Pith. sign in

REVIEW 5 major objections 6 minor 34 references

Gryphon-v2: One Model in Place of a Cascade - Generate-and-Rank Recommender with Rollout Distillation

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A single generate-and-rank model, Gryphon-v2, can replace the full production recommender cascade—more than 15 candidate generators, pre-ranking, and final ranking—while improving active users by 1.41% at comparable serving latency.

desk verdict One model can replace a cascade in live traffic, but the distillation-specific claim leans on an unshown teacher edge. read the letter →

arxiv 2608.06213 v1 pith:6J5H3PKA submitted 2026-08-06 cs.IR

Anna Lipkina , Daria Tikhonovich , Viktor Yanush , Mariia Ulianova , Oleg Sorokin , Vladislav Dodonov , Ilya Murzin , Denis Burshtein
show 1 more author
Nikolay Savushkin
This is my paper · ORCID
classification cs.IR
keywords generativerecommendationSemanticIDsrolloutdistillationknowledgelearningtorankrecommendersystemsgenerate-and-rank
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 claims that a mature industrial recommender cascade—more than 15 candidate generators, a pre-ranker, and a final ranker—can be replaced by one model that both generates and ranks recommendations. The crucial training idea is to have a large, training-only teacher ranker score candidates from the model's own decoder rollouts as well as from logged impressions, and to distill those scores into a lightweight ranking module that reuses the shared user-history encoding. If the claim holds, the multi-stage serving stack collapses to a single model encoded once per request, and the online A/B test reported here shows the replacement increasing the number of active users by 1.41% at serving latency comparable to the cascade. The paper offers this as evidence that unified generate-and-rank recommenders are practically viable end-to-end alternatives to production cascades.

What carries the argument

The central mechanism is Rollout Distillation built on Gryphon's shared-encoder generate-and-rank architecture. The model encodes each user's history once; the decoder generates Semantic-ID candidates under a catalogue-trie-constrained beam search; and an item-level Ranking Module—a lightweight cross-attention head over the same encoded history—produces task-specific scores, combined with fixed weights into the final ordering. Rollout Distillation supplies the ranking supervision: a training-only Teacher Ranker scores candidates from the current decoder's beam search and from logged impressions, and the Ranking Module is trained with per-task mean-absolute-error losses against those scores while the decoder is trained with next-token prediction. Because the teacher is never part of the serving graph, fine-grained ranking preferences are transferred without adding a second model to the serving path.

What would settle it

Run the same online A/B experiment with a Gryphon-v2 variant whose Ranking Module is distilled from the production ranker's scores instead of the Teacher Ranker's; if user engagement does not decline, the claim that teacher superiority drives the reported gains is contradicted.

Watch

Extended reading notes

Core claim

Gryphon-v2 proposes that one served model can do the work of the entire production cascade. The model encodes a user's history once with a bidirectional Transformer; an autoregressive decoder then generates Semantic IDs—hierarchical discrete codes assigned to catalogue items—under a trie-constrained beam search, resolves them to items, and an item-level Ranking Module reusing the shared encoder states produces task-specific scores whose fixed combination determines the final order. The Ranking Module is not supervised by next-item prediction; instead, Rollout Distillation trains it to match the scores of a high-capacity, training-only Teacher Ranker on two complementary candidate sets: beam-search rollouts from the current decoder (the same decoding mechanism used at serving) and logged impressions from production traffic. In an online A/B experiment on a large-scale music recommendation surface, a single Gryphon-v2 model replaced a cascade of more than 15 candidate generators, pre-ranking, and final ranking, increasing active users by 1.41% and total listening time by 1.62% at serving latency comparable to the cascade. The paper's central claim is that this validates a generate-and-rank recommender with a teacher-distilled Ranking Module as a complete substitute for a multi-stage cascade.

Load-bearing premise

The whole method rests on the assumption that the Teacher Ranker is actually a better ranker than the production ranker; this superiority is asserted from earlier internal evaluations not reported as controlled results in the paper, so if the teacher is not superior the distilled Ranking Module inherits a worse preference signal and the reported gains cannot be attributed to Rollout Distillation.

Editorial extensions

If this is right

  • A single shared-encoder model can be the only learned model on the serving path, eliminating separate candidate-generation, pre-ranking, and final-ranking stages.
  • Distilling on decoder rollouts makes the Ranking Module faithful to the candidate distribution the model actually produces: rollout + impression training reaches TeacherRecall@10 of 0.5654, versus 0.2983 with impressions only.
  • Serving becomes cheaper in structure: the unified path ranks at most 1,200 resolved items rather than the cascade's ~10,000 generated and 3,000 pre-ranked candidates, at comparable end-to-end latency.
  • The teacher's expensive inference pass is removed from serving; throughput is about 4x that of the same generative backbone followed online by the Teacher Ranker.
  • The online gains are aggregate effects of the full replacement, not isolated contributions of distillation, the Ranking Module, or online updates—as the paper's own limitations section notes.

Reading between the lines

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

  • The decisive untested premise is the Teacher Ranker's superiority over the production ranker; a direct matched comparison between those two models on identical data would either support or break the paper's story.
  • If rollout distillation transfers to other domains, the candidate-source mix (rollouts vs logged impressions) would likely need re-tuning per application, since the ablation shows each source alone favours different metrics.
  • Because serving is updated on a ten-minute cadence while the teacher is refreshed daily, part of the reported gains may come from online learning rather than the distillation scheme; an A/B with a frozen checkpoint would separate these effects.
  • The paper's single-surface, short-window A/B on 8% of users leaves open whether the gains persist at full population scale or over longer horizons, which is exactly the limitation flagged in the conclusion.
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

5 major / 6 minor

Summary. The paper presents Gryphon-v2, a unified generate-and-rank recommender in which a Semantic-ID autoregressive decoder proposes candidates and an item-level Ranking Module, sharing the same user-history encoder, produces the final ordering. The Ranking Module is trained by distilling a training-only Teacher Ranker over two candidate sources: synchronous decoder rollouts and logged impressions, with next-token prediction retained on the generative path. The authors report offline diagnostics (R@1000, TeacherRecall@k, WPA) and a user-level online A/B experiment in which a single Gryphon-v2 model replaces a production cascade of more than 15 candidate generators, a pre-ranker, and a final ranker, increasing active users by 1.41% at allegedly comparable serving latency.

Significance. If the online result is reliable, this is a practically significant industrial demonstration that one learned model can replace a mature multi-stage recommender cascade. The architecture is clearly described, the supervision comparison against Gryphon is well matched, and the paper is unusually candid about limitations: it explicitly calls TeacherRecall@k a fidelity measure, notes that WPA is measured under the existing cascade's exposure policy, and acknowledges in Section 6 that the online experiment does not isolate the marginal contributions of components. These strengths make the paper worth serious consideration. However, several load-bearing pieces of evidence are missing or underreported: the teacher-superiority premise, quantitative latency parity, and uncertainty estimates for the offline metrics. The central claims are therefore defensible but not yet fully supported.

major comments (5)
  1. [Section 3.5, Table 1] The distillation mechanism rests on the claim that the Teacher Ranker is superior to the production ranker, but the only evidence in the paper is Table 1, which reports WPA 0.6199 vs. 0.6141 on logged impressions with no confidence intervals, significance test, or online engagement comparison. Section 3.5 states that the validating evaluations are prior internal results not included as controlled experiments. Since the teacher supplies the only ranking supervision (Section 3.6), this missing evidence is load-bearing for attributing the online gains specifically to Rollout Distillation.
  2. [Section 5.5, Table 4] The abstract and Section 5.5 claim 'serving latency comparable to the production cascade,' but Table 4 reports only the label 'Comparable' with no measured latency values, percentiles, or description of the measurement conditions. Given that latency parity is one of the two headline deployment claims, the absence of quantitative data makes the operational claim unverifiable as reported.
  3. [Section 5.1, Table 1] All offline comparisons are reported as point estimates without confidence intervals or significance tests. The R@1000 differences are acknowledged as within run-to-run standard deviation, but WPA differences such as 0.5892 vs. 0.5528 between Gryphon-v2 and Gryphon are discussed as meaningful without error bars. Since these numbers support the claim that distillation recovers 'the majority of the gap' to the reference rankers, uncertainty quantification is needed before the offline ranking conclusion can be accepted.
  4. [Section 5.3, Section 6] The online A/B experiment is a single treatment that bundles the new architecture, Rollout Distillation, the reduced candidate pool, and ten-minute online updates, so the +1.41% active-user gain cannot be attributed specifically to Rollout Distillation or to the unified architecture. The paper acknowledges this in Section 6, but the abstract and conclusion nevertheless make mechanism-level claims about the 'distilled Ranking Module' that the experiment cannot isolate. The authors should either provide a decomposition that separates the contributions or revise the causal language to match the aggregate nature of the evidence.
  5. [Section 5.1, Section 6] TeacherRecall@k is defined as agreement with the same Teacher Ranker that generates the distillation targets, so the large jump from 0.0392 to 0.5654 is partly by construction. The paper does label this as a fidelity metric in Section 6, but the Results narrative in Section 5.2 presents it as the primary evidence that distillation 'works.' The authors should present T-R@k only as a sanity check of optimization fidelity, or supply an independent ranking-quality metric evaluated on the candidate distribution that Gryphon-v2 itself serves.
minor comments (6)
  1. [Section 3.2] The sentence defining the teacher target contains a formatting artifact: 'producing a teacher target r^{T,t}_{u,i} ... where the T denotes the teacher' has misplaced spacing and reads awkwardly; please correct the typesetting.
  2. [Section 5.1] The definition of TeacherRecall@k uses intersection of Top-k sets without specifying tie handling; if ties are possible in the teacher's or student's scores, the metric is ambiguous and should be defined with a deterministic tie-break rule.
  3. [Section 3.4] The task combination weights w_t in the final score R(u,i) are called fixed, but their values are not reported anywhere, including in the appendix; please state the values or cite where they are defined.
  4. [Table 3] The caption correctly notes that bold does not imply statistical significance, but the surrounding text in Section 5.4 says 'MSE performs best on T-R@10 and WPA' without consistently qualifying that these are descriptive point estimates; please add the same caveat in the prose.
  5. [References] Several references are 2026 arXiv preprints and may not yet be publicly available; please verify that all cited works are accessible and appropriately dated, and add arXiv identifiers where missing.
  6. [Figure 1] Figure 1 is referenced and describes the framework, but the caption uses abbreviations such as NTP and SID without expansion; please spell out all acronyms in the caption for readability.

Circularity Check

1 steps flagged · score 2.0 of 10

No significant circularity: the online A/B result is independent, and the only self-referential element is the TeacherRecall@k fidelity metric, which the paper explicitly discloses as non-independent.

  1. self definitional [Section 3.6 (Rollout Distillation loss) and Section 5.1 (TeacherRecall@k definition); acknowledged in Section 6 Limitations]
    "We use the Teacher Ranker’s top-𝑘 selections as an offline reference for evaluating distillation fidelity. This metric does not independently measure recommendation quality; rather, it measures how closely the student reproduces the teacher on the student’s serving-time candidate distribution. ... TeacherRecall@𝑘 uses the same Teacher Ranker that provides the distillation targets and therefore measures distillation fidelity."

    The Ranking Module is trained by minimizing Ldistill = Lroll-distill + Limpr-distill, where each term is an MAE between the student score \widehat{r}^t_{u,i} and the Teacher Ranker score r^{T,t}_{u,i}. TeacherRecall@k is defined as the fraction of the Teacher Ranker’s top-k items recovered in the student’s top-k. The reported jump from 0.0392 to 0.5654 therefore measures agreement with the very teacher that supplied the training targets; it is a fit/fidelity check, not an independent prediction of quality. The paper says so itself in Section 6, so the circularity is disclosed and does not support the online engagement claim, which is an external A/B benchmark.

full rationale

The only exhibitable circular step is the TeacherRecall@k diagnostic: the distillation loss (Sec. 3.6) fits student scores to the Teacher Ranker, and the metric (Sec. 5.1) measures top-k overlap with that same Teacher Ranker. This is self-referential by construction, but the paper explicitly labels it as distillation fidelity and disclaims independent quality measurement in Section 6. The headline online result (+1.41% active users vs. the production cascade) is an external A/B benchmark independent of the teacher, so the central claim does not reduce to the teacher's scores. The teacher-superiority premise in Section 3.5 rests on unreported prior internal evaluations and a point-estimate WPA gap (0.6199 vs. 0.6141) without significance testing; this is an evidence gap and correctness risk, but not circularity, because the teacher's quality is assessed separately from the student. The Gryphon architecture is adopted via a same-author citation [25], but it is prior work used as a building block, not a uniqueness theorem or an unverified premise on which the paper's own derivation depends. Score 2 reflects the minor disclosed self-referential metric while the central claim is independent.

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

The central claim does not rely on new physical or mathematical entities; it rests on empirical assumptions about teacher quality, semantic ID fidelity, and distribution match between training rollouts and serving, all of which are plausible but unverified outside the paper.

free parameters (4)
  • Task combination weights w_t = Not reported
    Final ranking score is a fixed weighted sum of task-specific head outputs; the weights are stated as fixed but their origin or fitting procedure is not given.
  • Distillation loss coefficients = 1.0 for each source
    Both rollout and impression losses enter with coefficient one; these equal weights are chosen by hand rather than tuned.
  • Training rollout beam size = 32
    Used for all offline and online training; the paper notes that matching serving beam size 1024 was prohibitive, and beam size was fixed before the ablation sweep.
  • Candidate budget B_item = 1,200
    Resolved candidate pool cap used at inference; chosen to balance recall and latency.
assumptions (4)
  • domain assumption Teacher Ranker outperforms the production ranker
    Section 3.5: prior internal offline and online validation predated this experiment and is not included; if false, distillation targets are not better than production.
  • domain assumption Semantic IDs preserve item identity with small collision classes
    Section 3.3: residual K-means quantization yields Independent Code Rate 0.98; the ranking module relies on resolved item identities.
  • domain assumption On-policy rollouts at beam 32 are representative of serving-time beam 1024 candidates
    Section 5.4: matching serving beam is prohibitive; training uses beam 32, while serving uses 1,024, and the paper does not measure distribution shift.
  • domain assumption Logged impressions under the production cascade's exposure policy are valid ranking supervision
    Section 3.6 and Section 6: WPA and impression distillation use logged impressions, which are biased by the old cascade's exposure; the paper acknowledges this does not measure ordering on Gryphon-v2's own items.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Gryphon-v2: One Model in Place of a Cascade - Generate-and-Rank Recommender with Rollout Distillation." pith.science (2026). https://pith.science/paper/6J5H3PKA

@misc{pith2026260806213,
  author       = {Pith},
  title        = {Pith review of: Gryphon-v2: One Model in Place of a Cascade - Generate-and-Rank Recommender with Rollout Distillation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6J5H3PKA}},
  note         = {Machine review of arXiv:2608.06213}
}
read the original abstract

Industrial recommender systems are commonly deployed as multi-stage cascades with separate candidate generators, pre-rankers, and final rankers. Although effective, these cascades require repeated user-history processing, complex feature pipelines, and multiple serving stages. Semantic-ID-based generative retrieval offers a path toward simpler end-to-end systems, but next-item prediction alone does not capture the fine-grained preferences encoded by production ranking objectives. We present Gryphon-v2, a unified generate-and-rank architecture for end-to-end recommendation. The model encodes a user history once, generates Semantic-ID candidates with an autoregressive decoder, resolves them to catalogue items, and ranks them with an item-level Ranking Module that reuses the shared encoder states. To transfer fine-grained production ranking preferences without adding an expensive second model to the serving path, we distill a high-capacity, training-only Teacher Ranker into the Ranking Module. Gryphon-v2 is trained with Rollout Distillation: teacher scores are the only ranking supervision, and they are collected over two complementary candidate distributions. Rollouts from the current decoder expose the Ranking Module to candidates produced by the same generation mechanism used at serving time, while logged impressions cover items users were actually shown. In an online A/B experiment on a large-scale recommendation surface at Yandex Music, a single Gryphon-v2 model replaces a production cascade comprising more than 15 candidate generators, pre-ranking, and final ranking. The deployment increases the number of active users by 1.41% at serving latency comparable to the production cascade. These results support the practical viability of a generative retriever with a Ranking Module distilled from the Teacher Ranker as an end-to-end alternative to a production cascade.

Figures

Figures reproduced from arXiv: 2608.06213 by the authors.

Figure 1
Figure 1. Gryphon-v2 framework: a unified generate-and-rank architecture that replaces the production cascade with one [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 14 canonical work pages

  1. [1]

    Rishabh Agarwal, Nino Vieillard, Yongchao Zhou, Piotr Stanczyk, Sabela Ramos Garea, Matthieu Geist, and Olivier Bachem. 2024. On-policy distilla- tion of language models: Learning from self-generated mistakes. InInternational Conference on Learning Representations, Vol. 2024. 21246–21263

    work page 2024

  2. [2]

    Benjamin Coleman, Wang-Cheng Kang, Matthew Fahrbach, Ruoxi Wang, Lichan Hong, Ed H Chi, and Derek Zhiyuan Cheng. 2023. Unified Embedding: Battle-tested feature representations for web-scale ML systems.arXiv preprint arXiv:2305.12102(2023)

    arXiv 2023

  3. [3]

    Haoran Ding, Wenlin Zhao, Yuchen Jiang, Jie Zhu, Juren Li, Xinchun Li, Yishujie Zhao, Yi Zhang, Ao Qiao, Jianhui Dong, Cheng Chen, Ziyan Gong, Deping Xie, Peng Xu, Zikai Wang, Yuwei Wang, Huizhi Yang, Zhe Chen, and Yuchao Zheng. 2026. Rec-Distill: An Industrial Distillation Pipeline for Large-Scale Recommendation Models.arXiv preprint arXiv:2605.29755(202...

    work page Pith review arXiv 2026

  4. [4]

    Vianne R Gao, Chen Xue, Marc Versage, Xie Zhou, Zhongruo Wang, Chao Li, Yeon Seonwoo, Nan Chen, Zhen Ge, Gourab Kundu, et al . 2025. Synergen: Contextualized generative recommender for unified search and recommendation. arXiv preprint arXiv:2509.21777(2025)

    arXiv 2025

  5. [5]

    Chengcheng Guo, Kuo Cai, Yu Zhou, Qiang Luo, Ruiming Tang, Han Li, Kun Gai, and Guorui Zhou. 2026. PROMISE: Process Reward Models Unlock Test-Time Gryphon-v2 Scaling Laws in Generative Recommendations.arXiv preprint arXiv:2601.04674 (2026). doi:10.48550/arXiv.2601.04674

    doi:10.48550/arxiv.2601.04674 2026

  6. [6]

    Xuegang Hao, Ming Zhang, Alex Li, et al. 2025. OxygenREC: An Instruction- Following Generative Framework for E-commerce Recommendation.arXiv preprint arXiv:2512.22386(2025). https://arxiv.org/abs/2512.22386

    arXiv 2025

  7. [7]

    Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. 2015. Distilling the Knowledge in a Neural Network.arXiv preprint arXiv:1503.02531(2015). https://arxiv.org/abs/ 1503.02531

    arXiv 2015

  8. [8]

    Shali Jiang, Hua Zheng, Boyang Liu, Laming Chen, Kenny Lov, Chuanqi Xu, Lisang Ding, Qinghai Zhou, Can Cui, Xiaolong Liu, Xiaoyi Liu, Yasmine Badr, Xin Xu, Jiyan Yang, Ellie Dingqiao Wen, and Huayu Li. 2026. LoopFM: Learning frOm HistOrical RePresentations of Foundation Model for Recommendation. arXiv preprint arXiv:2605.29280(2026). https://arxiv.org/abs...

    work page Pith review arXiv 2026

Show all 34 references
  1. [9]

    SeongKu Kang, Junyoung Hwang, Wonbin Kweon, and Hwanjo Yu. 2020. DE- RRD: A knowledge distillation framework for recommender system. InPro- ceedings of the 29th ACM international conference on information & knowledge management. 605–614

    2020

  2. [10]

    Kirill Khrylchenko, Artem Matveev, Sergei Makeev, and Vladimir Baikalov. 2025. Scaling Recommender Transformers to One Billion Parameters.arXiv preprint arXiv:2507.15994(2025)

    2025

  3. [11]

    Jae-woong Lee, Minjin Choi, Jongwuk Lee, and Hyunjung Shim. 2019. Collabora- tive distillation for top-N recommendation. In2019 IEEE international conference on data mining (ICDM). IEEE, 369–378

    2019

  4. [12]

    Dai Li, Kevin Course, Wei Li, Hongwei Li, Jie Hua, Yiqi Chen, Zhao Zhu, Rui Jian, Xuan Cao, Bi Xue, Yu Shi, Jing Qian, Kai Ren, Matt Ma, Qunshu Zhang, and Rui Li. 2025. Realizing Scaling Laws in Recommender Systems: A Foundation–Expert Paradigm for Hyperscale Model Deployment....

    2025 arXiv

  5. [13]

    Hanyu Li, Yi-Ping Hsu, Aditya Mantha, Prabhat Agarwal, Laksh Bhasin, Jialu Wang, Hongtao Lin, Bella Huang, Yaxin Li, Xinyi Li, et al . 2026. UniPinRec: Unifying Generative Retrieval and Ranking at Pinterest Scale.arXiv preprint arXiv:2606.00422(2026)

    2026 arXiv

  6. [14]

    Yaxuan Li, Yuxin Zuo, Bingxiang He, Jinqian Zhang, Chaojun Xiao, Cheng Qian, Tianyu Yu, Huan-ang Gao, Wenkai Yang, Zhiyuan Liu, et al. 2026. Rethinking on-policy distillation of large language models: Phenomenology, mechanism, and recipe.arXiv preprint arXiv:2604.13016(2026)

    2026 arXiv

  7. [15]

    Mingfu Liang, Xi Liu, Rong Jin, Boyang Liu, Qiuling Suo, Qinghai Zhou, Song Zhou, Laming Chen, Hua Zheng, Zhiyuan Li, et al. 2025. External Large Foun- dation Model: How to Efficiently Serve Trillions of Parameters for Online Ads Recommendation. InCompanion Proceedings of the ...

    2025

  8. [16]

    Yu Liu and Jiangxia Cao. 2026. Harmonizing Generative Retrieval and Ranking in Chain-of-Recommendation.arXiv preprint arXiv:2604.25787(2026). https: //arxiv.org/abs/2604.25787

    2026 arXiv

  9. [17]

    Xinchen Luo, Jiangxia Cao, Tianyu Sun, Jinkai Yu, Rui Huang, Wei Yuan, Hezheng Lin, Yichen Zheng, Shiyao Wang, Qigen Hu, et al . 2025. Qarm: Quantitative alignment multi-modal recommendation at kuaishou. InProceedings of the 34th ACM International Conference on Information and...

    2025

  10. [18]

    OneRec Team. 2025. OneRec Technical Report.arXiv preprint arXiv:2506.13695 (2025). https://arxiv.org/abs/2506.13695

    2025

  11. [19]

    Junyan Qiu et al. 2025. EGA-V1: Unifying Online Advertising with End-to-End Learning.arXiv preprint arXiv:2505.19755(2025)

    2025 arXiv

  12. [20]

    Tran, Jonah Samost, Maciej Kula, Ed H

    Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan Hulikal Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Q. Tran, Jonah Samost, Maciej Kula, Ed H. Chi, and Maheswaran Sathiamoorthy. 2023. Recommender Systems with Generative Retrieval. InAdvances in Neural In...

    2023

  13. [21]

    Hao-Jun Michael Shi, Dheevatsa Mudigere, Maxim Naumov, and Jiyan Yang

  14. [22]

    Jiawei Sun, Jun Yang, Ziyue Guo, Dongyue Xu, Jianan Yan, Lifang Deng, and Xiaoyi Zeng. 2026. UniSGR: Unified Framework for Semantic ID Generation and Ranking.arXiv preprint arXiv:2607.04068(2026)

    2026 arXiv

  15. [23]

    Yijia Sun, Shanshan Huang, Zhiyuan Guan, Qiang Luo, Ruiming Tang, Kun Gai, and Guorui Zhou. 2026. GRank: Towards Target-Aware and Streamlined Industrial Retrieval with a Generate-Rank Framework. InProceedings of the ACM Web Conference 2026. 7798–7808

    2026

  16. [24]

    Jiaxi Tang and Ke Wang. 2018. Ranking Distillation: Learning Compact Ranking Models with High Performance for Recommender System. InProceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 2289–2298. doi:10.1145/3219819.3220021

    2018

  17. [25]

    Daria Tikhonovich, Oleg Sorokin, Vladislav Dodonov, Mariia Ulianova, and Ilya Murzin. 2026. Gryphon: A Unified Architecture for Semantic-ID Gener- ation and Item-Level Scoring in Industrial Recommendations.arXiv preprint arXiv:2606.08604(2026)

    2026 arXiv

  18. [26]

    Shen Wang et al. 2026. OneLive: Dynamically Unified Generative Framework for Live-Streaming Recommendation.arXiv preprint arXiv:2602.08612(2026)

    2026

  19. [27]

    Zhipeng Wei et al. 2025. OneLoc: Geo-Aware Generative Recommender Systems for Local Life Service.arXiv preprint arXiv:2508.14646(2025)

    2025 arXiv

  20. [28]

    Jun Zhang, Yi Li, Yue Liu, et al. 2025. GPR: Towards a Generative Pre-trained One- Model Paradigm for Large-Scale Advertising Recommendation.arXiv preprint arXiv:2511.10138(2025). https://arxiv.org/abs/2511.10138

    2025

  21. [29]

    Kun Zhang et al. 2026. OneMall: One Architecture, More Scenarios — End-to- End Generative Recommender Family at Kuaishou E-Commerce.arXiv preprint arXiv:2601.21770(2026)

    2026

  22. [30]

    Carolina Zheng, Minhui Huang, Dmitrii Pedchenko, Kaushik Rangadurai, Siyu Wang, Fan Xia, Gaby Nahum, Jie Lei, Yang Yang, Tao Liu, et al. 2025. Enhancing embedding representation stability in recommendation systems with semantic id. InProceedings of the Nineteenth ACM Conferenc...

    2025

  23. [31]

    Zuowu Zheng et al. 2025. EGA-V2: An End-to-End Generative Framework for Industrial Advertising.arXiv preprint arXiv:2505.17549(2025)

    2025 arXiv

  24. [32]

    Guorui Zhou, Hengrui Hu, Hongtao Cheng, et al. 2025. OneRec-V2 Technical Report.arXiv preprint arXiv:2508.20900(2025). https://arxiv.org/abs/2508.20900

    2025 arXiv

  25. [33]

    Jieming Zhu, Jinyang Liu, Weiqi Li, Jincai Lai, Xiuqiang He, Liang Chen, and Zibin Zheng. 2020. Ensembled CTR prediction via knowledge distillation. In Proceedings of the 29th ACM international conference on information & knowledge management. 2941–2958. A Hyperparameters and ...

    2020

  26. [2020]

    InProceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining

    Compositional embeddings using complementary partitions for memory- efficient recommendation systems. InProceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 165–175

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.