openvibers/openvibe.network Commits

Platform AI: all five ways of paying less act (#785)

OP OpenVibers committed 3 days ago · 98a3322e1c · 1 file, +6 −5

Changes

ddiff --git a/public/js/admin/inference.js b/public/js/admin/inference.js
iindex 394d6919..8baf4b47 100644
---- a/public/js/admin/inference.js
++++ b/public/js/admin/inference.js
@@@ -27,13 +27,14 @@ const OIN_STRATEGIES = [
['fastest', 'Fastest', 'The quickest answer, at any price under the ceiling'],
['fixed', 'My order', 'Your list, top first; the next one when it fails'],
];
--// The ways of paying less. `acts`: whether the switch changes calls today (OpenVibe.AI #445: the output limit and the
--// small model first do; the other three are recorded and their savings measured, but do not yet rewire the calls).
++// The ways of paying less. `acts`: whether the switch changes calls today. All five act now: the
++// output limit and the small model first since #445, and the three economy switches since this release
++// (OpenVibe.AI: the batch path, the cache-control markers and warm bias, and the private answer cache).
const OIN_ECONOMY = [
['small_first', 'Try a small model first', 'A large one answers only when the small one cannot.', true],
-- ['batches', 'Send it to a batch when it can wait', 'About half price; the answer comes within a day.', false],
-- ['prompt_cache', 'Keep repeated prompts cached', 'The same opening is paid once, then at a fraction.', false],
-- ['answer_cache', 'Reuse an identical answer', 'Only for the same person asking the same thing.', false],
++ ['batches', 'Send it to a batch when it can wait', 'About half price; the answer comes within a day.', true],
++ ['prompt_cache', 'Keep repeated prompts cached', 'The same opening is paid once, then at a fraction.', true],
++ ['answer_cache', 'Reuse an identical answer', 'Only for the same person asking the same thing.', true],
];
const OIN = { data: null, window: '24h', busy: false, drafts: {} };