Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

EMR Serverless StarRocks adds multimodal Lakehouse analytics

AI summary

Alibaba Cloud's EMR Serverless StarRocks product introduction details Stella 3.0, a commercial engine that combines OLAP, KV lookups, full-text search and vector retrieval in one SQL workflow. StarRocks says this unified approach can support multimodal hybrid search, AI training-sample selection and RAG retrieval over Paimon, Lance and Iceberg lake tables. The article also reports benchmark results and agent skills, but these capabilities and performance figures come from vendor claims.

In-depth insight

The release puts multimodal retrieval and analysis behind a single SQL interface, directly addressing workflows that otherwise split search, filtering and aggregation across separate systems.

作者:范振,StarRocks TSC Member;阿里云开源 OLAP 负责人

构建 Lakehouse 原生 OLAP 引擎,一直是 StarRocks 的首要方向。用户可以把业务数据留在开放湖表里,直接用 StarRocks 做多表关联、聚合和交互式分析。报表和临时查询都可以围绕同一份湖上数据展开。

近两年,随着 AI 应用和模型训练越来越多地用到图片、视频、音频和文本,StarRocks 加大了对多模态分析与检索的投入。做报表时,分析师通常知道该查哪些表。准备训练数据时,工程师手里可能只有一句描述或一张参考图,需要先找到一批相关样本,再决定怎样筛选和分析这批数据。

为什么要统一分析与检索

拿雨夜道路图片做训练集为例,团队希望用同一条 SQL 完成两件事。先取少量样本预览,确认找出的图片符合预期。条件确定后,只需调大返回数量,就能用同一套检索和筛选条件生成训练样本清单。

过去,这类任务往往由两个系统接力完成。检索服务先找一批相似图片,把候选 ID 交给 OLAP 引擎,再关联标签和采集记录,筛选并统计样本。小批量预览时,跨系统传一份清单的开销有限。要生成几万张图的训练集,候选清单也跟着变大,筛完数量不够还得回到检索端补查。

#Stella 3.0#EMR Serverless StarRocks#StarRocks#Alibaba Cloud#Paimon#Lance#Iceberg#Qwen

Original · 13 min readRead the original
Group chat

No comments yet. Start the conversation.