# Fine tuning existing semantic models

**URL:** <https://community.pinecone.io/t/fine-tuning-existing-semantic-models/223>\
**Category:** General\
**Tags:** nlp, semantic-search\
**Created:** [February 11, 2022, 6:02pm UTC](https://community.pinecone.io/t/fine-tuning-existing-semantic-models/223 "2022-02-11T18:02:19Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![Rash123](https://avatars.discourse-cdn.com/v4/letter/r/7bcc69/32.png) [@Rash123](https://community.pinecone.io/u/Rash123)\
**Post date:** [February 11, 2022, 6:02pm UTC](https://community.pinecone.io/t/fine-tuning-existing-semantic-models/223/1 "2022-02-11T18:02:19Z")

</div>

Semantically if we compare 2 sentences, like “I have a cat” and " I am a cat", they sit very close in the vector space. Is there a way where we can build a model that could differentiate it by the meaning and not the distance.

---

<div class="post-metadata">

**Author:** ![greg](https://sea2.discourse-cdn.com/flex020/user_avatar/community.pinecone.io/greg/32/54_2.png) [@greg](https://community.pinecone.io/u/greg)\
**Post date:** [February 11, 2022, 9:33pm UTC](https://community.pinecone.io/t/fine-tuning-existing-semantic-models/223/2 "2022-02-11T21:33:55Z")

</div>

@Rash123 A decent [sentence transformer](https://www.pinecone.io/learn/sentence-embeddings/) should not put them together in the first place. Is this from an example, or are you just guessing they might be close together since the words are similar?

---

<div class="post-metadata">

**Author:** ![Sanjog](https://avatars.discourse-cdn.com/v4/letter/s/da6949/32.png) [@Sanjog](https://community.pinecone.io/u/Sanjog)\
**Post date:** [February 14, 2022, 4:07am UTC](https://community.pinecone.io/t/fine-tuning-existing-semantic-models/223/3 "2022-02-14T04:07:13Z")

</div>

Agree with @greg “A decent [sentence transformer](https://www.pinecone.io/learn/sentence-embeddings/) should not put them together in the first place.”