←back to thread

Production RAG: what I learned from processing 5M+ documents

(blog.abdellatif.io)
548 points tifa2up | 6 comments | 20 Oct 25 15:55 UTC | HN request time: 0.02s | source | bottom
1. 383toast ◴[20 Oct 25 17:39 UTC] No.45646734[source]▶
They should've tested other embedding models, there are better ones than openai's (and cheaper)
replies(1): >>45646823 #
2. prettyblocks ◴[20 Oct 25 17:44 UTC] No.45646823[source]▶
Which do you suggest?
replies(2): >>45646987 #>>45647899 #
3. roze_sha ◴[20 Oct 25 17:57 UTC] No.45646987[source]▶
https://huggingface.co/spaces/mteb/leaderboard
replies(2): >>45647106 #>>45649118 #
4. 383toast ◴[20 Oct 25 18:06 UTC] No.45647106{3}[source]▶
yep
5. leftnode ◴[20 Oct 25 19:10 UTC] No.45647899[source]▶
The Qwen3 600M and 4B embedding models are near state of the art and aren't too computationally intensive.
6. remz14 ◴[20 Oct 25 20:45 UTC] No.45649118{3}[source]▶
You should use RTEB instead. See here for why: https://huggingface.co/blog/rteb

Here is that leaderboard https://huggingface.co/spaces/mteb/leaderboard?benchmark_nam...

Voyage-3-large seems like SOTA right now