# How to start embedding models

**URL:** <https://discuss.vllm.ai/t/how-to-start-embedding-models/55>\
**Category:** General\
**Created:** [March 20, 2025, 3:55am UTC](https://discuss.vllm.ai/t/how-to-start-embedding-models/55 "2025-03-20T03:55:06Z")\
**Posts on this page:** 8\
**Page:** 1

<div class="post-metadata">

**Author:** ![dugutianxue](https://yyz2.discourse-cdn.com/flex036/user_avatar/discuss.vllm.ai/dugutianxue/32/12_2.png) [@dugutianxue](https://discuss.vllm.ai/u/dugutianxue)\
**Post date:** [March 20, 2025, 3:55am UTC](https://discuss.vllm.ai/t/how-to-start-embedding-models/55/1 "2025-03-20T03:55:06Z")

</div>

提个问题 vllm如何跑 embadding 用正常跑模型的参数 貌似无法跑embadding模型 求大佬解答  
A question: How to run embadding with vllm? I can’t seem to run an embadding model with the normal model parameters, so I’m looking for an answer.

---

<div class="post-metadata">

**Author:** ![youkaichao](https://avatars.discourse-cdn.com/v4/letter/y/ac8455/32.png) [@youkaichao](https://discuss.vllm.ai/u/youkaichao)\
**Post date:** [March 20, 2025, 5:39am UTC](https://discuss.vllm.ai/t/how-to-start-embedding-models/55/2 "2025-03-20T05:39:48Z")

</div>

you can check [Pooling Models — vLLM](https://docs.vllm.ai/en/latest/models/pooling_models.html)

---

<div class="post-metadata">

**Author:** ![dugutianxue](https://yyz2.discourse-cdn.com/flex036/user_avatar/discuss.vllm.ai/dugutianxue/32/12_2.png) [@dugutianxue](https://discuss.vllm.ai/u/dugutianxue)\
**Post date:** [March 20, 2025, 5:49am UTC](https://discuss.vllm.ai/t/how-to-start-embedding-models/55/3 "2025-03-20T05:49:34Z")

</div>

这个没有看明白用途，有使用案例可以参考么！！

---

<div class="post-metadata">

**Author:** ![youkaichao](https://avatars.discourse-cdn.com/v4/letter/y/ac8455/32.png) [@youkaichao](https://discuss.vllm.ai/u/youkaichao)\
**Post date:** [March 20, 2025, 5:56am UTC](https://discuss.vllm.ai/t/how-to-start-embedding-models/55/4 "2025-03-20T05:56:39Z")

</div>

better to use english in this forum.

---

<div class="post-metadata">

**Author:** ![dugutianxue](https://yyz2.discourse-cdn.com/flex036/user_avatar/discuss.vllm.ai/dugutianxue/32/12_2.png) [@dugutianxue](https://discuss.vllm.ai/u/dugutianxue)\
**Post date:** [March 20, 2025, 6:09am UTC](https://discuss.vllm.ai/t/how-to-start-embedding-models/55/5 "2025-03-20T06:09:43Z")

</div>

Didn’t see the use for this one, is there a use case to look at!

---

<div class="post-metadata">

**Author:** ![Robin](https://avatars.discourse-cdn.com/v4/letter/r/e5b9ba/32.png) [@Robin](https://discuss.vllm.ai/u/Robin)\
**Post date:** [March 20, 2025, 6:14am UTC](https://discuss.vllm.ai/t/how-to-start-embedding-models/55/6 "2025-03-20T06:14:41Z")

</div>

It is possible to calculate similarity based on the hidden state embedding for recall.

---

<div class="post-metadata">

**Author:** ![DarkLight1337](https://yyz2.discourse-cdn.com/flex036/user_avatar/discuss.vllm.ai/darklight1337/32/570_2.png) [@DarkLight1337](https://discuss.vllm.ai/u/DarkLight1337)\
**Post date:** [March 20, 2025, 6:25am UTC](https://discuss.vllm.ai/t/how-to-start-embedding-models/55/7 "2025-03-20T06:25:58Z")

</div>

> [@dugutianxue](#):
>
> Didn’t see the use for this one, is there a use case to look at!

You can follow the example at [vllm/examples/offline\_inference/basic/embed.py at main · vllm-project/vllm · GitHub](https://github.com/vllm-project/vllm/blob/main/examples/offline_inference/basic/embed.py)

---

<div class="post-metadata">

**Author:** ![DarkLight1337](https://yyz2.discourse-cdn.com/flex036/user_avatar/discuss.vllm.ai/darklight1337/32/570_2.png) [@DarkLight1337](https://discuss.vllm.ai/u/DarkLight1337)\
**Post date:** [March 20, 2025, 6:27am UTC](https://discuss.vllm.ai/t/how-to-start-embedding-models/55/8 "2025-03-20T06:27:40Z")

</div>

> It is possible to calculate similarity based on the hidden state embedding for recall.

vLLM supports `LLM.score`, please check out this example: [vllm/examples/offline\_inference/basic/score.py at main · vllm-project/vllm · GitHub](https://github.com/vllm-project/vllm/blob/main/examples/offline_inference/basic/score.py)

Our OpenAI-compatible server also has a Score API that does the same thing.