Pith. sign in

REVIEW 3 major objections 5 minor 128 references

FreqDepthKV compresses long-context LLM key-value caches by sharing low-frequency depth components across layers while keeping sparse high-frequency residuals that protect retrieval and reasoning.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-11 00:10 UTC pith:FDDICGXT

load-bearing objection Plausible MiniCache extension with a clean frequency+routing idea, but Table 1 is uncheckable until they name the model and clean the paper. the 3 major comments →

arxiv 2607.06519 v1 pith:FDDICGXT submitted 2026-07-07 cs.AI

FreqDepthKV: Frequency-Guided Depth Sharing for Robust KV Cache Compression in Long-Context LLM Inference

Anna C\'ordoba , Adam Puente Tercero , Nerea Angulo Hijo , Mar Linares Tercero , Julia Barrientos , Ainhoa Miranda , Jes\'us Olivera This is my paper
classification cs.AI
keywords KV cache compressionlong-context LLM inferencedepth-frequency factorizationattention head routinghigh-frequency residualsreconstruction-aware lossinference-time compression
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Long-context language model inference is bottlenecked by the memory and bandwidth of key-value caches, yet aggressive compression often erases the layer-specific evidence that retrieval and multi-step reasoning need. This paper introduces FreqDepthKV, an inference-time method that factorizes adjacent-layer caches into shared low-frequency depth components and sparse high-frequency residuals, then uses a lightweight prefill probe to assign each attention head to shared-depth, residual-depth, or exact mode according to how compression would change attention logits. The policy adapts to prompt structure without retraining. On long-context QA, needle retrieval, summarization, and code generation with a 32k-token prefill, it closely matches full-cache accuracy while cutting peak KV memory to 6.2 GB (about 3.9 imes compression) and raising decoding throughput to 70.4 tokens/s. A sympathetic reader cares because many long prompts hide decisive evidence in a few token-head-layer interactions that average-error depth sharing would discard.

Core claim

FreqDepthKV establishes that inter-layer KV redundancy is frequency-structured: sharing low-frequency depth components while selectively preserving sparse high-frequency residuals—and exact entries for reconstruction-sensitive heads—lets inference-time compression nearly match full KV accuracy on long-context QA, summarization, and code tasks at roughly 3.9× lower peak KV memory and higher throughput.

What carries the argument

Depth-frequency factorization with reconstruction-aware head routing: adjacent-layer KV states are transformed with a fixed DCT basis into shared low-frequency components and sparse high-frequency residuals; a prefill probe routes each head to shared-depth, residual-depth, or exact mode by minimizing a memory-penalized attention-logit reconstruction loss.

Load-bearing premise

A one-shot prefill probe over a small set of query positions is assumed to be a stable enough proxy for which heads must keep residual or exact caches so that answers stay correct throughout decoding without re-routing.

What would settle it

On a long needle-retrieval or multi-step code task, if the prefill probe routes heads to shared-depth mode yet later decoding produces wrong answers that full KV gets right—and restoring residual or exact storage for those same heads recovers correctness—then the prefill logit proxy fails as a stand-in for end-to-end task fidelity.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Aggressive depth compression can retain retrieval and code accuracy when high-frequency residuals are kept for logit-sensitive heads.
  • A prefill-only reconstruction-aware router adapts compression to prompt structure without model retraining.
  • Depth-frequency sharing stacks with token eviction and quantization for joint sequence, precision, and depth savings.
  • Mixed layer-block sizes (tighter near model boundaries) balance memory and quality better than uniform blocks.
  • At 32k prefill, peak KV memory can fall to 6.2 GB with 70.4 tokens/s throughput while closely matching full-KV task scores.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Generation-time re-routing, which the paper only sketches as future work, may be necessary when relevant evidence shifts mid-decode in multi-turn or long-generation settings.
  • Assigning lower bit-width to shared low-frequency coefficients than to sparse residuals could cut bandwidth further without changing the routing logic.
  • The same logit-sensitivity probe could identify heads that are chronically non-compressible across models, guiding architecture choices toward more cache-friendly layers.
  • Prompts with diffuse evidence and weak document boundaries may force more exact-mode heads, shrinking the compression advantage relative to retrieval-heavy workloads.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes FreqDepthKV, an inference-time KV-cache compression method that stacks adjacent-layer keys/values, applies a fixed DCT depth transform (Eq. 1), and stores a shared low-frequency component plus sparse high-frequency residuals. An online prefill probe routes each head to shared-depth, residual-depth, or exact mode by minimizing a reconstruction-aware attention-logit loss plus memory penalty (Eq. 2–3). On a 32k prefill setup the method reports 58.3 EM / 63.0 F1 / 32.5 ROUGE-L / 48.1 pass@1, nearly matching Full KV while reducing peak KV memory to 6.2 GB (3.9×) and raising throughput to 70.4 tokens/s (Table 1), with ablations in Table 2 attributing gains to the frequency factorization, residuals, and routing.

Significance. If the empirical claims hold, the work is a useful systems contribution: it refines MiniCache-style depth sharing with frequency residuals and logit-aware head routing, remains training-free, and is complementary to token eviction and quantization. The design is concrete (DCT blocks, three cache modes, residual scoring) and the ablations isolate components. Significance is currently limited by incomplete experimental provenance—the base model is unnamed, budgets and variance are underspecified, and the reference list is unreliable—so the headline Table 1 deltas cannot yet be treated as established facts.

major comments (3)
  1. §4 Experiments never names the base long-context decoder (architecture, size, or checkpoint), only “the same base long-context decoder with a 32k-token prefill window.” Without this, Table 1’s Full-KV and baseline numbers (including the MiniCache comparison that is the main depth-sharing foil) are not reproducible or comparable to the literature, so the central claim that FreqDepthKV nearly matches Full KV at 3.9× cannot be verified.
  2. Table 1 / §4: no error bars, multi-seed runs, or multi-model results are reported, and compression budgets are “tuned on a held-out subset of LongBench” without disclosing the selected budgets or residual rates r_b,h for FreqDepthKV or baselines. Aggregate EM/F1/ROUGE/pass@1 and systems metrics are therefore hard to interpret as stable gains rather than single-run outcomes under an opaque budget protocol.
  3. §3, Eq. (2) and Future Work: the load-bearing assumption that a one-shot prefill probe over |P|=128 positions (recent tokens, document boundaries, high-entropy rows), using key-induced logit reconstruction, is a sufficient proxy for which heads must keep residual/exact storage throughout decoding is asserted but not stress-tested. The paper itself notes routing is fixed after prefill; without generation-time re-routing experiments or failure cases on multi-step/needle tasks, the claim that task answers remain correct under aggressive residual sparsity is only weakly supported.
minor comments (5)
  1. Header claims “37th Conference on Neural Information Processing Systems (NeurIPS 2023)” while the arXiv stamp is 2026; this should be corrected.
  2. References contain many off-topic PMC/sensor/biology/fluid-dynamics entries unrelated to KV cache compression; the bibliography needs a thorough cleanup.
  3. Figure 1 is described as summarizing routing patterns in §4, but the caption only restates the high-level idea; a figure that actually shows head-mode assignments would help.
  4. Notation: B is used both for block size and (implicitly) for the DCT basis F_B; residual rate r_b,h and the exact split of “first coefficient group” vs remaining groups should be stated more precisely.
  5. Systems metrics (tokens/s, TTFT, peak GB) lack hardware/software stack details (GPU, batch size, kernel fusion), which limits interpretability of the 70.4 tokens/s and 6.2 GB claims.

Circularity Check

0 steps flagged

No significant circularity: empirical systems method with external task metrics, not a self-forcing derivation.

full rationale

FreqDepthKV is an inference-time engineering method, not a first-principles derivation that claims to force task accuracy from its own inputs. The depth-frequency factorization (Eq. 1) and reconstruction-aware routing loss L_b,h(m) (Eq. 2) define how heads are assigned to shared/residual/exact modes; those choices minimize a measured logit-reconstruction-plus-memory objective during prefill. Downstream claims (EM, F1, ROUGE-L, pass@1, tokens/s, TTFT) are then reported as separate empirical measurements on LongBench, Needle-in-a-Haystack, summarization, and code benchmarks against Full KV and prior compressors (Table 1, ablations in Table 2). Nothing in the equations algebraically implies those task scores. Tuning λ to a target compression budget is ordinary systems hyperparameter selection and does not rename a fit as a prediction of accuracy. Citations to MiniCache and token/quantization baselines are external prior art by other authors, not load-bearing self-citation uniqueness theorems. There is no self-definitional loop, no fitted parameter re-labeled as the headline result, and no ansatz smuggled in as a forced theorem. Score 0 is appropriate; experimental unverifiability (unnamed base model, etc.) is a reproducibility concern, not circularity.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 3 invented entities

The central accuracy-under-compression claim rests on empirical design choices (block size, probe, λ budget, residual rate) and domain assumptions about inter-layer redundancy and logit sensitivity, not on free physical constants or formal theorems. Invented entities are algorithmic constructs (modes, residual index sets, routing loss), not new physical objects. No independent external evidence is given beyond the paper's own tables.

free parameters (5)
  • layer block size B (default 4; 2 at boundaries)
    Hand-chosen depth-block width controlling how many layers share low-frequency coefficients; ablated as B=2/8.
  • memory penalty λ
    Tuned so estimated peak KV footprint hits ~3.8–3.9× compression; directly steers mode assignment.
  • residual retention rate r_b,h
    Fraction of tokens keeping high-frequency residuals, chosen from routing loss budget.
  • probe size |P|=128 and probe sampling policy
    Number and selection of prefill query positions used to score reconstruction-sensitive heads.
  • DCT/frequency split (first coefficient group as shared low-frequency)
    Ad hoc partition of depth spectrum into shared vs residual bands without learned basis.
axioms (4)
  • domain assumption Adjacent transformer layers have substantial low-frequency KV redundancy that a fixed short orthonormal depth transform (DCT) can separate from sparse high-frequency residuals.
    Core premise of §3 Eq. (1) and MiniCache lineage; not proved, only motivated.
  • ad hoc to paper Key-induced attention-logit reconstruction error on a small prefill probe set predicts which heads need residual/exact storage for correct long-context task behavior.
    Routing objective Eq. (2) and mode assignment; load-bearing and not independently validated outside reported tables.
  • ad hoc to paper Budgeted routing decided once after prefill can be reused throughout autoregressive decoding (optional rare refresh).
    Stated in §3; Future Work notes generation-aware routing may be needed.
  • domain assumption Standard transformer attention and KV caching semantics hold; method is inference-only and needs no weight updates.
    Background systems assumption throughout method and experiments.
invented entities (3)
  • shared-depth / residual-depth / exact head cache modes no independent evidence
    purpose: Per-head discrete compression policy adapting to prompt structure.
    Defined by the paper's routing procedure; no external independent definition.
  • depth-frequency factorization of stacked adjacent-layer KV (Z = F_B X) no independent evidence
    purpose: Separate shared low-frequency depth components from sparse high-frequency residuals.
    Algorithmic representation introduced in §3; evidence is only internal ablations.
  • reconstruction-aware routing loss L_b,h(m) over cached attention logits no independent evidence
    purpose: Choose compression mode by measured logit change plus memory cost.
    Paper-specific objective; not a standard external metric with independent calibration.

pith-pipeline@v1.1.0-grok45 · 16852 in / 3592 out tokens · 42224 ms · 2026-07-11T00:10:34.383023+00:00 · methodology

0 comments
read the original abstract

Long-context LLM inference is increasingly limited by the memory and bandwidth cost of KV caches, yet aggressive compression can remove the layer-specific evidence needed for retrieval and multi-step reasoning. We introduce FreqDepthKV, an inference-time cache compression method that factorizes adjacent-layer KV states into shared low-frequency depth components and sparse high-frequency residuals. A lightweight online probe assigns attention heads to shared-depth, residual-depth, or exact cache modes according to their contribution to reconstruction-sensitive attention logits, allowing the compression policy to adapt to prompt structure without retraining. Across long-context question answering, needle retrieval, summarization, and code generation benchmarks, FreqDepthKV preserves task accuracy under substantially smaller cache budgets. With a 32k-token prefill window, FreqDepthKV reaches 58.3 Exact Match, 63.0 F1, 32.5 ROUGE-L, and 48.1 pass@1, closely matching full KV while outperforming prior compressed-cache methods. It also improves decoding throughput to 70.4 tokens/s, reduces TTFT to 2.06 seconds, and lowers peak KV memory to 6.2 GB, achieving a 3.9x effective compression ratio.

Figures

Figures reproduced from arXiv: 2607.06519 by Adam Puente Tercero, Ainhoa Miranda, Anna C\'ordoba, Jes\'us Olivera, Julia Barrientos, Mar Linares Tercero, Nerea Angulo Hijo.

Figure 1
Figure 1. Figure 1: Overview of the core idea: FreqDepthKV removes redundant depth information shared [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Architecture of FreqDepthKV: KV states are decomposed across layer depth into shared [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

128 extracted references · 128 canonical work pages · 7 internal anchors

  1. [1]

    2026 , journal =

    PolyKV: A Shared Asymmetrically-Compressed KV Cache Pool for Multi-Agent LLM Inference , author =. 2026 , journal =

    work page 2026

  2. [2]

    2024 , journal =

    MiniCache: KV Cache Compression in Depth Dimension for Large Language Models , author =. 2024 , journal =

    work page 2024

  3. [3]

    2026 , journal =

    Make Each Token Count: Towards Improving Long-Context Performance with KV Cache Eviction , author =. 2026 , journal =

    work page 2026

  4. [4]

    2026 , journal =

    ReST-KV: Robust KV Cache Eviction with Layer-wise Output Reconstruction and Spatial-Temporal Smoothing , author =. 2026 , journal =

    work page 2026

  5. [5]

    2026 , journal =

    The risk of KV cache compression , author =. 2026 , journal =

    work page 2026

  6. [6]

    2025 , journal =

    Attention Is All You Need for KV Cache in Diffusion LLMs , author =. 2025 , journal =

    work page 2025

  7. [7]

    2025 , journal =

    R-KV: Redundancy-aware KV Cache Compression for Reasoning Models , author =. 2025 , journal =

    work page 2025

  8. [8]

    2024 , journal =

    Lossless KV Cache Compression to 2\ author =. 2024 , journal =

    work page 2024

  9. [9]

    2026 , journal =

    Knowledge Packs: Zero-Token Knowledge Delivery via KV Cache Injection , author =. 2026 , journal =

    work page 2026

  10. [10]

    2025 , journal =

    The Pitfalls of KV Cache Compression , author =. 2025 , journal =

    work page 2025

  11. [11]

    2025 , journal =

    ChunkKV: Semantic-Preserving KV Cache Compression for Efficient Long-Context LLM Inference , author =. 2025 , journal =

    work page 2025

  12. [12]

    2024 , journal =

    A Survey on Large Language Model Acceleration based on KV Cache Management , author =. 2024 , journal =

    work page 2024

  13. [13]

    2024 , journal =

    KVSharer: Efficient Inference via Layer-Wise Dissimilar KV Cache Sharing , author =. 2024 , journal =

    work page 2024

  14. [14]

    2025 , journal =

    Towards Efficient Large Language Model Serving: A Survey on System-Aware KV Cache Optimization , author =. 2025 , journal =

    work page 2025

  15. [15]

    2025 , journal =

    TailorKV: A Hybrid Framework for Long-Context Inference via Tailored KV Cache Optimization , author =. 2025 , journal =

    work page 2025

  16. [16]

    2026 , journal =

    Compressing Kv Cache for Long-Context LLM Inference with Inter-Layer Attention Similarity , author =. 2026 , journal =

    work page 2026

  17. [17]

    2025 , journal =

    SpindleKV: A Novel KV Cache Reduction Method Balancing Both Shallow and Deep Layers , author =. 2025 , journal =

    work page 2025

  18. [18]

    2024 , journal =

    Model Tells You Where to Merge: Adaptive KV Cache Merging for LLMs on Long-Context Tasks , author =. 2024 , journal =

    work page 2024

  19. [19]

    2025 , journal =

    Hcattention: Extreme Kv Cache Compression Via Heterogeneous Attention Computing for Llms , author =. 2025 , journal =

    work page 2025

  20. [20]

    2025 , journal =

    EvolKV: Evolutionary KV Cache Compression for LLM Inference , author =. 2025 , journal =

    work page 2025

  21. [21]

    2025 , journal =

    Key, Value, Compress: A Systematic Exploration of KV Cache Compression Techniques , author =. 2025 , journal =

    work page 2025

  22. [22]

    2026 , journal =

    The kv4 Trade-off Is Workload-Dependent: A Depth-and Workload-Resolved Study of 4-bit KV-Cache Quantization on a 4 GB Turing GPU , author =. 2026 , journal =

    work page 2026

  23. [23]

    2025 , journal =

    TableKV: KV Cache Compression for In-Context Table Processing , author =. 2025 , journal =

    work page 2025

  24. [24]

    2026 , journal =

    Q-KVComm: Efficient Multi-Agent Communication via Adaptive KV Cache Compression , author =. 2026 , journal =

    work page 2026

  25. [25]

    2026 , journal =

    HCAttention: Extreme KV cache compression via heterogeneous attention computing for LLMs , author =. 2026 , journal =

    work page 2026

  26. [26]

    2025 , journal =

    Enhancing Large Multimodal Models with Adaptive Sparsity and KV Cache Compression , author =. 2025 , journal =

    work page 2025

  27. [27]

    2024 , journal =

    Eigen Attention: Attention in Low-Rank Space for KV Cache Compression , author =. 2024 , journal =

    work page 2024

  28. [28]

    , author =

    Structural optimization principles for edge AI in motorsport telemetry. , author =. 2026 , journal =

    work page 2026

  29. [29]

    , author =

    Building large-scale English-Romanian literary translation resources with open models. , author =. 2026 , journal =

    work page 2026

  30. [30]

    , author =

    Limit-Cycle Proliferation Under Parametric Delayed Feedback in a Conductance-Based Neuron: Bifurcation Landscape, Orbit Catalog, and Capacity Analysis. , author =. 2026 , journal =

    work page 2026

  31. [31]

    , author =

    Towards sustainable AI knowledge-base assistants in computer science education: on-premise deployment and optimization with open educational resources. , author =. 2026 , journal =

    work page 2026

  32. [32]

    , author =

    Multi-Hardware Benchmarking of Open-Source Large Language Models with Retrieval-Augmented Generation for Mitsubishi FX-Series PLC Instruction List Code Generation. , author =. 2026 , journal =

    work page 2026

  33. [33]

    , author =

    Infrared Imaging for Autonomous Power Inspection: A Review from Detector to System Integration. , author =. 2026 , journal =

    work page 2026

  34. [34]

    , author =

    A survey on LLM-as-a-judge. , author =. 2026 , journal =

    work page 2026

  35. [35]

    , author =

    The Rise of Small Language Models in Healthcare: A Comprehensive Survey. , author =. 2026 , journal =

    work page 2026

  36. [36]

    , author =

    Development of a Rapid Deep-Learning-Assisted Multiphase Multicomponent Numerical Simulation Protocol. , author =. 2026 , journal =

    work page 2026

  37. [37]

    , author =

    F-Transformer: a federated transformer for efficient and privacy-preserving sequence generation. , author =. 2026 , journal =

    work page 2026

  38. [38]

    2021 , journal =

    The Modern Mathematics of Deep Learning , author =. 2021 , journal =

    work page 2021

  39. [39]

    2023 , journal =

    Learn to Accumulate Evidence from All Training Samples: Theory and Practice , author =. 2023 , journal =

    work page 2023

  40. [40]

    2023 , journal =

    Deep Learning and Computational Physics (Lecture Notes) , author =. 2023 , journal =

    work page 2023

  41. [41]

    2026 , journal =

    Monodense Deep Neural Model for Determining Item Price Elasticity , author =. 2026 , journal =

    work page 2026

  42. [42]

    2017 , journal =

    A multitask deep learning model for real-time deployment in embedded systems , author =. 2017 , journal =

    work page 2017

  43. [43]

    2026 , journal =

    Training Transformers for KV Cache Compressibility , author =. 2026 , journal =

    work page 2026

  44. [44]

    2019 , journal =

    Deep learning observables in computational fluid dynamics , author =. 2019 , journal =

    work page 2019

  45. [45]

    2024 , journal =

    Mooncake: A KVCache-centric Disaggregated Architecture for LLM Serving , author =. 2024 , journal =

    work page 2024

  46. [46]

    2025 , journal =

    Efficient Many-Shot In-Context Learning with Dynamic Block-Sparse Attention , author =. 2025 , journal =

    work page 2025

  47. [47]

    2025 , journal =

    SwiftKV: Fast Prefill-Optimized Inference with Knowledge-Preserving Model Transformation , author =. 2025 , journal =

    work page 2025

  48. [48]

    2026 , journal =

    Sub-Token Routing for KV Cache Compression , author =. 2026 , journal =

    work page 2026

  49. [49]

    2024 , journal =

    Hymba: A Hybrid-head Architecture for Small Language Models , author =. 2024 , journal =

    work page 2024

  50. [50]

    2025 , journal =

    KV-RAPTOR: Scalable Tree-Structured Retrieval with KV Cache Compression for Question-Answering Systems , author =. 2025 , journal =

    work page 2025

  51. [51]

    1970 , journal =

    Deep Learning-Based Point Cloud Compression: An In-Depth Survey and Benchmark\_supp1-3594355.pdf , author =. 1970 , journal =

    work page 1970

  52. [52]

    2026 , journal =

    KV Pareto: Systems-Level Optimization of KV Cache and Model Compression for Long Context Inference , author =. 2026 , journal =

    work page 2026

  53. [53]

    2025 , journal =

    FAEDKV: Infinite-Window Fourier Transform for Unbiased KV Cache Compression , author =. 2025 , journal =

    work page 2025

  54. [54]

    2026 , journal =

    FastKV: Decoupling of Context Reduction and KV Cache Compression for Prefill-Decoding Acceleration , author =. 2026 , journal =

    work page 2026

  55. [55]

    , author =

    Large-Scale Model-Enhanced Vision-Language Navigation: Recent Advances, Practical Applications, and Future Challenges. , author =. 2026 , journal =

    work page 2026

  56. [56]

    , author =

    SS-AdaMoE: Spatio-Spectral Adaptive Mixture of Experts with Global Structural Priors for Graph Node Classification. , author =. 2026 , journal =

    work page 2026

  57. [57]

    , author =

    AI Agents as Universal Task Solvers. , author =. 2026 , journal =

    work page 2026

  58. [58]

    , author =

    Green prompt engineering for sustainable generative AI. , author =. 2026 , journal =

    work page 2026

  59. [59]

    , author =

    Anatomic Interactive Atlas of the Loggerhead Sea Turtle (Caretta caretta) Coelomic Cavity. , author =. 2026 , journal =

    work page 2026

  60. [60]

    , author =

    On Demand Secure Scalable Video Streaming for Both Human and Machine Applications. , author =. 2026 , journal =

    work page 2026

  61. [61]

    , author =

    Compute-in-memory implementation of state space models for event sequence processing. , author =. 2026 , journal =

    work page 2026

  62. [62]

    , author =

    Kilohertz volumetric imaging of in vivo dynamics using squeezed light field microscopy. , author =. 2025 , journal =

    work page 2025

  63. [63]

    Ai agents as universal task solvers

    Achille A and Soatto S. Ai agents as universal task solvers. PMC, 2026. doi:10.3390/e28030332

    work page doi:10.3390/e28030332 2026

  64. [64]

    Anatomic interactive atlas of the loggerhead sea turtle (caretta caretta) coelomic cavity

    Arencibia A, Melian A, and Oros J. Anatomic interactive atlas of the loggerhead sea turtle (caretta caretta) coelomic cavity. PMC, 2026 a . doi:10.3390/ani16050754

    work page doi:10.3390/ani16050754 2026

  65. [65]

    On demand secure scalable video streaming for both human and machine applications

    Zain A, Fan Y, and Zhou J. On demand secure scalable video streaming for both human and machine applications. PMC, 2026 b . doi:10.3390/s26041285

    work page doi:10.3390/s26041285 2026

  66. [66]

    Rest-kv: Robust kv cache eviction with layer-wise output reconstruction and spatial-temporal smoothing

    Yongqi An, Chang Lu, Kuan Zhu, Tao Yu, Chaoyang Zhao, Hong Wu, Ming Tang, and Jinqiao Wang. Rest-kv: Robust kv cache eviction with layer-wise output reconstruction and spatial-temporal smoothing. arXiv preprint, 2026

    work page 2026

  67. [67]

    The modern mathematics of deep learning

    Julius Berner, Philipp Grohs, Gitta Kutyniok, and Philipp Petersen. The modern mathematics of deep learning. arXiv preprint, 2021. doi:10.1017/9781009025096.002

    work page doi:10.1017/9781009025096.002 2021

  68. [68]

    Make each token count: Towards improving long-context performance with kv cache eviction

    Ngoc Bui, Hieu Trung Nguyen, Arman Cohan, and Rex Ying. Make each token count: Towards improving long-context performance with kv cache eviction. arXiv preprint, 2026

    work page 2026

  69. [69]

    R-kv: Redundancy-aware kv cache compression for reasoning models

    Zefan Cai, Wen Xiao, Hanshi Sun, Cheng Luo, Yikai Zhang, Ke Wan, Yucheng Li, Yeyang Zhou, Li-Wen Chang, Jiuxiang Gu, Zhen Dong, Anima Anandkumar, Abedelkadir Asi, and Junjie Hu. R-kv: Redundancy-aware kv cache compression for reasoning models. arXiv preprint, 2025

    work page 2025

  70. [70]

    The pitfalls of kv cache compression

    Alex Chen, Renato Geh, Aditya Grover, Guy Van den Broeck, and Daniel Israel. The pitfalls of kv cache compression. arXiv preprint, 2025

    work page 2025

  71. [71]

    Tablekv: Kv cache compression for in-context table processing

    Giulio Corallo, Elia Faure-Rolland, Miriam Lamari, and Paolo Papotti. Tablekv: Kv cache compression for in-context table processing. Journal, 2025. doi:10.18653/v1/2025.trl-1.13

    work page doi:10.18653/v1/2025.trl-1.13 2025

  72. [72]

    da Silva, Savio S

    Joao Gabriel J. da Silva, Savio S. T. de Oliveira, Lucas Alexandria Alves, Nicolas Eiris, and Arlindo Rodrigues Galvao Filho. Kv-raptor: Scalable tree-structured retrieval with kv cache compression for question-answering systems. Journal, 2025. doi:10.5753/sbbd.2025.247245

    work page doi:10.5753/sbbd.2025.247245 2025

  73. [73]

    Hymba: A Hybrid-head Architecture for Small Language Models

    Xin Luna Dong, Yonggan Fu, Shizhe Diao, Wonmin Byeon, Zhaoliang Chen, Ameya Sunil Mahabaleshwarkar, Shih-Yang Liu, Matthijs Van Keirsbilck, Min-Hung Chen, Yoshi Suhara, Yingyan Lin, Jan Kautz, and Pavlo Molchanov. Hymba: A hybrid-head architecture for small language models. Journal, 2024. doi:10.48550/arxiv.2411.13676

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2411.13676 2024

  74. [74]

    Deep learning-based point cloud compression: An in-depth survey and benchmark\_supp1-3594355.pdf

    Wei Gao. Deep learning-based point cloud compression: An in-depth survey and benchmark\_supp1-3594355.pdf. Journal, 1970. doi:10.1109/tpami.2025.3594355/mm1

    work page doi:10.1109/tpami.2025.3594355/mm1 1970

  75. [75]

    Monodense deep neural model for determining item price elasticity

    Lakshya Garg, Sai Yaswanth, Deep Narayan Mishra, Karthik Kumaran, Anupriya Sharma, and Mayank Uniyal. Monodense deep neural model for determining item price elasticity. arXiv preprint, 2026. doi:10.1109/AAIML67890.2026.11498150

    work page doi:10.1109/aaiml67890.2026.11498150 2026

  76. [76]

    Training transformers for kv cache compressibility

    Yoav Gelberg, Yam Eitan, Michael Bronstein, Yarin Gal, and Haggai Maron. Training transformers for kv cache compressibility. arXiv preprint, 2026

    work page 2026

  77. [77]

    Kv pareto: Systems-level optimization of kv cache and model compression for long context inference

    Sai Gokhale, Devleena Das, Rajeev Patwari, Ashish Sirasao, and Elliott Delaye. Kv pareto: Systems-level optimization of kv cache and model compression for long context inference. Journal, 2026. doi:10.18653/v1/2026.eacl-industry.9

    work page doi:10.18653/v1/2026.eacl-industry.9 2026

  78. [78]

    Minicache: Kv cache compression in depth dimension for large language models

    Gholamreza Haffari, Yefei He, Akide Liu, Jing Liu, Zizheng Pan, and Bohan Zhuang. Minicache: Kv cache compression in depth dimension for large language models. Journal, 2024. doi:10.52202/079017-4443

    work page doi:10.52202/079017-4443 2024

  79. [79]

    The risk of kv cache compression

    Lukas Haverbeck, Carmen Amo Alonso, Andres Felipe Posada-Moreno, Sebastian Trimpe, and Marco Pavone. The risk of kv cache compression. arXiv preprint, 2026

    work page 2026

  80. [80]

    A survey on llm-as-a-judge

    Gu J, Jiang X, Shi Z, Tan H, Zhai X, Xu C, Li W, Shen Y, Ma S, Liu H, Wang S, Zhang K, Lin Z, Zhang B, Ni L, Gao W, Wang Y, and Guo J. A survey on llm-as-a-judge. PMC, 2026. doi:10.1016/j.xinn.2025.101253

    work page doi:10.1016/j.xinn.2025.101253 2026

Showing first 80 references.