跳过导航
People are finding really creative ways to use EmbeddingGemma 2 beyond search.
0:00 / 0:30

People are finding really creative ways to use EmbeddingGemma 2 beyond search.

Google Gemma

@googlegemma11万 位关注者

在 X 上查看
1,931 个赞下载原帖

98,165 次观看

People are finding really creative ways to use EmbeddingGemma 2 beyond search. Instead of running a full multimodal LLM, this community project trains a lightweight classifier on top of EmbeddingGemma 2 in real time. Because the model is compact and natively multimodal, it can learn new categories on the fly and classify inputs locally in ~20ms for text and ~100ms for images.

18 条评论

评论来自 X (Twitter) 上的原帖

  • Google Gemma 的头像

    @googlegemma

    Project by @mmastrac! He also open-sourced the approach as ‘hunch’ if you want to try it yourself: Demo post: GitHub:

    24 个赞

  • Rakesh Gohel 🇨🇦 的头像

    @rakeshgohel01

    EmbeddingGemma 2 is proving that you don't need a massive multimodal LLM for every task.

    2 个赞

  • ApexAiTech 的头像

    @ApexAITech

    Local classification in milliseconds without a huge model is brilliant

    1 个赞

  • Juan Galeas🇭🇳🇨🇦 的头像

    @JuanGaleas7

    I am in love with embedding Gemma 2.

  • Mister SnowCrash 的头像

    @MisterSnowCrash

    Can a category taught with text examples transfer to images in hunch, or does each modality need its own labels? With EmbeddingGemma 2, that could make “teach once, sort both documents and photos” a useful product pattern.

  • Anand Butani 的头像

    @AnandButani

    This is great for app builders who need cheap on-device sorting, like tagging photos or flagging support tickets. I'd use EmbeddingGemma 2 plus a tiny classifier instead of calling a big model per item, since it costs far less and keeps data on the phone.

  • AmirMuhammad Nateghi 的头像

    @AM_Nateghi

    This guy should have studied at MIT 😐😅

  • Shivay Lamba 的头像

    @HowDevelop

    Hey @googlegemma is it possible to get a shout out for this project - using EmbeddingGemma model on a @arduino VentunoQ for multi-modal embedding for clothing suggestions 😁

  • AI Drop 的头像

    @AIDropIntel

    Detail from the hunch repo: it runs on plain CPUs and only takes over once it agrees with its teacher model often enough, then keeps getting spot checked. Cheap reflexes for agents 🧠

  • ReignSwords 的头像

    @ReignSwords

    embedding 顶上套个轻量分类器这招太实用了,实时训练加端侧推理,成本直接砍掉一个量级。好奇问下,实时训练的标注是用户现场给的吗?冷启动时类别很少的情况下效果怎么样

  • Dominik 的头像

    @EternitiesAI

    how small is the classifier trained on top of embeddinggemma?

  • Stoic Tiger 的头像

    @stoic_tiger

    Carbon - where is it?

  • Gabe 的头像

    @fishjar

    Freezing the embedding and only training the head is the real win here — but real-time retraining means your classifier drifts with every new label batch. Version the embedding index or you'll chase ghost regressions.

  • Bojodanchev 的头像

    @BojoDanchev

    how many examples does a new category need before it holds up? if it's under 10, a support team could train its own ticket router without ever calling an ml person

  • Max Petrusenko 的头像

    @petrusenko_max

    a head trained live is only as good as its handful of labeled examples, and drift shows up as inputs landing between categories

  • Alibel Fragoso 的头像

    @Alibel_Fragoso0

    Yup nice

  • Bala Venkatesh 的头像

    @balavenkatesh22

    Amazing

  • Alina Fomina 的头像

    @fominaaalina

    embedding models quietly became the most useful part of the stack. everyone's chasing bigger llms while the real production wins keep coming from small, fast, multimodal embeddings doing 80% of the job at 1% of the cost