Page MenuHomePhabricator

Nola | Gpt should not try to complete if result wasn't truncated and new prompt is blank
Open, HighPublic15 PointsBUG

Description

104257 < bavaria>    ↪ vhp — 274 (19/255) tokens — e8c77f4c6217ae3ff80fb422fe5d7c9b
104308 < Tree> +gpt vhp
104309 < bavaria> GPT had nothing else to say :( ↪ q6w — 273 (273/0) tokens — e8c77f4c6217ae3ff80fb422fe5d83e0

The second call was useless as vhp was stopped, not truncated, and the new prompt empty. This made inference costs (273 tokens!) for a call that has no possible output.

Event Timeline

href triaged this task as High priority.Dec 20 2022, 10:58 AM
href created this task.
href shifted this object from the Restricted Space space to the S6 Nola space.
href changed the visibility from "All Users" to "Public (No Login Required)".
href changed the edit policy from "All Users" to "Nola Devs (Project)".