--- 格式版本: 2 标题: "Sona Technical Report" 原文链接: "https://arxiv.org/abs/2608.11015" 发布日期: "2026-08-11" 发布时间校准状态: "found" 发布时间需复核: "否" 发布时间来源: "rule:local:strict_original_body" 发布时间证据: "**\\[v1\\]** Tue, 11 Aug 2026 14:58:26 UTC (520 KB)" 发布时间校准原因: "规则确认唯一严格发布时间，来源 local:strict_original_body" 发布时间校准置信度: "high" 发布时间候选数量: 18 发布时间严格候选数量: 6 发布时间原页读取状态: "source template page reused from URL open" 发布时间未找到原因: "" 发布时间校准时间: "2026-08-12T11:15:10+08:00" 发布时间仲裁状态: "skipped" 发布时间仲裁尝试次数: 0 发布时间仲裁耗时毫秒: 0 发现时间: "2026-08-12T11:13:54+08:00" 入库时间: "2026-08-12T03:15:10.387Z" 来源平台: "arXiv 学术论文搜索" 搜索渠道: "source_template" 搜索词: "https://arxiv.org/search/?query=deployment&searchtype=all" 匹配关键词: - "deployment" 相关厂家: [] 相关专家: [] 内容类型: "网页" 抓取工具: "Free Fetch + Defuddle" 清洗工具: "Defuddle Markdown + Defuddle/Readability 正文提取" 原始附件: [] AI优质: "否" AI打分: 10 AI分档: "非优质" AI质检状态: "不通过" AI打分理由: "论文主题为Yandex Music推荐系统，属于信息检索领域，与超节点/AI Rack/机柜级AI基础设施完全无关。" AI质检模型: "ali-deepseek-v4-flash" AI质检时间: "2026-08-12T11:15:15+08:00" AI主题相关性: 0 AI来源权威性: 5 AI新颖性: 0 AI技术细节: 0 AI商业部署信号: 0 AI完整性: 5 采集批次: "2026年8月12日2点04分20秒" 采集批次ID: "20260812-020420-289" 去重键: "https://arxiv.org/abs/2608.11015" --- ## Computer Science > Information Retrieval ## Title:Sona Technical Report Authors:[Sona Team](https://arxiv.org/search/cs?searchtype=author&query=Sona+Team): [Alexandr Udeneev](https://arxiv.org/search/cs?searchtype=author&query=Udeneev,+A), [Aleksei Krasilnikov](https://arxiv.org/search/cs?searchtype=author&query=Krasilnikov,+A), [Alexey Nadtochiy](https://arxiv.org/search/cs?searchtype=author&query=Nadtochiy,+A), [Andrey Semenov](https://arxiv.org/search/cs?searchtype=author&query=Semenov,+A), [Andrey Tsyrkunov](https://arxiv.org/search/cs?searchtype=author&query=Tsyrkunov,+A), [Anna Krivonos](https://arxiv.org/search/cs?searchtype=author&query=Krivonos,+A), [Anna Lipkina](https://arxiv.org/search/cs?searchtype=author&query=Lipkina,+A), [Artem Matveev](https://arxiv.org/search/cs?searchtype=author&query=Matveev,+A), [Daniil Burlakov](https://arxiv.org/search/cs?searchtype=author&query=Burlakov,+D), [Daniil Leschev](https://arxiv.org/search/cs?searchtype=author&query=Leschev,+D), [Daria Tikhonovich](https://arxiv.org/search/cs?searchtype=author&query=Tikhonovich,+D), [Denis Burshtein](https://arxiv.org/search/cs?searchtype=author&query=Burshtein,+D), [Ekaterina Dmitrieva](https://arxiv.org/search/cs?searchtype=author&query=Dmitrieva,+E), [Eugene Krofto](https://arxiv.org/search/cs?searchtype=author&query=Krofto,+E), [Grigorii Khlystov](https://arxiv.org/search/cs?searchtype=author&query=Khlystov,+G), [Ilya Murzin](https://arxiv.org/search/cs?searchtype=author&query=Murzin,+I), [Kirill Golovko](https://arxiv.org/search/cs?searchtype=author&query=Golovko,+K), [Ksenia Sycheva](https://arxiv.org/search/cs?searchtype=author&query=Sycheva,+K), [Leonid Dmitriev](https://arxiv.org/search/cs?searchtype=author&query=Dmitriev,+L), [Mariia Rozaeva](https://arxiv.org/search/cs?searchtype=author&query=Rozaeva,+M), [Mariia Ulianova](https://arxiv.org/search/cs?searchtype=author&query=Ulianova,+M), [Mikhail Sandul](https://arxiv.org/search/cs?searchtype=author&query=Sandul,+M), [Nikolai Savushkin](https://arxiv.org/search/cs?searchtype=author&query=Savushkin,+N), [Oleg Sorokin](https://arxiv.org/search/cs?searchtype=author&query=Sorokin,+O), [Roman Odobesku](https://arxiv.org/search/cs?searchtype=author&query=Odobesku,+R), [Semyon Panenko](https://arxiv.org/search/cs?searchtype=author&query=Panenko,+S), [Sergei Liamaev](https://arxiv.org/search/cs?searchtype=author&query=Liamaev,+S), [Sergei Makeev](https://arxiv.org/search/cs?searchtype=author&query=Makeev,+S), [Vadim Shilov](https://arxiv.org/search/cs?searchtype=author&query=Shilov,+V), [Veronika Ivanova](https://arxiv.org/search/cs?searchtype=author&query=Ivanova,+V), [Viktor Yanush](https://arxiv.org/search/cs?searchtype=author&query=Yanush,+V), [Vladimir Baikalov](https://arxiv.org/search/cs?searchtype=author&query=Baikalov,+V), [Vladislav Dodonov](https://arxiv.org/search/cs?searchtype=author&query=Dodonov,+V), [Vladislav Tytskiy](https://arxiv.org/search/cs?searchtype=author&query=Tytskiy,+V) [View PDF](https://arxiv.org/pdf/2608.11015) [HTML (experimental)](https://arxiv.org/html/2608.11015v1) > Abstract:We introduce Sona, a single-model generative recommender for Yandex Music. In an online A/B test, Sona replaced the entire production cascade, comprising more than 15 candidate generators followed by pre-ranking and ranking models that consume hundreds of features, including signals from large transformer models such as Argus and target-attention scorers, while significantly improving key engagement metrics. > The architecture of Sona unifies candidate generation and ranking around a shared user representation. Its encoder transforms the user's chronological sequence of logged engagement events into hidden states consumed by both the autoregressive decoder and the Ranking Module. The next-token-prediction and distillation objectives jointly update the encoder, coupling generation and ranking through the same user state. Neither Sona nor its Teacher Ranker uses hand-engineered features; both operate on logged event fields and learned item representations. In the final Sona configuration, the larger teacher supplies ranking targets during training but is absent from serving, leaving the encoder, decoder, and Ranking Module as a single deployed model. > We evaluate Sona in an online A/B experiment using live traffic from My Vibe on smart speakers, one of Yandex Music's largest recommendation surfaces. Relative to the production control, Sona produced statistically significant uplifts of 4.53% in Active Users, the primary metric, 6.30% in Total Listening Time, and 11.42% in Likes. These effects were incremental to improvements retained from preceding deployments. The Active Users uplift was 2.35 times the increment previously delivered by Argus, the strongest model deployed on this surface before Sona. These results show that a single jointly trained model can replace a mature multi-stage recommendation cascade while improving recommendation quality on live traffic. | Subjects: | Information Retrieval (cs.IR) | | --- | --- | | Cite as: | [arXiv:2608.11015](https://arxiv.org/abs/2608.11015) \[cs.IR\] | | | (or [arXiv:2608.11015v1](https://arxiv.org/abs/2608.11015v1) \[cs.IR\] for this version) | | | [https://doi.org/10.48550/arXiv.2608.11015](https://doi.org/10.48550/arXiv.2608.11015) | ## Submission history From: Daria Tikhonovich \[[view email](https://arxiv.org/show-email/28151bd6/2608.11015)\] **\[v1\]** Tue, 11 Aug 2026 14:58:26 UTC (520 KB) [Which authors of this paper are endorsers?](https://arxiv.org/auth/show-endorsers/2608.11015) | Disable MathJax ([What is MathJax?](https://info.arxiv.org/help/mathjax.html))