalate (DEHP) case, the system processed 12,983 articles within 30 min and included 1,093 of them, from which it extracted 1,589 biological events (including 458 official key events [KEs] and 1,131 candidate KEs) and 2,519 key event relationships (KERs). In the amyotrophic lateral sclerosis (ALS) case, the system identified 55 stressors from 437 articles and extracted 278 biological events and 548 event relationships. Following manual curation and Weight of Evidence (WoE) evaluation, a mechanistic network comprising 16 KEs and 23 KERs was ultimately constructed. This framework can either use existing AOPs as knowledge guidance for mechanistic expansion and evidence supplementation, or discover novel event connections and construct mechanistic networks around user-specified research topics. It provides a high-throughput, traceable, human-AI hybrid methodology for complex biological mechanism elucidation and AOP discovery driven by large-scale literature.

arXiv AI Papershace 4 días

Verify Before You Fetch: Reliable KV Cache Compression for LLM Inference

Long-context applications, long-lived agents, and multi-agent systems require large language models (LLMs) to process and retain extensive information. In these settings, the KV cache can exceed GPU high-bandwidth memory (HBM) capacity, necessitating offloading to host memory, where CPU-to-GPU transfers can dominate inference latency. While KV cache compression can reduce data movement, query-agnostic compression that guarantees accurate attention for arbitrary queries is extremely challenging, if not impossible. We introduce VerKV, a hybrid approach that provides (ε, δ) guarantees for approximating attention on any given query. VerKV maintains a compact set of query-agnostic pivotal tokens on the GPU, together with a token-level sketch of the full KV cache, while retaining the complete cache in CPU memory. We introduce a novel stochastic sampling procedure, PoissonComp, to select the pivotal tokens. At inference time, our proposed KVFetch algorithm uses the pivotal tokens and the sketch to estimate the relative error of the resulting attention computation. If the estimate satisfies a user-specified (ε, δ) tolerance, inference proceeds without accessing host memory. Otherwise, VerKV performs query-aware retrieval to fetch additional tokens from the CPU, refining the attention computation to meet the required tolerance. Thus, VerKV enables attention with bounded approximation error while minimizing CPU-to-GPU transfers. We show that PoissonComp outperforms state-of-the-art approaches to KV cache compression. Combined with KVFetch, it achieves near-dense performance (90-100% of dense quality at 32K context length, averaged over RULER and LOFT), using Qwen3-4B, Qwen3.5-9B and Ministral-3-8B, while reducing the GPU memory footprint by 8-32x and keeping CPU-to-GPU memory fetches within 2-23% of the cache per generated token, depending on the model.

arXiv AI Papershace 5 días

DeepFinance: Can AI Diligence a $100MM Deal?

Existing evaluations of AI models on high-finance work focus on retrieving answers from financial documents and producing numerically correct outputs. Real private equity (PE) and investment banking (IB) work also depends on choosing between conflicting sources, making sound assumptions, and keeping every step auditable. We introduce DeepFinance, an ultra long-horizon benchmark of 21 end-to-end PE and IB tasks that take a human professional an average of 27.1 hours. Each task pairs a request with a data room averaging 19 files and is graded against an average of 84 expert-written rubric criteria covering output, workflow, and judgment. We evaluate 22 model configurations, each in its native agent harness. The best configuration, GPT-6 Astra with max reasoning, satisfies 57.6% of rubric criteria, and the top four configurations are not statistically distinguishable. Frontier models often calculate correctly from the wrong inputs and find decision-relevant evidence without acting on it.

arXiv AI Papershace 5 días

Zero-Shot VLA Adaptation with Retrieved Guidance

Pre-training large parametric "Vision-Language-Action" (VLA) models on large demonstration datasets has recently yielded promising results for general-purpose manipulation. However, these models still struggle under even small domain shifts relative to the training data, such as changes in camera viewpoint, background, object identity, or robot morphology. Motivated by the success of Retrieval-Augmented Generation (RAG) in language models, we pursue retrieval to boost VLA performance. Specifically, we seek to leverage retrieval from domain-mismatched, "in-the-wild" demonstration datasets like DROID to guide a VLA's behavior in a new target domain. To achieve this, we introduce RAGu, which overcomes robotics-specific challenges for RAG: (1) It incorporates simple geometric calibration to transform source demonstrations into the target setup and reduce cross-domain differences. (2) It uses a coarse-to-fine retrieval strategy that selects, from a large pool of pre-warped candidates, the trajectory most compatible with the base VLA. (3) It guides the VLA's flow-matching inference procedure rather than seeking to retrain the base model or expand its limited context window. We evaluate RAGu across real-world, simulation, sim-to-real transfer, and cross-embodiment humanoid manipulation. We find it improves real-world VLA success rates from 20% to 80% and outperforms the closest-related baselines in sim. Project website: raguvla.github.io

arXiv AI Papershace 5 días

Interactive Person Retrieval with On-the-Fly Visual Feedback

Traditional text-based person retrieval (TPR) assumes that a witness provides a complete description at once, whereas interactive person retrieval (Inter-PR) refines a partial description through question-answer exchanges. In this setting, the questioner inspects retrieved candidates to decide what evidence to request next, while the answerer sees only the witness-side memory and cannot report candidate-specific visual differences. Existing methods mainly learn offline question orderings, leaving this visual-feedback loop underexplored. To address this limitation, we introduce visual-feedback interactive person retrieval (VFI-PR), a new interactive retrieval paradigm in which retrieved candidates produce visual feedback that becomes evidence for subsequent questioning and re-ranking. To support training and evaluation in this setting, we construct VFI-PEDES, a dataset and evaluation protocol for visual-feedback interaction, and propose VFI-ReID, a framework that learns to use visual feedback during interactive person retrieval. Through the Dynamic Accumulated-evidence Retriever (DAR) and retrieval-aware question optimization, VFI-ReID adapts retrieval and question generation to evolving evidence states, turning visual feedback into complementary evidence for subsequent retrieval. Experiments show that VFI-ReID consistently improves retrieval across VFI-PR interaction stages, demonstrating the utility of visual feedback for interactive person retrieval and outperforming existing Inter-PR baselines.

arXiv AI Papershace 5 días

Polynomial Denoisers on the Image Manifold: Volterra Networks for Self-Contained Pixel Diffusion

Natural images are often conjectured to lie on a low-dimensional manifold within the high-dimensional pixel space. Recent work leverages this to build self-contained pixel-space diffusion models without specialized machinery or pretrained components. A plain Transformer on raw image patches suffices, provided it predicts the clean image rather than the noise and compresses each patch through an aggressive bottleneck. The mixing operator, however, has remained softmax attention, inherited from sequence modeling rather than derived from denoising. For Gaussian noise, the optimal denoiser admits a canonical expansion in polynomial functionals of the observation, the Wiener--Volterra series. We prove that softmax attention, and any normalized mixing rule (including linear attention), cannot realize any polynomial map beyond an affine one. In contrast, Volterra layers realize this class directly and, on a data manifold, achieve approximation rates governed by the manifold dimension rather than the ambient pixel dimension. We therefore propose the architecture these results prescribe; raw image patches passed through a bottleneck embedding, followed by cascaded low-rank Volterra layers, which carry no activation function because each layer is nonlinear in its own right. To our knowledge this is the first pixel-space diffusion architecture whose mixing operators are Volterra layers throughout, with no attention, no convolution, no pretrained components, and no activation function on the residual path. On ImageNet 256 256 and 512 512, the resulting model is competitive with self-contained pixel-space diffusion at matched parameter and FLOP budgets.

arXiv AI Papershace 5 días

A Geometry-Guided Feature-Space Framework for Interpretable Unsupervised 4D-STEM Analysis of Complex Multiphase Materials

Rapid, high-fidelity analysis of four-dimensional scanning transmission electron microscopy (4D-STEM) datasets from heterogeneous materials remains challenging because coexisting phases, orientation variants, local strain gradients, and intrinsic TEM thin-foil effects produce complex, overlapping diffraction signatures. Here, we present a machine-learning (ML) framework that combines physics-informed geometric descriptors with weighted high-dimensional embeddings to identify latent structural patterns with minimal prior knowledge. By constructing a regularized descriptor space before embedding and density-based clustering, the method improves cluster stability, suppresses noise-driven fragmentation, and reduces sensitivity to embedding hyperparameters. Spatial weighting produced the largest stability gain, increasing median cluster persistence from 0.04 to 0.55 while reducing the noise fraction from approximately 22% to 0%. Rotation weighting increased DBCV from 0.80 to 0.92, although the noise fraction rose from 0.2% to 1.0%, highlighting a trade-off between improved density separation and noise rejection. We validate the framework on two benchmark systems: Si/Si_xGe_{1-x} epitaxial layers with continuous lattice variation and deformation, and multiphase Ti-Nb alloys containing distinct phases and orientation variants. Comparisons with a pretrained vanilla autoencoder and model-free pattern vectorization show that the physics-informed geometric descriptor provides higher cluster fidelity, greater interpretability, and substantially lower dimensionality.

arXiv AI Papershace 5 días

KeyASRBench: Measuring Whole-Phrase Keyword Preservation in Real-World ASR

Public automatic speech recognition (ASR) benchmarks increasingly report low sequence error rates, yet production systems still use hotword lists, contextual biasing, retrieval, and post-correction. This motivates a complementary question: does a transcript preserve selected names, products, technical terms, versions, and relational phrases? We present KeyASRBench, a bilingual pilot that operationalizes occurrence-based Keyword Error Rate (KER) over frozen whole-phrase annotations. The 10-hour model-conditioned challenge set contains 1,501 Chinese and English clips with 4,723 keyword occurrences. Across five locally inferred checkpoints and two official API systems, language-specific MER/WER ranges from 5.31% to 12.91%, while KER ranges from 13.61% to 33.81%. A disjoint 30-minute model-blind control yields KER of 10.65%-27.31%. The control confirms that selected-phrase errors extend beyond challenge-set selection.

arXiv AI Papershace 5 días

Learning Biological Hierarchies in Single-Cell Foundation Models

Cell types are organized by taxonomic relationships that define lineages in the so-called cell ontology. However, most existing foundation models typically ignore cell-type lineages encoded in this cell ontology. We introduce a biologically informed foundation model called scOntoFM, which embeds hierarchical cell ontology into representation learning. Using Lowest Common Ancestor (LCA) distances from the ontology graph, our framework pairs efficient offline triplet sampling with hierarchical cell-ontology learning. This approach jointly preserves local neighborhoods and shapes the global embedding geometry to reflect the ontology structure. Across diverse cell- and gene-level benchmarks, the model consistently improves performance, with especially strong gains in zero-shot settings. Our model also balances batch integration with biological structure preservation, providing complementary perspectives to existing embedding evaluation frameworks and supporting scalable biological discovery.

arXiv AI Papershace 5 días

WARP-JEPA: World-Model-in-the-Loop Amortized Rollout Planning for Frozen Hierarchical Joint-Embedding Predictive Architectures

Hierarchical Joint-Embedding Predictive Architectures (H-JEPA) enable long-horizon visual control but rely on costly test-time gradient descent for planning. Our analysis reveals that iterative optimization consumes up to 97.7% of episode runtime and that low predicted rollout costs do not necessarily translate into successful control. We further identify a planning-horizon asymmetry between long-horizon top-level composition and short two-step subordinate planning. To address these challenges, we propose WARP-JEPA, a hierarchical amortized planning framework that replaces online gradient optimization with learned multi-candidate trajectory proposers while keeping the world models frozen. WARP-JEPA combines a lightweight world-model-in-the-loop causal Transformer at the top level with parallel residual MLPs at subordinate levels. The proposers are trained using model-based rollout costs and best-of-K behavior cloning, enabling efficient candidate selection without test-time gradient optimization. Across 30 frozen H-JEPA models on four benchmarks, WARP-JEPA achieves 9.6-18.8x faster episode inference on long-horizon tasks (2.9-3.0x on Cube) and improves task success by up to 33.3 percentage points. These results demonstrate that amortized planning can substantially improve both inference efficiency and control performance without retraining the world models. Code is available at https://github.com/hideinsewer/WARP-JEPA.

arXiv AI Papershace 6 días

Sampling Is Not Curiosity: Why LLM Agents Should Investigate

LLM agents are increasingly used for open-ended tasks, very common in research scenarios where they are required to explore, investigate, or act curiously. At inference time, an LLM is a fixed map from context to next-token distribution, the pretrained weights determine the function, and only its conditioning changes as context accumulates. Mechanisms like temperature, best-of-N, tree search, or RAG vary how the distribution is sampled or what it is conditioned on, but do not update the agent's model of the environment. We argue that current LLM agents do not explore curiously; they explore stochastically. From how curiosity is understood in the RL literature, it requires a predictive structure over how the environment responds to the agent's actions, which improves online from current interaction, drives action selection through its predicted improvement, and persists beyond the context window. Additionally, in its open-ended form, the agent should intrinsically originate what to investigate. Thus, we propose a framework defining curiosity in LLM agents, reserving the term for systems that satisfy it. In doing so, we identify the research directions its implementation opens up.

arXiv AI Papershace 6 días

PatternBloom: Empowering Agentic RAG with Externalized RL-Distilled Graph Patterns

Reinforcement-learning-trained agents for retrieval-augmented generation retrieve content (passages, knowledge-graph fragments, or hyperedges) and supervise training with outcome rewards on the final answer. Across these paradigms the reasoning route through retrieved content never leaves the language model's hidden state, conflating declarative knowledge with the procedural knowledge that recurs across queries of the same logical type; content-only methods consequently plateau as reasoning depth grows. We introduce PatternBloom, an agentic RAG framework that shifts the unit of retrieval from content (what to read) to logic (how to reason). A first RL stage scores the agent's constructed evidence graph by a frozen oracle's information gain through the Information-Density Reward (IDR), a size-normalized mutual-information surrogate. High-reward trajectories are distilled into a Graph Pattern Memory (GPM) of type-abstracted reasoning skeletons, and a second RL stage uses the Pattern-Augmented Reward (PAR) to couple policy to memory, turning procedural reasoning into a learnable, externalized structural prior. A 7B PatternBloom outperforms 27 baselines by +15.0 Avg-OOD EM, with the gain compounding with reasoning depth precisely because procedural knowledge, once distilled, is reused rather than re-derived at every query. We release our code at https://github.com/EllenYiGe/PatternBloom.

arXiv AI Papershace 6 días

Mental Imagery That Matters: Curing Latent Collapse in Multimodal Reasoning

Mental simulation, the ability to reason by constructing and updating internal visual states, is central to flexible human cognition, but remains elusive in current multimodal large language models. Latent multimodal reasoning offers a promising route toward this capability by interleaving text with continuous visual latent blocks, allowing models to maintain internal visual workspaces without decoding them into pixels. In this paper, we ask whether these latent visual states actually matter. Through a direct perturbation analysis across representative latent-reasoning MLLMs, we find that they often do not: dropping, replacing, or shuffling generated visual latents has only minor effects on the model's own final-answer probability, while equivalent perturbations to the linguistic reasoning trace are substantially more damaging. We call this failure mode Mental Simulation Collapse (MSC): the model appears to perform interleaved visual reasoning, but routes most answer-relevant computation through text. We further diagnose that MSC follows naturally from existing training objectives: visual alignment can make generated latents resemble helper-image embeddings, but it does not require the decoder to rely on them; subsequent text-only supervision further leaves the semantic role of the latent pathway unidentifiable. To address this, we introduce AnchorSim (Anchored Mental Simulation), a contrastive distillation framework that trains models to make their generated visual latents consequential. Across multiple visual reasoning benchmarks, AnchorSim improves performance, while diagnostic re-evaluation shows that visual perturbations now meaningfully affect predictions. Our results expose a fundamental failure mode of latent multimodal reasoning and demonstrate that mental simulation must be anchored to model behavior in order to matter.

arXiv AI Papershace 6 días

Beyond Query Similarity: Safety-Conditioned Cache Routing for LLM Response Reuse

Semantic response caching reuses a cached response for a semantically similar query to improve LLM inference efficiency. However, query similarity alone can lead to unsafe reuse when safety conditions change. Existing approaches improve cache safety through stricter similarity checks, context-aware validation, or response verification, yet they may still miss safety conflicts between the cached response and the new request or introduce unnecessary verification overhead. We argue that safe and efficient semantic caching requires a reuse policy that makes the cached response's safety conditions explicit and uses them to guide the reuse decision with a balance between safety and utility. In this paper, we propose SafeCache, a safety-aware framework that separates candidate retrieval from safety approval and formulates semantic response reuse as a safety-conditioned routing problem. SafeCache matches each cached-response condition against the new request and aggregates the per-condition mismatches into an overall risk score. A separate contradiction guard prevents explicit conflicts from being diluted by strong matches. It further estimates matching uncertainty and calibrates the routing policy under a required unsafe-reuse target and a verifier-call budget to choose among direct reuse, selective verification, or regeneration. Experiments show that SafeCache achieves the better safety-utility trade-off among the evaluated baselines, enabling safer response reuse at comparable cache utility.