Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

FAEDKV: Infinite-Window Fourier Transform for Unbiased KV Cache Compression

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read FAEDKV claims a training-free Fourier transform of the KV cache compresses long-context memory to 9% while preserving all-token information.

desk verdict Plausible frequency-domain KV compression recipe, but the 'unbiased' claim is contradicted by its own equations and the memory comparison is misleading. read the letter →

arxiv 2507.20030 v1 pith:2KIHN6SG submitted 2025-07-26 cs.CL

Runchao Li , Yao Fu , Mu Sheng , Xianxuan Long , Haotian Yu , Pan Li This is my paper
classification cs.CL
keywords KVcachecompressionFouriertransforminfinitewindowfrequencydomainlong-contexttraining-freeattentionsinksLLMinference
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper proposes FAEDKV, a KV cache compression method that keeps a running frequency-domain summary of the entire token history instead of evicting or repeatedly recompressing old tokens. The central claim is that an 'Infinite-Window Fourier Transform' can update this summary in O(M) per step with all tokens contributing equally, so early and recent context are both preserved. On LongBench with a 9% cache, FAEDKV beats eviction baselines by up to 22% in accuracy, and it shows position-agnostic retrieval in needle-in-a-haystack tests. If the claims hold, long-context inference can keep most of the original attention quality at a fraction of the memory cost and without any fine-tuning.

What carries the argument

The Infinite-Window Fourier Transform (IWDFT) is a recursive update to a frequency-domain KV cache: each new token is rotated by a fixed twiddle factor and blended into the stored spectral state, as in Equation 9. It is what lets the cache grow without storing time-domain history; the paper also uses a frequency ablation to pick, per layer, which spectral chunks survive compression, and a sparse IDFT to reconstruct only those components. The key claim is that this preserves all tokens with equal weight, in contrast to eviction's recency bias and learned compression's repeated degradation of old tokens.

What would settle it

Construct a deterministic sequence of length N, compute its exact DFT, and run the IWDFT update over the same sequence. If the resulting state differs from the DFT coefficients—which the (N-1)/N-to-1 substitution predicts—the equal-weight property is false. A simpler check is to measure the state magnitude after N steps: an exact normalized DFT stays bounded, while the paper's approximation grows logarithmically.

Watch

Extended reading notes

Core claim

FAEDKV's central discovery is that the KV cache can be represented in the frequency domain and updated recursively as new tokens arrive, so that a fixed-size spectral state summarizes the unbounded history. The paper introduces the Infinite-Window Fourier Transform (IWDFT), given by S_{t+1}[k] = W_k ((N-1)/N S_t[k] + (1/N) x[t+1]), to fold each new token into the state while avoiding the overflow of a naive recursive DFT. A layer-wise frequency ablation on WikiText identifies which spectral chunks are critical per layer, and only those are retained at the chosen compression ratio. FAEDKV keeps the first S and last R tokens as time-domain anchors, reconstructs the compressed middle segment with a sparse IDFT, and combines them for attention. The paper reports average LongBench gains of 2.91 points over H2O and 2.12 over SnapKV, with the largest gains at the tightest cache budgets.

Load-bearing premise

The entire 'unbiased' claim rests on the IWDFT recursion being an exact Fourier transform of the full token history with equal per-token weights; the paper asserts this and then replaces (N-1)/N with 1, which changes the effective weights so early tokens dominate.

Editorial extensions

If this is right

  • At a 9% cache the method reports average LongBench accuracy gains of about 2.9 points over H2O and 2.1 over SnapKV, and up to 22% at the tightest budgets.
  • Needle-in-a-haystack accuracy stays roughly flat across needle positions, unlike compression methods that favor recent tokens, supporting the position-agnostic claim.
  • FAEDKV requires no fine-tuning and works within the model's existing context length, so it can be dropped into a deployed LLM with a one-time per-layer ablation.
  • Memory drops from O(N) to O(N·r) for the transformed segment, and the sparse IDFT keeps reconstruction cost proportional to the retained components.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the equal-weight property is what drives the gains, then restoring the exact (N-1)/N factor instead of approximating it to 1 could change the story at very long contexts; the paper does not report that ablation.
  • The frequency-ablation principle suggests a testable extension where the selected spectral bands adapt per input rather than being fixed per layer, which could combine the robustness of frequency retention with per-query flexibility.
  • Since FAEDKV explicitly does not extend the model's context limit, its real-world value is in compressing existing long-context deployments; pairing it with an eviction-based recent-token cache could yield further memory savings without changing the frequency-side claim.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes FAEDKV, a training-free KV-cache compression method that stores a pruned frequency-domain representation of the middle segment of the context, keeps small sink and recent token windows, and updates the frequency representation with a recursive "Infinite-Window Fourier Transform" (IWDFT). A one-time layer-wise frequency ablation on WikiText-103 selects which frequency chunks are retained, and the representation is pruned and reconstructed via sparse IDFT during decoding. Experiments on LongBench with Llama3-8B report gains over H2O and SnapKV at small cache sizes, and a Needle-in-a-Haystack study with Qwen2.5-7B reports more position-consistent retrieval than LoCoCo. The central stated contribution is that the method is unbiased because the IWDFT gives equal weight to all tokens.

Significance. If the method genuinely provided an equal-weight, fixed-memory spectral summary of the entire KV history, it would be a valuable training-free alternative to eviction-based and learned compression. The paper has several strengths: the ablation hyperparameters are tuned on WikiText-103/PG-19 while the headline LongBench and NIAH benchmarks are held out; the authors compare with official baseline code; and the Limitations section is candid about single-GPU resource constraints. However, the central theoretical claim is not supported by the paper's own equations: the IWDFT recursion in Eq. (9) is not a Fourier transform of the growing sequence, and the approximation of (N-1)/N to 1 changes the per-token weighting. The empirical numbers may still indicate a useful heuristic, but the paper's novelty and its 'unbiased' characterization rest on an incorrect mathematical premise.

major comments (3)
  1. [Section 4.1, Eq. (9)] The IWDFT recursion is not a DFT of the growing KV history. In the DFT definition of Eq. (4), the twiddle factor W_k = e^{-j2πk/N} depends on the current sequence length N, so the bin frequencies k/N change as N grows; a fixed W_k in Eq. (9) cannot reproduce the DFT coefficients of the full sequence. Moreover, the paper immediately approximates (N-1)/N to 1, which changes the per-token weight from 1/N (equal magnitude) to roughly 1 per token, so early tokens do not contribute with equal weight and the state magnitude grows linearly with t. This contradicts the claim in §4.1 that the normalization prevents overflow and the Abstract's 'equalized contribution of all tokens'. The authors must either use the exact normalization and provide a rigorous statement of what equal weighting means for the phase-rotated accumulation, or remove the unbiased/equal-weight claims and rename the update accordingly.
  2. [Table 1, Section 5.2] The caption of Table 1 describes the scores as 'perplexity', but LongBench tasks are scored with accuracy/F1 metrics and the reported values (e.g., FullKV 22.53 on NtrvQA) are not perplexity numbers. Since Table 1 is the main quantitative evidence for the paper's performance claims, the metric must be corrected and all prose statements should be checked against the correct metric.
  3. [Section 5.3 and Figure 3] The Needle-in-a-Haystack setup is described inconsistently: the text says the two methods were evaluated on 1024 cache size with r=0.05 of 24K, while the Figure 3 caption says LoCoCo uses cache size 2048 and FAEDKV uses a 5% compression rate; the text says contexts are 8K-30K tokens, while the caption says the x-axis runs from 8K to 300K tokens. This experiment is used to support the position-agnostic retrieval claim, so the actual configuration must be stated precisely and consistently.
minor comments (4)
  1. [Eq. (13)] In the assembly of V_t, the second block should be the reconstructed value cache eVt, not eKt as written.
  2. [Section 5.1 vs 5.3] The model is named 'Qwen2-7B-Instruct' in Section 5.1 and 'Qwen2.5-7B-Instruct' in Section 5.3; please use one consistent name.
  3. [Figure 6] The heatmap's x-axis is labeled 1-20, while the text states C=22 chunks; please align the figure labels with the stated hyperparameter.
  4. [Throughout] There are several typos and style inconsistencies, including 'mathmatical', 'convulution', and the typography of (N-1)/N; a careful proofreading pass is needed.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: FAEDKV's benchmark claims are evaluated on held-out LongBench and NIAH, and the 'unbiased' property is a construction-level property of Eq. 9 rather than a fitted prediction.

full rationale

The paper's main empirical claims are not circular by construction. The per-layer frequency-chunk masks are obtained from a WikiText-103 perplexity ablation, and the hyperparameters (C=22, S=10, R=50) are tuned on PG-19, but the headline LongBench and Needle-in-a-Haystack results are held-out evaluations, so the reported gains are not forced by the fitted masks or tuned constants. The 'unbiased information retention' claim follows directly from the design of Eq. 9, where the (N-1)/N and 1/N coefficients make each past token contribute with equal weight; this is a definitional property of the algorithm, not a benchmark prediction derived from it. The paper does not rely on a self-citation chain: the references are all external prior work, and no load-bearing claim is justified only by the present authors' earlier publications. The possible mathematical objection that Eq. 9 is not literally the DFT of a growing sequence concerns correctness or naming, not circularity, because the benchmark results are empirically measured rather than deduced from the Fourier interpretation. No circular step can be exhibited from the paper's own equations or citations.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a calibration transfer (WikiText-103 perplexity importances applied to other datasets and lengths), an unproven frequency-representation property of the recursive update, and an unexamined positional-semantics premise, plus hand-chosen hyperparameters (C=22, S=10, R=50) and retention ratios. The empirical gains further depend on a comparison whose memory accounting omits complex-valued storage and sink/recent tokens. No free parameters are derived from first principles; the retained chunk masks B*_l are fitted and unpublished.

free parameters (4)
  • C = 22 (number of frequency chunks) = 22
    Chosen via PG-19 perplexity ablation (Section 5.5); controls pruning granularity and the number of importance scores.
  • S = 10 sink tokens and R = 50 recent tokens = 10, 50
    Hand-picked in Section 5.1 to keep attention-sink and recent tokens in time domain; they escape the frequency compression but add memory overhead that the reported ratios ignore.
  • Retention ratio r = 0.094, 0.125, 0.25 (LongBench); 0.05 (NIAH)
    Experimental design choices; the smallest budget (r=0.094 of N=8192, about 768 bins) exceeds the baselines' 512-entry budget.
  • Per-layer retained chunk sets B*_l = not published
    Fitted to WikiText-103 perplexity sensitivities (Eq 6 and Eq 7) in the frequency ablation study (Section 3.2); these masks determine what is pruned and are never released.
assumptions (4)
  • domain assumption The token-axis KV cache can be treated as a time-domain signal whose DFT coefficients can be pruned without destroying the information attention needs.
    Sections 3.1.2 and 3.2: the entire compression scheme rests on the idea that a lossy spectral summary of K and V still supports downstream attention.
  • domain assumption Perplexity-based chunk importance measured on WikiText-103 at prefill length transfers to other datasets and context lengths.
    Section 3.2 selects B*_l on 100 WikiText-103 samples; Section 5.2 applies it to LongBench (average 11K tokens) and Section 5.3 to 8K-30K contexts without re-measuring.
  • ad hoc to paper The recursion in Eq 9 with fixed W_k is a valid Fourier representation of the full KV history with equal token weights.
    Section 4.1: a fixed-rotation recursion is not the DFT of a growing-length sequence, and the (N-1)/N approximately 1 approximation changes weights to 1/(arrival index). This is the central unproven premise.
  • domain assumption The model's positional embeddings (RoPE) tolerate the fixed position grid of the reconstructed middle block as the true context grows.
    Section 4.2.2, Eq 13: the assembled cache always places the reconstructed block in the same position range; the paper never analyzes how this interacts with RoPE.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FAEDKV: Infinite-Window Fourier Transform for Unbiased KV Cache Compression." pith.science (2026). https://pith.science/paper/2KIHN6SG

@misc{pith2026250720030,
  author       = {Pith},
  title        = {Pith review of: FAEDKV: Infinite-Window Fourier Transform for Unbiased KV Cache Compression},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2KIHN6SG}},
  note         = {Machine review of arXiv:2507.20030}
}
read the original abstract

The efficacy of Large Language Models (LLMs) in long-context tasks is often hampered by the substantial memory footprint and computational demands of the Key-Value (KV) cache. Current compression strategies, including token eviction and learned projections, frequently lead to biased representations -- either by overemphasizing recent/high-attention tokens or by repeatedly degrading information from earlier context -- and may require costly model retraining. We present FAEDKV (Frequency-Adaptive Infinite-Window for KV cache), a novel, training-free KV cache compression framework that ensures unbiased information retention. FAEDKV operates by transforming the KV cache into the frequency domain using a proposed Infinite-Window Fourier Transform (IWDFT). This approach allows for the equalized contribution of all tokens to the compressed representation, effectively preserving both early and recent contextual information. A preliminary frequency ablation study identifies critical spectral components for layer-wise, targeted compression. Experiments on LongBench benchmark demonstrate FAEDKV's superiority over existing methods by up to 22\%. In addition, our method shows superior, position-agnostic retrieval accuracy on the Needle-In-A-Haystack task compared to compression based approaches.

Figures

Figures reproduced from arXiv: 2507.20030 by the authors.

Figure 1
Figure 1. Conceptual illustration of how different KV cache management strategies process past tokens, highlighting [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of the FAEDKV workflow. (a) Pre-filling: The middle segment of the initial context is converted [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Results of Fact Retrieval Across Context [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Results of Pre-filling and Decoding Latency [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 6
Figure 6. Figure 6: Overview of the Frequency Ablation Study. (a) The workflow illustrates the process: time-domain [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Blurry Window Attention

    cs.LG 2026-05 unverdicted novelty 6.0 of 10

    Blurry Window Attention stores a frequency window and reconstructs blurry KV history via Dirichlet kernel interpolation, achieving 8x better state efficiency than sliding window attention on the MQAR synthetic task.

Reference graph

Works this paper leans on

31 extracted references · 8 canonical work pages · cited by 1 Pith paper

  1. [1]

    Joshua Ainslie, James Lee-Thorp, Michiel de Jong, Yury Zemlyanskiy, Federico Lebron, and Sumit Sanghai. 2023. https://doi.org/10.18653/v1/2023.emnlp-main.298 GQA : Training generalized multi-query transformer models from multi-head checkpoints . In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 4895--4901, Si...

    doi:10.18653/v1/2023.emnlp-main.298 2023

  2. [2]

    Yushi Bai, Xin Lv, Jiajie Zhang, Yuze He, Ji Qi, Lei Hou, Jie Tang, Yuxiao Dong, and Juanzi Li. 2024 a . https://doi.org/10.18653/v1/2024.findings-emnlp.74 L ong A lign: A recipe for long context alignment of large language models . In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 1376--1395, Miami, Florida, USA. Association...

    doi:10.18653/v1/2024.findings-emnlp.74 2024

  3. [3]

    Yushi Bai, Xin Lv, Jiajie Zhang, Hongchang Lyu, Jiankai Tang, Zhidian Huang, Zhengxiao Du, Xiao Liu, Aohan Zeng, Lei Hou, Yuxiao Dong, Jie Tang, and Juanzi Li. 2024 b . https://arxiv.org/abs/2308.14508 Longbench: A bilingual, multitask benchmark for long context understanding . Preprint, arXiv:2308.14508

    arXiv 2024

  4. [4]

    Ruisi Cai, Yuandong Tian, Zhangyang Wang, and Beidi Chen. 2024 a . https://arxiv.org/abs/2406.05317 Lococo: Dropping in convolutions for long context compression . Preprint, arXiv:2406.05317

    arXiv 2024

  5. [5]

    Zefan Cai, Yichi Zhang, Bofei Gao, Yuliang Liu, Tianyu Liu, Keming Lu, Wayne Xiong, Yue Dong, Baobao Chang, Junjie Hu, and Wen Xiao. 2024 b . https://arxiv.org/abs/2406.02069 Pyramidkv: Dynamic kv cache compression based on pyramidal information funneling . Preprint, arXiv:2406.02069

    arXiv 2024

  6. [6]

    RangRang Ge, ShiYe Song, Zhaorui Liu, Wei Liu, Yuesheng Wang, Dongling Wang, Bofang Zhou, Zhicheng Dou, and Ji-Rong Wen. 2023. A survey on KV cache compression for large language models. arXiv preprint arXiv:2312.10546

    work page Pith review arXiv 2023

  7. [7]

    Suyu Ge, Yunan Zhang, Liyuan Liu, Minjia Zhang, Jiawei Han, and Jianfeng Gao. 2024. https://arxiv.org/abs/2310.01801 Model tells you what to discard: Adaptive kv cache compression for llms . Preprint, arXiv:2310.01801

    arXiv 2024

  8. [8]

    Chi Han, Qifan Wang, Hao Peng, Wenhan Xiong, Yu Chen, Heng Ji, and Sinong Wang. 2024. https://arxiv.org/abs/2308.16137 Lm-infinite: Zero-shot extreme length generalization for large language models . Preprint, arXiv:2308.16137

    arXiv 2024

Show all 31 references
  1. [9]

    Ziwei He, Meng Yang, Minwei Feng, Jingcheng Yin, Xinbing Wang, Jingwen Leng, and Zhouhan Lin. 2023. https://doi.org/10.18653/v1/2023.findings-acl.570 Fourier transformer: Fast long range modeling by removing sequence redundancy with fft operator . In Findings of the Associatio...

    2023 doi

  2. [10]

    Jushi Kai, Boyi Zeng, Yixuan Wang, Haoli Bai, Ziwei He, Bo Jiang, and Zhouhan Lin. 2025. https://arxiv.org/abs/2505.00570 Freqkv: Frequency domain key-value compression for efficient context window extension . Preprint, arXiv:2505.00570

    2025

  3. [11]

    Byungchan Lee, Seokmin Lee, Donghyun Kim, and Beomseok Heo. 2024. Scaling FFNs for better transformer. arXiv preprint arXiv:2403.15916

    2024 arXiv

  4. [12]

    Heejun Lee, Geon Park, Jaduk Suh, and Sung Ju Hwang. 2025. https://arxiv.org/abs/2502.08910 Infinitehip: Extending language model context up to 3 million tokens on a single gpu . Preprint, arXiv:2502.08910

    2025 arXiv

  5. [13]

    James Lee-Thorp, Joshua Ainslie, Ilya Eckstein, and Santiago Ontanon. 2021. FNet : Mixing tokens with fourier transforms. arXiv preprint arXiv:2105.03824

    2021 arXiv

  6. [14]

    Mo Li, Songyang Zhang, Taolin Zhang, Haodong Duan, Yunxin Liu, and Kai Chen. 2024 a . https://doi.org/10.48550/ARXIV.2407.11963 Needlebench: Can llms do retrieval and reasoning in information-dense context? Preprint, arXiv:2407.11963

    2024 doi

  7. [15]

    Shaoyi Li, Zhaowen Ni, Tiecheng Li, Hong Ma, and Zheng Wang. 2024 b . AFFormer : Resolution-agnostic and frequency-adaptive recurrent transformer for image super-resolution. arXiv preprint arXiv:2403.05088

    2024 arXiv

  8. [16]

    Yuhong Li, Yingbing Huang, Bowen Yang, Bharat Venkitesh, Acyr Locatelli, Hanchen Ye, Tianle Cai, Patrick Lewis, and Deming Chen. 2024 c . https://arxiv.org/abs/2404.14469 Snapkv: Llm knows what you are looking for before generation . Preprint, arXiv:2404.14469

    2024 arXiv

  9. [17]

    Liyue Liu, Shijie Li, Zhuo Chen, Tianyi Li, and Yu Wang. 2023 a . https://openreview.net/forum?id=aZIFC4NFfE Scissorhands: Exploiting the persistence of importance hypothesis for llm kv cache compression at test time . In Thirty-seventh Conference on Neural Information Process...

    2023

  10. [18]

    Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang

    Nelson F. Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. 2023 b . https://arxiv.org/abs/2307.03172 Lost in the middle: How language models use long contexts . Preprint, arXiv:2307.03172

    2023 arXiv

  11. [19]

    Zichang Liu, Jue Wang, Tri Zhao, Zirui Li, Yixin Bai, and Jeff Yu. 2024. Deja vu: Contextual sparsity for efficient LLM inference. arXiv preprint arXiv:2401.09486

    2024 arXiv

  12. [20]

    Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. 2016. https://arxiv.org/abs/1609.07843 Pointer sentinel mixture models . Preprint, arXiv:1609.07843

    2016 arXiv

  13. [21]

    Morris, Neel Nanda, Michael J

    Sharan Murty, John D. Morris, Neel Nanda, Michael J. Li, Jacob Andreas, Michael I. Hudson, and Divya Misra. 2024. Exactly solving acrostic puzzles with a language model. arXiv preprint arXiv:2403.04534

    2024 arXiv

  14. [22]

    Neel Nanda, Lawrence Chan, Tom Liberum, Jess Smith, and Jacob Steinhardt. 2023. https://openreview.net/forum?id=zNHz3V8xS5 Emergent linear representations in world models of self-supervised transformers . In International Conference on Learning Representations

    2023

  15. [23]

    Jack W Rae, Anna Potapenko, Siddhant M Jayakumar, Chloe Hillier, and Timothy P Lillicrap. 2019. https://arxiv.org/abs/1911.05507 Compressive transformers for long-range sequence modelling . arXiv preprint

    2019 arXiv

  16. [24]

    Jianlin Su, Yu Lu, Shengfeng Pan, Ahmed Murtadha, Bo Wen, and Yunfeng Liu. 2024. RoFormer : Enhanced transformer with rotary position embedding. arXiv preprint arXiv:2104.09864

    2024 arXiv

  17. [25]

    Gonzalez, and Raluca Ada Popa

    Sijun Tan, Xiuyu Li, Shishir Patil, Ziyang Wu, Tianjun Zhang, Kurt Keutzer, Joseph E. Gonzalez, and Raluca Ada Popa. 2024. https://arxiv.org/abs/2404.07979 Lloco: Learning long contexts offline . Preprint, arXiv:2404.07979

    2024 arXiv

  18. [26]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2023. https://arxiv.org/abs/1706.03762 Attention is all you need . Preprint, arXiv:1706.03762

    2023 arXiv

  19. [27]

    Peitian Zhang, Zheng Liu, Shitao Xiao, Ninglu Shao, Qiwei Ye, and Zhicheng Dou. 2024. https://arxiv.org/abs/2401.03462 Long context compression with activation beacon . Preprint, arXiv:2401.03462

    2024 arXiv

  20. [29]

    Zhenyu Zhang, Ying Sheng, Tianyi Zhou, Tianlong Chen, Lianmin Zheng, Ruisi Cai, Zhao Song, Yuandong Tian, Christopher Ré, Clark Barrett, Zhangyang Wang, and Beidi Chen. 2023 b . https://arxiv.org/abs/2306.14048 H _2 o: Heavy-hitter oracle for efficient generative inference of ...

    2023 arXiv

  21. [30]

    Kaixuan Zhou, Jianing Wang, Zhiyuan Yin, Yan Zhou, Xin Cao, Yang Gao, and Sheng Zhao. 2024. SFM-LLM : A efficient long-term time series forecasting framework based on spectrum frequency mix. arXiv preprint arXiv:2403.15912

    2024 arXiv

  22. [31]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  23. [32]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.