From c9e651f7e5346f8fff56bcc5a75506a6868b1c65 Mon Sep 17 00:00:00 2001
From: Concedo <39025047+LostRuins@users.noreply.github.com>
Date: Sun, 1 Mar 2026 00:41:56 +0800
Subject: [PATCH] updated lite, fix some cuda spams, fix qwen3tts voice loading

---
 embd_res/klite.embd             | 65 +++++++++++++++++++++++----------
 ggml/src/ggml-cuda/ggml-cuda.cu |  4 +-
 koboldcpp.py                    | 19 +++++-----
 3 files changed, 57 insertions(+), 31 deletions(-)
diff --git a/embd_res/klite.embd b/embd_res/klite.embd
index fe3d71552..0f64532b2 100644
--- a/embd_res/klite.embd
+++ b/embd_res/klite.embd
@@ -12,7 +12,7 @@ Current version indicated by LITEVER below.
 -->
 <head>
 	<script id="init-config">
-	const LITEVER = 316;
+	const LITEVER = 317;
 	const urlParams = new URLSearchParams(window.location.search);
 	var localflag = urlParams.get('local'); //this will be replaced automatically in embedded kcpp
 	const STORAGE_PREFIX = (localflag?"e_":"")+"kaihordewebui_";
@@ -4291,7 +4291,6 @@ Current version indicated by LITEVER below.
 	var passed_ai_warning_local = false;
 	var welcome = "";
 	var personal_notes = "";
-	var voice_typing_mode = 0; //0=off, 1=on, 2=ptt, 3=ttt
 	var koboldcpp_has_whisper = false; //does backend support voice typing
 	var voice_is_recording = false; //currently recording voice?
 	var voice_is_processing = false; //currently processing voice?
@@ -4407,6 +4406,7 @@ Current version indicated by LITEVER below.
 		tts_stream: false,
 		voice_end_delay: 300,
 		voice_suppress_nonspeech: false,
+		voice_typing_mode: 0,  //0=off, 1=on, 2=ptt, 3=ttt
 		voice_langcode: "auto",
 		tts_speed: 1.0,
 		image_styles: "",
@@ -15645,6 +15645,7 @@ Current version indicated by LITEVER below.
 		document.getElementById("tts_speed").value = localsettings.tts_speed;
 		document.getElementById("voice_end_delay").value = localsettings.voice_end_delay;
 		document.getElementById("voice_suppress_nonspeech").checked = localsettings.voice_suppress_nonspeech;
+		document.getElementById("voice_typing_mode").value = localsettings.voice_typing_mode;
 		document.getElementById("voice_langcode").value = localsettings.voice_langcode;
 		toggle_opmode();
 
@@ -16484,6 +16485,7 @@ Current version indicated by LITEVER below.
 		localsettings.tts_speed = document.getElementById("tts_speed").value;
 		localsettings.voice_end_delay = document.getElementById("voice_end_delay").value;
 		localsettings.voice_suppress_nonspeech = (document.getElementById("voice_suppress_nonspeech").checked?true:false);
+		localsettings.voice_typing_mode = document.getElementById("voice_typing_mode").value;
 		localsettings.voice_langcode = document.getElementById("voice_langcode").value;
 		localsettings.auto_ctxlen = (document.getElementById("auto_ctxlen").checked ? true : false);
 		localsettings.auto_genamt = (document.getElementById("auto_genamt").checked ? true : false);
@@ -16586,8 +16588,8 @@ Current version indicated by LITEVER below.
 		toggle_theme_colors();
 		toggle_sidepanel_mode();
 
-		voice_typing_mode = document.getElementById("voice_typing_mode").value;
-		if(voice_typing_mode>0 && is_using_kcpp_with_whisper())
+		localsettings.voice_typing_mode = document.getElementById("voice_typing_mode").value;
+		if(localsettings.voice_typing_mode>0 && is_using_kcpp_with_whisper())
 		{
 			init_voice_typing();
 		}
@@ -19197,13 +19199,13 @@ Current version indicated by LITEVER below.
 	var recent_voice_duration = 0;
 	function ptt_start()
 	{
-		if(voice_typing_mode==3 && voice_is_speaking)
+		if(localsettings.voice_typing_mode==3 && voice_is_speaking)
 		{
 			//stop toggle to talk
 			ptt_end(true);
 			return;
 		}
-		if(voice_typing_mode>0)
+		if(localsettings.voice_typing_mode>0)
 		{
 			voice_is_speaking = true;
 			++voice_speaking_counter;
@@ -19219,16 +19221,21 @@ Current version indicated by LITEVER below.
 				update_submit_button(false);
 				ptt_start_timestamp = performance.now();
 			}
+			else if(!isVoiceInputConfigured)
+			{
+				//trigger popup for audio permissions
+				init_voice_typing();
+			}
 		}
 	}
 	function ptt_end(force=false)
 	{
-		if(voice_typing_mode==3 && !force)
+		if(localsettings.voice_typing_mode==3 && !force)
 		{
 			return; //do nothing for toggle to talk unless forced
 		}
 		var voice_end_delay = localsettings.voice_end_delay;
-		if(voice_typing_mode>0)
+		if(localsettings.voice_typing_mode>0)
 		{
 			voice_is_speaking = false;
 			let check_speak_counter = voice_speaking_counter;
@@ -19236,7 +19243,7 @@ Current version indicated by LITEVER below.
 				if (voice_is_recording && !voice_is_speaking && voice_speaking_counter == check_speak_counter) {
 					//generate prerecorder blobs (prebuffer 1sec)
 					preaudioblobs = [];
-					if(voice_typing_mode==1)
+					if(localsettings.voice_typing_mode==1)
 					{
 						for(let i=0;i<preaudiobuffers.length;++i)
 						{
@@ -19266,7 +19273,7 @@ Current version indicated by LITEVER below.
 	}
 	function submit_generation_button(aesthetic_ui)
 	{
-		if(voice_typing_mode==0) //only when voice is off
+		if(localsettings.voice_typing_mode==0) //only when voice is off
 		{
 			if(aesthetic_ui)
 			{
@@ -20795,6 +20802,12 @@ Current version indicated by LITEVER below.
 
 				if(targetep.toLowerCase().includes("openrouter.ai"))
 				{
+					let wantsearch = document.getElementById("openroutersearch").checked;
+					if(wantsearch && oai_payload.model && !oai_payload.model.endsWith(":online"))
+					{
+						oai_payload.model += ":online";
+					}
+
 					oaiheaders["HTTP-Referer"] = "https://lite.koboldai.net";
 					if (document.getElementById("openrouterproviders").value.trim() != "") {
 						let provarr = document.getElementById("openrouterproviders").value.trim().split(",").map(item => item.trim());
@@ -22933,7 +22946,7 @@ Current version indicated by LITEVER below.
 		if ('speechSynthesis' in window) {
 			currentlySpeaking = window.speechSynthesis.speaking;
 		}
-		return (voice_typing_mode>0 && is_using_kcpp_with_whisper()
+		return (localsettings.voice_typing_mode>0 && is_using_kcpp_with_whisper()
 			&& !document.getElementById("btnsend").disabled
 			&& !voice_is_processing && !voice_is_recording && isVoiceInputConfigured
 			&& !currentlySpeaking && !tts_is_playing && !is_popup_open());
@@ -23125,7 +23138,7 @@ Current version indicated by LITEVER below.
 		if(navigator.mediaDevices==null)
 		{
 			msgbox("Cannot initialize microphone. If you're using a non-localhost URL, it needs to be served over HTTPS!","Error Starting Microphone");
-			voice_typing_mode = document.getElementById("voice_typing_mode").checked = 0;
+			localsettings.voice_typing_mode = document.getElementById("voice_typing_mode").checked = 0;
 			return;
 		}
 		if (isVoiceInputConfigured) {
@@ -23193,7 +23206,7 @@ Current version indicated by LITEVER below.
 				if (voiceprerecorder.state !== "inactive") {
 					voiceprerecorder.stop();
 				}
-				if(ready_to_record() && voice_typing_mode==1){ //only voice detect needs it
+				if(ready_to_record() && localsettings.voice_typing_mode==1){ //only voice detect needs it
 					voiceprerecorder.start();
 				}
 			}, 500);
@@ -23206,14 +23219,14 @@ Current version indicated by LITEVER below.
 			let options = {
 				source: source,
 				voice_stop: function () {
-					if(voice_typing_mode==1)
+					if(localsettings.voice_typing_mode==1)
 					{
 						console.log("speech stopped");
 						ptt_end();
 					}
 				},
 				voice_start: function () {
-					if(voice_typing_mode==1)
+					if(localsettings.voice_typing_mode==1)
 					{
 						console.log("speech started");
 						ptt_start();
@@ -24171,7 +24184,7 @@ Current version indicated by LITEVER below.
 			document.getElementById("chat_msg_send_btn").classList.remove("showmic");
 			document.getElementById("chat_msg_send_btn").classList.remove("showmiclive");
 			document.getElementById("chat_msg_send_btn").classList.remove("showmicoff");
-			if(voice_typing_mode>0 && is_using_kcpp_with_whisper())
+			if(localsettings.voice_typing_mode>0 && is_using_kcpp_with_whisper())
 			{
 				if (voice_is_processing) {
 					document.getElementById("chat_msg_send_btn").classList.add("showmicoff");
@@ -24181,10 +24194,18 @@ Current version indicated by LITEVER below.
 					document.getElementById("btnsend").innerHTML = "<div class='showmiclivebig'></div><span style='font-size:12px'>Record</span>";
 				} else if (ready_to_record()) {
 					document.getElementById("chat_msg_send_btn").classList.add("showmic");
-					document.getElementById("btnsend").innerHTML = "<div class='showmicbig'></div><span style='font-size:12px'>"+(voice_typing_mode==1?"Standby":(voice_typing_mode==2?"PTT":"TTT"))+"</span>";
+					document.getElementById("btnsend").innerHTML = "<div class='showmicbig'></div><span style='font-size:12px'>"+(localsettings.voice_typing_mode==1?"Standby":(localsettings.voice_typing_mode==2?"PTT":"TTT"))+"</span>";
 				} else {
-					document.getElementById("chat_msg_send_btn").classList.add("showmicoff");
-					document.getElementById("btnsend").innerHTML = "<div class='showmicoffbig'></div><span style='font-size:12px'>Busy</span>";
+					if(isVoiceInputConfigured)
+					{
+						document.getElementById("chat_msg_send_btn").classList.add("showmicoff");
+						document.getElementById("btnsend").innerHTML = "<div class='showmicoffbig'></div><span style='font-size:12px'>Busy</span>";
+					}
+					else
+					{
+						document.getElementById("chat_msg_send_btn").classList.add("showmicoff");
+						document.getElementById("btnsend").innerHTML = "<div class='showmicoffbig'></div><span>Need<br>Access</span>";
+					}
 				}
 			}
 		}
@@ -30908,7 +30929,11 @@ Current version indicated by LITEVER below.
 					</div>
 				</div>
 				<div id="openrouterproviderbox" class="hidden">
-					<div style="display:inline;">Prefer Provider: <input title="Enter Preferred AI Provider" style="height: 25px; font-size:12px;padding:4px;display:inline;width:calc(100% - 110px)" class="form-control" type="text" id="openrouterproviders" placeholder="(Automatic)" value=""></div>
+					<div style="display:inline;">Prefer Provider: <input title="Enter Preferred AI Provider" style="height: 25px; font-size:12px;padding:4px;display:inline;width:calc(100% - 226px)" class="form-control" type="text" id="openrouterproviders" placeholder="(Automatic)" value=""></div>
+					<div style="margin-left:3px;display:inline;">
+						<input type="checkbox" title="OpenRouter Websearch" id="openroutersearch">
+						<div style="padding-left:2px;padding-right:2px" class="box-label">WebSearch</div>
+                    </div>
 				</div>
 
 			</div>
diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
index 763179f18..89d2c35e5 100644
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -3960,7 +3960,7 @@ static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t backend,
                 // Warmup: need at least 2 calls with no property change on the 2nd call
                 if (!properties_changed) {
                     graph->warmup_complete = true;
-                    GGML_LOG_DEBUG("%s: CUDA graph warmup complete\n", __func__);
+                    // GGML_LOG_DEBUG("%s: CUDA graph warmup complete\n", __func__);
                     use_cuda_graph = true;
                     cuda_graph_update_required = true;
                 }
@@ -3970,7 +3970,7 @@ static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t backend,
                 if (properties_changed) {
                     // Properties changed - reset warmup, execute directly until stable again
                     graph->warmup_complete = false;
-                    GGML_LOG_DEBUG("%s: CUDA graph warmup reset\n", __func__);
+                    // GGML_LOG_DEBUG("%s: CUDA graph warmup reset\n", __func__);
                 } else {
                     use_cuda_graph = true;
                     cuda_graph_update_required = graph->instance == nullptr;
diff --git a/koboldcpp.py b/koboldcpp.py
index 4c7b43ed9..ce66d44e6 100755
--- a/koboldcpp.py
+++ b/koboldcpp.py
@@ -8946,7 +8946,7 @@ def kcpp_main_process(launch_args, g_memory=None, gui_launcher=False):
         print("Could not find Embedded MusicUI.")
 
     # load all TTS audio files
-    if args.ttsdir and args.ttsmodel and os.path.isdir(args.ttsdir):
+    if args.ttsmodel:
         try:
             global voicebank, voicelist
             voicebank = {}
@@ -8965,14 +8965,15 @@ def kcpp_main_process(launch_args, g_memory=None, gui_launcher=False):
             voicelist.append("random")
             voicebank["random"] = ""
 
-            for filename in os.listdir(args.ttsdir):
-                if filename.lower().endswith((".mp3", ".wav")):
-                    full_path = os.path.join(args.ttsdir, filename)
-                    with open(full_path, "rb") as f:
-                        encoded = base64.b64encode(f.read()).decode("utf-8")
-                        voicebank[filename] = encoded
-                        voicecount += 1
-                        voicelist.append(os.path.basename(filename))
+            if args.ttsdir and os.path.isdir(args.ttsdir):
+                for filename in os.listdir(args.ttsdir):
+                    if filename.lower().endswith((".mp3", ".wav")):
+                        full_path = os.path.join(args.ttsdir, filename)
+                        with open(full_path, "rb") as f:
+                            encoded = base64.b64encode(f.read()).decode("utf-8")
+                            voicebank[filename] = encoded
+                            voicecount += 1
+                            voicelist.append(os.path.basename(filename))
             print(f"Loaded {voicecount} TTS voices.")
         except Exception:
             print("Could not load TTS voices.")
