Pith. sign in

REVIEW 3 major objections 5 minor 5 cited by

MARM: Unlocking the Future of Recommendation Systems through Memory Augmentation and Scalable Complexity

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read By caching intermediate masked-self-attention outputs, MARM makes multi-layer user-interest modeling affordable at linear per-candidate cost and reports a cache scaling law.

desk verdict A genuinely useful industrial caching trick for user-sequence modeling, wrapped in overclaimed 'scaling-law' language and missing the one comparison that would validate the core equivalence. read the letter →

arxiv 2411.09425 v3 pith:ORFT6NAY submitted 2024-11-14 cs.IR

Xiao Lv , Jiangxia Cao , Shijie Guan , Xiaoyou Zhou , Zhiguang Qi , Yaqiang Zang , Ming Li , Ben Wang
show 2 more authors
Kun Gai Guorui Zhou
This is my paper · ORCID
classification cs.IR
keywords recommendationsystemscachescaling-lawuserinterestmodelingtargetattentionmaskedself-attentionmodelFLOPsstreamingtrainingmemoryaugmentation
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

Industrial recommendation is compute-bound: models serve billions of daily samples within milliseconds, so adding parameters or data is cheap while adding operations is expensive. This paper proposes MARM (Memory Augmented Recommendation Model), which attacks that bottleneck by caching the intermediate outputs of masked self-attention layers over a user's item history. At serving time, a candidate item only pays for a target-attention pass — attention from the candidate query to the stored vectors — so per-candidate cost drops from $O(n^2 d)$ to $O(n d)$ and a single-layer interest module can become multi-layer. The paper reports a cache scaling law: quality improves as cache size $C = L \cdot n \cdot d$ grows, with production A/B tests showing 0.43% GAUC offline and 2.079% more playtime per user online.

What carries the argument

The load-bearing object is the external cache memory: a key-value store whose keys identify a (user, historical item, layer) triple and whose values are that layer's masked self-attention output vectors. In the MARM forward pass, the first layer is ordinary target attention from the candidate item's embedding to the item sequence; each later layer is another target attention from the previous layer's query output to the cached vectors of the corresponding layer. This replaces the quadratic full-sequence masked self-attention with a chain of linear target attentions, and it is what makes cache size $C = L \cdot n \cdot d$ the resource that controls both storage and performance. Because cached values are written once and frozen, the backpropagation graph connects only the query path, a deliberate approximation to the true multi-layer model.

What would settle it

Train the exact uncached multi-layer masked self-attention model with the same depth, sequence length, features, and streaming data, and compare its GAUC to MARM; if the exact model beats MARM by more than the reported margins, the frozen-cache approximation is the reason. A cheaper probe is to periodically recompute cached entries with current parameters and check whether GAUC rises.

Watch

Extended reading notes

Core claim

The central claim is that caching lets a recommendation model deepen its user-interest module without paying the full cost of multi-layer attention. MARM stores, for every user, historical item, and attention layer, the masked self-attention output vector of that item; a candidate item's query then attends over these stored vectors layer by layer through target attention. Because the expensive self-attention computation is done once and reused across all candidate items, the per-candidate complexity falls from $O(n^2 d)$ to $O(n d)$, where $n$ is the sequence length and $d$ is the representation dimension. The paper further claims a cache scaling law: model quality rises with cache size $C = L \cdot n \cdot d$, and configurations with the same cache size behave comparably once $C$ is large enough. In the production short-video setting, the deployed MARM (depth $L=4$, sequence length $n=6000$, $d=128$, 60TB cache) reports 0.43% GAUC improvement offline, 2.079% playtime per user online, and one-eighth the combined compute and storage overhead of an uncached multi-layer self-attention module.

Load-bearing premise

The load-bearing premise is that intermediate results written to the cache once and then frozen stay accurate enough as the model's other parameters keep updating, so that the cached pipeline still behaves like a truly multi-layer attention model.

Editorial extensions

If this is right

  • A ranking model can grow user-history depth to several layers (e.g., $L=4$) and sequence length into the thousands (e.g., $n=6000$) while keeping per-candidate FLOPs linear, measured at roughly one-eighth the cost of an uncached multi-layer self-attention module.
  • Cache size $C = L \cdot n \cdot d$ behaves as a scaling resource: within the tested range, increasing $C$ improves GAUC, and configurations with the same $C$ (trading depth against sequence length) reach similar performance once $C$ is large.
  • The cache can be shared across recommendation stages — retrieval, cascading, and ranking — so the same stored interest representations serve multiple models rather than being recomputed.
  • MARM can be stacked on top of existing interest modules; the largest ensemble experiment shows 0.22% AUC and 0.43% GAUC gains over a baseline that already includes DIN, SIM, TWIN, and TWIN V2.
  • Online A/B tests show the deployed version adds roughly 15 ms per request and delivered +2.079% average playtime per user over the multi-stage system.

Reading between the lines

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

  • Editorial inference: the equal-performance-at-equal-cache-size pattern points to a compute-storage equivalence in attention-based sequence modeling, but the paper only tests a limited grid, so the equivalence should not be extrapolated far beyond the measured range.
  • Editorial inference: because cache entries are frozen in a one-epoch streaming regime, the method is most likely to pay off where user-item exposures are not replayed; recommendation systems with re-exposure or multi-epoch training could see the frozen cache become stale.
  • Editorial inference: the cached vectors are per-user, per-item interaction summaries, so the same store could unify retrieval, cascading, and ranking; the paper reports using the cache in all three stages but does not ablate the contribution of each.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes MARM, a memory-augmented sequence modeling module for industrial recommendation ranking. The core idea is to cache intermediate outputs of a multi-layer masked self-attention stack in an external key-value store, then approximate the stack at inference/training time by a sequence of target-attention layers that read the cached vectors, reducing per-candidate complexity from O(n^2·d) to O(n·d). The authors report a 'cache scaling law' in which performance improves with cache size C = L·n·d, offline GAUC gains of 0.43% over a strong ensemble baseline, and online A/B gains of 2.079% in play-time per user, with deployment at Kuaishou across retrieval, cascading, and ranking stages. The paper also describes a 60 TB cache storage center and resource-cost estimates.

Significance. If the central claim holds, the paper makes a practically significant contribution: it introduces a storage-for-compute trade-off in streaming RecSys, extends single-layer target-attention user modeling to multiple layers at controlled FLOPs, and demonstrates gains in a deployed system with online A/B validation. The scale of the deployment and the multi-stage online results are genuine strengths. However, the conceptual novelty over LLM KV-cache ideas is incremental, and the paper's core scientific assertion—that cached frozen representations faithfully emulate a full multi-layer attention model—is not directly validated. The significance therefore depends on whether that equivalence can be established experimentally, which the current manuscript does not do.

major comments (3)
  1. [Section 4.0.2, Eqs. (5)-(6), Figure 2] The central equivalence claim is not tested. Section 4.0.2 states that cached key/value vectors are not nodes in the computation graph, are written once, and are frozen, while Q/K/V and FFN parameters continue to update. Equations (5)-(6) replace a multi-layer masked self-attention stack with target-attention over these cached vectors, and Figure 2 claims 'The results are same!' No experiment compares MARM against the exact uncached multi-layer attention model on a tractable sequence length, nor against a version in which the cache is refreshed with current parameters. Without such a comparison, the offline and online gains could be attributed to the added target-attention parameters, to the cached vectors acting as additional dense input features, or to the extra FLOPs actually incurred, rather than to faithful approximation of deep self-attention. This is a load-bearing gap for the paper's main claim.
  2. [Section 3.2.3, Figure 5(c)] The 'cache scaling law' confounds cache size with compute and parameters. The paper defines C = L·n·d and notes that 'the size of C is linearly positively correlated with the online latency' and with all training, inference, and storage resources. Figure 5(c) plots performance against sequence length for curves labeled by C, where equal C corresponds to equal FLOPs (e.g., 400*128 has variants 4*100*128, 2*200*128, 1*400*128). Thus the observed improvement could be an ordinary scaling curve in FLOPs, depth, or sequence length, not a specific effect of cache size. The paper needs an ablation that holds FLOPs and parameter count fixed while varying cache freshness, cache coverage, or cache staleness, or a comparison against an uncached model with the same total FLOPs, to support the claim that memory, rather than compute, is the scaling variable.
  3. [Section 3.3.3, Table 2] The comparison with HSTU* is not matched and therefore does not support the conclusion that MARM is preferable to a direct multi-layer self-attention approach at equivalent cost. Table 2 reports HSTU* with only item ID, author ID, tag, and user feedback features, a history length of 2000, depth 4, and 1.04B FLOPs, while MARM uses richer features, the existing TWIN module as its first layer, length 6000, depth 4, and 302.6M FLOPs. The two systems differ simultaneously in architecture, input features, sequence length, and training procedure, so the result that 'directly adding an HSTU-style module provided no significant improvements' is not attributable to MARM's caching mechanism. A matched comparison at a tractable scale is needed.
minor comments (5)
  1. [Section 3.3.1, MARM bullet] The text says 'maximum attention depth L of 41'; this appears to be a typo for L = 4, since the experiments use L = 1, 2, and 4.
  2. [Section 4.0.1] The heading 'Complexity and Resoruces' contains a typo; it should read 'Complexity and Resources'.
  3. [Section 2.1.2, Eq. (4)] Equation (4) uses the same notation on both sides of the equality: the hash keys [Iid1_i_UidX, ...] are passed to MARMCacheLookUp, and the returned cached values are also written as [Iid1_i_UidX, ...]. The notation should distinguish keys from values to avoid ambiguity.
  4. [Section 2.1.2] The phrase 'the the bottom target-attention input' contains a duplicated article and should be corrected.
  5. [Figure 5(c)] The label 'C=0*128' is unclear because C is defined as L·n·d; since n and d are positive, C=0 only if L=0. The figure caption should clarify what 'C=0*128' represents (e.g., the L=0 baseline).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the cache-scaling results are empirical and the central MSA-to-TA approximation, though unvalidated, is not an input-equivalent construction.

full rationale

MARM's claimed derivation is not circular. The central reduction (Eqs. 5-6, Sec 2.1.3) replaces masked self-attention with target-attention over cached representations; this is an architectural approximation with an explicit effectiveness argument in Sec 4.0.2, not an equation whose output is defined to equal its input. The cache size C = L*n*d (Sec 3.2) is an input hyperparameter, and Figure 5 reports measured GAUC/loss as C is swept; no fitted parameter is renamed as a prediction. Offline GAUC and online play-time gains are independently measured A/B results. Self-citations (DIN, SIM) are used as baselines and as a GSU/ESU scaffold, but MARM is benchmarked against them rather than derived from them, so they are not load-bearing. The genuine weaknesses—no direct comparison to an exact uncached multi-layer attention model and stale frozen cached K/Vs—are correctness/validity concerns, not circularity.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The method rests on a cache of frozen intermediate representations plus four tuned hyperparameters. No new physical entities are introduced. The main untested axiom is that stale cached values can substitute for freshly computed attention outputs without meaningful loss.

free parameters (4)
  • representation dimension d = 128
    Fixed after grid search in Section 3.2.1; used for all scaling and deployment results.
  • sequence length n = 6000 in deployment; sweeps up to 6400
    Deployed length set in Section 3.3.1; scaling study varies n to demonstrate cache size effects.
  • depth L = 4 in deployment; ablations 1 to 4
    Chosen from Table 1 and the scaling study; deeper setting improves GAUC.
  • GSU top-K = unspecified
    MARM with SIM selects top-K per layer; K is not disclosed but influences cache content and performance.
assumptions (3)
  • domain assumption In streaming RecSys, an item exposed to a user will not be re-exposed, so one epoch of training sustains a highly generalized model.
    Used in Section 4.0.2 to argue frozen caches do not drift out of distribution.
  • ad hoc to paper Cached final query outputs from an earlier model version can stand in for exact masked self-attention key/value outputs of the current model.
    Core approximation behind Figure 2 'The results are same!', contradicted by the frozen-cache update rule in Section 4.0.2.
  • standard math Attention complexity is O(n^2*d) for masked self-attention and O(n*d) for target-attention, and cache lookup is O(1).
    Standard complexity accounting stated in Sections 1 and 5; hash lookup assumed constant cost in Algorithm 1.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MARM: Unlocking the Future of Recommendation Systems through Memory Augmentation and Scalable Complexity." pith.science (2026). https://pith.science/paper/ORFT6NAY

@misc{pith2026241109425,
  author       = {Pith},
  title        = {Pith review of: MARM: Unlocking the Future of Recommendation Systems through Memory Augmentation and Scalable Complexity},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ORFT6NAY}},
  note         = {Machine review of arXiv:2411.09425}
}
read the original abstract

Scaling-law has guided the language model designing for past years, however, it is worth noting that the scaling laws of NLP cannot be directly applied to RecSys due to the following reasons: (1) The amount of training samples and model parameters is typically not the bottleneck for the model. Our recommendation system can generate over 50 billion user samples daily, and such a massive amount of training data can easily allow our model parameters to exceed 200 billion, surpassing many LLMs (about 100B). (2) To ensure the stability and robustness of the recommendation system, it is essential to control computational complexity FLOPs carefully. Considering the above differences with LLM, we can draw a conclusion that: for a RecSys model, compared to model parameters, the computational complexity FLOPs is a more expensive factor that requires careful control. In this paper, we propose our milestone work, MARM (Memory Augmented Recommendation Model), which explores a new cache scaling-laws successfully.

Figures

Figures reproduced from arXiv: 2411.09425 by the authors.

Figure 1
Figure 1. (a) A toy example of ranking model. (b) Perfor [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Motivation of MARM, using caching idea to reduce higher-FLOPs self-attention to lower-FLOPs target-attention. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. MARM realtime cache update&use workflow. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Using the MARM framework to handle long sequences with SIM GSU/ESU; Example to support Retrieval Model. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Model performance of MARM scaling up. of the cache size and the amount of storage resources are linearly proportional to C. This means that the size of C is linearly positively correlated with the online latency of the recommendation system’s MARM module, as well as wi…
Figure 6
Figure 6. Figure 6: GSU Top-k overlapped rate between different layers. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: The online ranking time cost of the deployed ver [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 5 Pith papers

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

  1. CAPTS: Channel-Aware, Preference-Aligned Trigger Selection for Multi-Channel Item-to-Item Retrieval

    cs.IR 2026-02 conditional novelty 6.0 of 10

    CAPTS selects retrieval triggers by predicted downstream engagement per I2I channel and routes them jointly, reporting offline Recall@K gains and a small online time-spent lift.

  2. Request-Only Optimization for Recommendation Systems

    cs.IR 2025-07 conditional novelty 6.0 of 10

    A request-level training data format eliminates duplicate user features, increasing storage efficiency and training throughput while enabling larger recommendation architectures.

  3. LONGER: Scaling Up Long Sequence Modeling in Industrial Recommenders

    cs.IR 2025-05 conditional novelty 6.0 of 10

    LONGER combines global tokens, token merging with inner transformers, and hybrid causal attention to model sequences of length 10,000 end-to-end, outperforming strong baselines offline and online at ByteDance.

  4. A Contextual-Aware Position Encoding for Sequential Recommendation

    cs.IR 2025-02 conditional novelty 5.0 of 10

    CAPE computes context-dependent positions from item dissimilarity and fuses them into attention, improving sequential recommenders on benchmark and live-traffic tests.

  5. Climber: Toward Efficient Scaling Laws for Large Recommendation Models

    cs.IR 2025-02 conditional novelty 4.0 of 10

    Climber reports that splitting user sequences by behavior type, adding adaptive temperature, and co-designed batching enable more efficient Transformer scaling in recommender systems.

Reference graph

Works this paper leans on

25 extracted references · 1 canonical work pages · cited by 5 Pith papers

  1. [1]

    Joshua Ainslie, James Lee-Thorp, Michiel de Jong, Yury Zemlyanskiy, Federico Lebrón, and Sumit Sanghai. 2023. Gqa: Training generalized multi-query trans- former models from multi-head checkpoints. arXiv preprint arXiv:2305.13245 (2023)

    arXiv 2023

  2. [2]

    Jinze Bai, Shuai Bai, Shusheng Yang, Shijie Wang, Sinan Tan, Peng Wang, Junyang Lin, Chang Zhou, and Jingren Zhou. 2023. Qwen-vl: A frontier large vision- language model with versatile abilities. arXiv preprint arXiv:2308.12966 (2023)

    arXiv 2023

  3. [3]

    Yue Cao, Xiaojiang Zhou, Jiaqi Feng, Peihao Huang, Yao Xiao, Dayao Chen, and Sheng Chen. 2022. Sampling is all you need on modeling long-term user behaviors for CTR prediction. In Proceedings of the 31st ACM International Conference on Information & Knowledge Management . 2974–2983

    2022

  4. [4]

    Jianxin Chang, Chenbin Zhang, Zhiyi Fu, Xiaoxue Zang, Lin Guan, Jing Lu, Yiqun Hui, Dewei Leng, Yanan Niu, Yang Song, et al. 2023. TWIN: TWo-stage interest network for lifelong user behavior modeling in CTR prediction at kuaishou. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 3785–3794

    2023

  5. [5]

    Qiwei Chen, Changhua Pei, Shanshan Lv, Chao Li, Junfeng Ge, and Wenwu Ou

  6. [6]

    Harry Dong, Xinyu Yang, Zhenyu Zhang, Zhangyang Wang, Yuejie Chi, and Beidi Chen. 2024. Get More with LESS: Synthesizing Recurrence with KV Cache Compression for Efficient LLM Inference. arXiv preprint arXiv:2402.09398 (2024)

    arXiv 2024

  7. [7]

    Daya Guo, Qihao Zhu, Dejian Yang, Zhenda Xie, Kai Dong, Wentao Zhang, Guanting Chen, Xiao Bi, Yu Wu, YK Li, et al. 2024. DeepSeek-Coder: When the Large Language Model Meets Programming–The Rise of Code Intelligence.arXiv preprint arXiv:2401.14196 (2024)

    arXiv 2024

  8. [8]

    Jordan Juravsky, Bradley Brown, Ryan Ehrlich, Daniel Y Fu, Christopher Ré, and Azalia Mirhoseini. 2024. Hydragen: High-Throughput LLM Inference with Shared Prefixes. arXiv preprint arXiv:2402.05099 (2024)

    arXiv 2024

Show all 25 references
  1. [9]

    Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. 2020. Scaling laws for neural language models. arXiv preprint arXiv:2001.08361 (2020)

    2020 arXiv

  2. [10]

    Qi Pi, Weijie Bian, Guorui Zhou, Xiaoqiang Zhu, and Kun Gai. 2019. Practice on long sequential user behavior modeling for click-through rate prediction. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining . 2671–2679

    2019

  3. [11]

    Qi Pi, Guorui Zhou, Yujing Zhang, Zhe Wang, Lejian Ren, Ying Fan, Xiaoqiang Zhu, and Kun Gai. 2020. Search-based user interest modeling with lifelong sequential behavior data for click-through rate prediction. In Proceedings of the 29th ACM International Conference on Informat...

    2020

  4. [12]

    Noam Shazeer. 2019. Fast transformer decoding: One write-head is all you need. arXiv preprint arXiv:1911.02150 (2019)

    2019 arXiv

  5. [13]

    Zihua Si, Lin Guan, ZhongXiang Sun, Xiaoxue Zang, Jing Lu, Yiqun Hui, Xingchao Cao, Zeyu Yang, Yichen Zheng, Dewei Leng, et al . 2024. TWIN V2: Scaling Ultra-Long User Behavior Sequence Modeling for Enhanced CTR Prediction at Kuaishou. arXiv preprint arXiv:2407.16357 (2024)

    2024 arXiv

  6. [14]

    A Vaswani. 2017. Attention is all you need. Advances in Neural Information Processing Systems (2017)

    2017

  7. [15]

    Guangxuan Xiao, Yuandong Tian, Beidi Chen, Song Han, and Mike Lewis

  8. [16]

    Zihao Ye, Lequn Chen, Ruihang Lai, Yilong Zhao, Size Zheng, Junru Shao, Bohan Hou, Hongyi Jin, Yifei Zuo, Liangsheng Yin, Tianqi Chen, and Luis Ceze. 2024. Accelerating Self-Attentions for LLM Serving with FlashInfer. https://flashinfer. ai/2024/02/02/introduce-flashinfer.html

    2024

  9. [17]

    Amir Zandieh, Insu Han, Vahab Mirrokni, and Amin Karbasi. 2024. SubGen: Token Generation in Sublinear Time and Memory.arXiv preprint arXiv:2402.06082 (2024)

    2024 arXiv

  10. [18]

    Jiaqi Zhai, Lucy Liao, Xing Liu, Yueming Wang, Rui Li, Xuan Cao, Leon Gao, Zhao- jie Gong, Fangda Gu, Michael He, et al. 2024. Actions speak louder than words: Trillion-parameter sequential transducers for generative recommendations.arXiv preprint arXiv:2402.17152 (2024)

    2024 arXiv

  11. [19]

    Buyun Zhang, Liang Luo, Yuxin Chen, Jade Nie, Xi Liu, Daifeng Guo, Yanli Zhao, Shen Li, Yuchen Hao, Yantao Yao, et al. 2024. Wukong: Towards a Scaling Law for Large-Scale Recommendation. arXiv preprint arXiv:2403.02545 (2024)

    2024 arXiv

  12. [20]

    Yuren Zhang, Enhong Chen, Binbin Jin, Hao Wang, Min Hou, Wei Huang, and Runlong Yu. 2022. Clustering based behavior sampling with long sequential data for CTR prediction. In Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information R...

    2022

  13. [21]

    Zhenyu Zhang, Ying Sheng, Tianyi Zhou, Tianlong Chen, Lianmin Zheng, Ruisi Cai, Zhao Song, Yuandong Tian, Christopher Ré, Clark Barrett, et al. 2024. H2o: Heavy-hitter oracle for efficient generative inference of large language models. Advances in Neural Information Processing...

    2024

  14. [22]

    Guorui Zhou, Na Mou, Ying Fan, Qi Pi, Weijie Bian, Chang Zhou, Xiaoqiang Zhu, and Kun Gai. 2019. Deep interest evolution network for click-through rate prediction. In Proceedings of the AAAI conference on artificial intelligence , Vol. 33. 5941–5948

    2019

  15. [23]

    Guorui Zhou, Xiaoqiang Zhu, Chenru Song, Ying Fan, Han Zhu, Xiao Ma, Yanghui Yan, Junqi Jin, Han Li, and Kun Gai. 2018. Deep interest network for click-through rate prediction. In Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining ...

    2018

  16. [2021]

    arXiv preprint arXiv:2108.04468 (2021)

    End-to-end user behavior retrieval in click-through rateprediction model. arXiv preprint arXiv:2108.04468 (2021)

    2021 arXiv

  17. [2023]

    arXiv preprint arXiv:2309.17453 (2023)

    Efficient streaming language models with attention sinks. arXiv preprint arXiv:2309.17453 (2023)

    2023 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.