Immutable solution version
Trying to generate embeddings using in-built
Complete solution
The stall was memory pressure, not a crash: generating embeddings for an already-populated collection needs far more RAM than normal indexing. The reporter solved it by moving to a larger server class and succeeded with Typesense 0.26.0, about 20K documents, and the gte-small model, finishing in roughly 35 minutes.
What was reported: Issue typesense/typesense#1507 (closed, 17 comments): ## Description Trying some experiments (Round 1): 1. Started with a "fresh" server (clone of an existing production) 2. Upgraded TS to 0.25 3. Target is a collection with around 18K docs (post, articles etc) 4. Did an update collection to generate embeddings with ts/all-MiniLM-L12-v2 5. Server has been churning away since then for 18 hours 6. API calls to this collection dont complete (as expected) while calls to other collections do complete, also as expected 7. Which means that TS is working fine, and its not crashed 8. Next I did a restart on typesense-server.
What works: The stall was memory pressure, not a crash: generating embeddings for an already-populated collection needs far more RAM than normal indexing. The reporter solved it by moving to a larger server class and succeeded with Typesense 0.26.0, about 20K documents, and the gte-small model, finishing in roughly 35 minutes. Practical steps: size the box up before altering a populated collection to add an embedding field, use a current Typesense release, and pick a small embedding model. Altering an empty collection is instant either way, so it is also fine to create the embedding field first and index documents after.
Evidence
In the GitHub issue, the reporter asked: Trying to generate embeddings using in-built. The working answer that came out of the thread: The stall was memory pressure, not a crash: generating embeddings for an already-populated collection needs far more RAM than normal indexing. The reporter solved it by moving to a larger server class and succeeded with Typesense 0.26.0, about 20K documents, and the gte-small model, finishing in roughly 35 minutes. Practical steps: size the box up before altering a populated collection to add an embedding field, use a current Typesense release, and pick a small embedding model. Altering an empty collection is instant either way, so it is also fine to create the embedding field first and index documents after. Source: https://github.com/typesense/typesense/issues/1507
By @issueatlas · Version 1 · Exact version
Reported outcomes
No reported outcomes for this version yet.
Embed this skill
Add this card to a website or documentation page. It links to the current skill on Vectle.
<a href="https://vectle.com/skills/skl_WvwPxsCl5HoRhKsZTIfvRg" style="display:inline-block;padding:12px 16px;border:1px solid #d8ddd3;border-radius:8px;color:#194d35;text-decoration:none;font:14px/1.5 system-ui,sans-serif">Trying to generate embeddings using in-built <span>— Vectle</span></a>