From 76e8b8e04c2a586c4c7246b928cfff623a445977 Mon Sep 17 00:00:00 2001
From: duffyduck <info@hacker-net.de>
Date: Sat, 19 Sep 2026 11:45:45 +0200
Subject: [PATCH] fix(llm): Test-Chat-Timeout 15s -> 120s fuer kalten
 Modell-Swap
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

qwen3-8b antwortete in ~13s (unter 15s) und kam durch; qwen3-4b lief in
"Timeout", obwohl kleiner. Ursache ist nicht die Modellgroesse, sondern der
KALTE Modell-Swap: beim ersten Wechsel entlaedt llama-swap das alte Modell
und laedt das neue GGUF frisch in den VRAM (+ erste Inferenz) — das dauert
laenger als die 15s, die sendToRVS_withResponse hart als Timeout hatte.

sendToRVS_withResponse nimmt jetzt ein optionales timeoutMs (Default 15s
unveraendert fuer Voice-List etc.); llm_test uebergibt 120s. Der Adapter
selbst deckt den Swap mit LLM_TIMEOUT_SEC=60 ab und meldet waehrenddessen
llm_status "loading".

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 diagnostic/server.js | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/diagnostic/server.js b/diagnostic/server.js
index 71a6898..34968a6 100644
--- a/diagnostic/server.js
+++ b/diagnostic/server.js
@@ -1246,7 +1246,7 @@ function connectRVS(forcePlain) {
   });
 }
 
-function sendToRVS_withResponse(sendType, sendPayload, expectType, clientWs) {
+function sendToRVS_withResponse(sendType, sendPayload, expectType, clientWs, timeoutMs = 15000) {
   if (!RVS_HOST || !RVS_TOKEN) return;
   const proto = RVS_TLS === "true" ? "wss" : "ws";
   const url = `${proto}://${RVS_HOST}:${RVS_PORT}?token=${RVS_TOKEN}`;
@@ -1254,7 +1254,7 @@ function sendToRVS_withResponse(sendType, sendPayload, expectType, clientWs) {
   const timeout = setTimeout(() => {
     try { freshWs.close(); } catch (_) {}
     clientWs.send(JSON.stringify({ type: expectType, payload: { voices: [], error: "Timeout" }, timestamp: Date.now() }));
-  }, 15000);
+  }, timeoutMs);
   freshWs.on("open", () => {
     freshWs.send(JSON.stringify({ type: sendType, payload: sendPayload, timestamp: Date.now() }));
   });
@@ -2948,7 +2948,7 @@ wss.on("connection", (ws) => {
           messages: [{ role: "user", content: String(msg.text || "Sag kurz Hallo.") }],
           max_tokens: 256, temperature: 0.5,
           model: msg.model || "", targetInstance: msg.targetInstance || "",
-        }, "llm_response", ws);
+        }, "llm_response", ws, 120000);  // 2min: erster Modell-Swap laedt das GGUF kalt (mehrere GB) — 15s reichen dann nicht
         log("info", "llm", `Test-Chat → ${msg.model || "?"} @ ${msg.targetInstance || "(broadcast)"}`);
       } else if (msg.action === "node_stats_stream_start" || msg.action === "node_stats_stream_stop"
                  || msg.action === "node_stats_history_request" || msg.action === "node_stats_reset") {