# Problemi con gli embeddings (argomenti correlati diventano troppo costosi)

**URL:** <https://meta.discourse.org/t/something-bad-with-embeddings-related-topics-getting-too-costly/340345>\
**Category:** Bug\
**Tags:** ai, related-topics\
**Created:** [4 Dicembre 2024, 11:40am UTC](https://meta.discourse.org/t/something-bad-with-embeddings-related-topics-getting-too-costly/340345 "2024-12-04T11:40:26Z")\
**Posts on this page:** 1\
**Showing post:** 14

<div class="post-metadata">

**Author:** ![Roman](https://sea3.discourse-cdn.com/meta/user_avatar/meta.discourse.org/roman/32/157504_2.png) [@Roman](https://meta.discourse.org/u/Roman)\
**Post date:** [4 Dicembre 2024, 9:09pm UTC](https://meta.discourse.org/t/something-bad-with-embeddings-related-topics-getting-too-costly/340345/14 "2024-12-04T21:09:56Z")

</div>

Sì, penso che il bug che ho corretto ne abbia rivelato un altro che il controllo del digest stava nascondendo.

Penso che il bug sia qui:

> <https://github.com/discourse/discourse-ai/blob/main/app/jobs/scheduled/embeddings_backfill.rb#L79>

L’ho cambiato da `find_each` a `find_in_batches` la settimana scorsa (il primo usa batch internamente) e poiché entrambi si basano su `limit` per specificare la dimensione del batch, il `limit` originale di `limit - rebaked` viene ignorato. Dovremmo usare `pluck` + `each_slice` invece.

---

_[View the full topic](https://meta.discourse.org/t/something-bad-with-embeddings-related-topics-getting-too-costly/340345)._