You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
System monitor example: an Applications list, the default in all
three renderers. Each desktop app shows its combined CPU, summed resident
memory, threads and process count; Show processes opens its members
as a parent/child tree, and All processes lists everything, as
before. Apps are recognized from desktop entries (Linux), .app bundles
(macOS) and window owners (Windows), with process ancestry. Its CI runs
the model tests and builds the Oriel renderer on Windows and macOS.
Fixed
oriel.chat: a Stop that comes after its reply ended no longer stops
the next one. Tokens can reach the page in a burst after the model
finished, and cancel() with no reply running stops the next to start,
so a Stop pressed during the burst left the next message empty. chat:token events now carry their reply's id (reply), and cancelReply(id) stops that reply only while it runs; the showcase's
Stop sends it. cancel() is unchanged.
The showcase's chat UI test (iOS CI) no longer waits 5 minutes for a
reply its own late Stop had cancelled: it stops only while the reply
runs, and asks again after a reply with no tokens.
System monitor (Oriel renderer): picking a sort key (CPU, Memory, ...)
no longer flips the direction; only the direction button does, as in the
web frontend.