Text & Chat Adds Jev Model
The Jev model has been added to VOD Text & Chat. Call it with POST https://mmu.vod-qcloud.com/v1/systemone: it takes a state (state) plus structured questions and returns typed answers with calibrated probabilities (no generated text).
Model IDs:
jev-1.14.0/jev-1.13.0/jev-latestQuestion types:
choice(classification / routing),noul(yes / no probability),score(graded scoring),box(locating a target in an image, 1.14+ experimental)Up to 8 questions per request; billed on input tokens only (output tokens are free)
Image input (PNG / JPEG) and Box are 1.14+ experimental features and require the
TypeSafe-Danger-Zone-Images: trueheader