gmp/trips
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

mock: disable LLM keep-alive by default, opt in with ?warm=1

Browse Source 
The 2-minute warmLLM() probe pinned Qwen3.8-27B-dual on the LM Studio
server (192.168.3.7:1234), preventing other models from loading
reliably. Default is now off; add ?warm=1 to a tab's URL to restore
the warm first-turn behavior for that session.
This commit is contained in:
Greg Pomerantz 2026-09-11 11:05:03 -04:00
parent 6457501520
commit c05e35cc08
1 changed files with 5 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
mock/app.js
View File

@ -122,7 +122,11 @@ async function warmLLM() {
} catch {}
warming = false;
}
setInterval(() => { if (llmOn && !scriptBusy) warmLLM(); }, 120000); // keep the model warm while the tab is open
// Keep-alive is DISABLED by default: the probe pins the big model (Qwen3.8-27B-dual)
// on the LM Studio server and prevents loading other models. Opt in per session
// with ?warm=1 in the URL (accepts the cold first turn cost, see above).
const warmEnabled = new URLSearchParams(location.search).has('warm');
setInterval(() => { if (warmEnabled && llmOn && !scriptBusy) warmLLM(); }, 120000);
// ---------------- web search client (SearXNG via the /search proxy) ------
// Live facts (opening hours, fees, seasonal notes) with provenance URLs —
// the third leg of the 3-source blend. Enrichment APPENDS to suggestion