Skip to content

fix(config): reduce default reranker_max_candidates from 300 to 50 (#2554)#2796

Closed
handnewb wants to merge 1 commit into
vectorize-io:mainfrom
handnewb:fix/reranker-max-candidates-default
Closed

fix(config): reduce default reranker_max_candidates from 300 to 50 (#2554)#2796
handnewb wants to merge 1 commit into
vectorize-io:mainfrom
handnewb:fix/reranker-max-candidates-default

Conversation

@handnewb

Copy link
Copy Markdown
Contributor

Summary

The default reranker_max_candidates=300 causes 30s+ recall latency because most reranker providers are optimized for batches of 50-100 documents, not 300. This latency cliff affects all users who don't explicitly configure this value.

Changes

  • Changed DEFAULT_RERANKER_MAX_CANDIDATES from 300 to 50

Operators who need larger batches can set HINDSIGHT_API_RERANKER_MAX_CANDIDATES explicitly.

Closes #2554

The default of 300 reranker candidates causes 30s+ recall latency
for the common case. Most reranker providers are optimized for batches
of 50-100 documents. 50 is a safer default that avoids the latency
cliff while still providing enough candidates for good recall quality.

Operators who need larger batches can set
HINDSIGHT_API_RERANKER_MAX_CANDIDATES explicitly.

Closes vectorize-io#2554
@handnewb handnewb closed this Jul 17, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[Bug] reranker_max_candidates default=300 causes 30s+ recall latency — should default to 50

1 participant