Skip to content

v2.48.1 — embedding input token-budget guard

Choose a tag to compare

View all tags
@CocoRoF CocoRoF released this 08 Jul 02:36
v2.48.1
2426c9b

OpenAI embedding client bounds every input to <=8192 bytes (>= token budget), so an over-long un-chunked text no longer 400s and wedges the embedding path. Truncates only over-budget inputs on a UTF-8 boundary. 4521 tests. See CHANGELOG [2.48.1].

Assets 2
Loading