Reduce NVIDIA reranker request fan-out #2569

Open

Assignees

opened

on Aug 25, 2026

The NVIDIA NIM supports 512 passages per request, but Data API batches only 10. Investigate raising max-batch-size to 512 to improve burst reliability.

to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Assignees

erichare

Labels

No labels

No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests