Default chunk size too big for default embedding model? #2391

Closed
opened 2025-11-11 15:06:17 -06:00 by GiteaMirror · 1 comment
Owner

Originally created by @lbluett on GitHub (Oct 16, 2024).

Hi all, this may be due to my own ignorance but isn't the default chunk size too big for the default embedding model?

The default embedding model is: sentence-transformers/all-MiniLM-L6-v2
The default chunk size is: 1500

Isn't the max sequence length for all-MiniLM-L6-v2 256 tokens? According to: https://www.sbert.net/docs/sentence_transformer/pretrained_models.html

I believe this would result in the 1500 token chunks being truncuated to 256 tokens which means a lot of context is dropped.

Originally created by @lbluett on GitHub (Oct 16, 2024). Hi all, this may be due to my own ignorance but isn't the default chunk size too big for the default embedding model? The default embedding model is: sentence-transformers/all-MiniLM-L6-v2 The default chunk size is: 1500 Isn't the max sequence length for all-MiniLM-L6-v2 256 tokens? According to: https://www.sbert.net/docs/sentence_transformer/pretrained_models.html I believe this would result in the 1500 token chunks being truncuated to 256 tokens which means a lot of context is dropped.
Author
Owner

@thiswillbeyourgithub commented on GitHub (Oct 16, 2024):

I brought it up multiple times and got ignored each time. See #5323

I think it's counting letters instead of tokens, which is complete bonkers IMO

@thiswillbeyourgithub commented on GitHub (Oct 16, 2024): I brought it up multiple times and got ignored each time. See #5323 I think it's counting letters instead of tokens, which is complete bonkers IMO
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/open-webui#2391