ui : stop the conversation model from switching backends

Assisted-by: pi:llama.cpp/DeepSeek-V4.1-Flash

tools/ui/src/lib/components/app/chat/ChatForm/ChatFormActions/ChatFormActionModels.svelte

@@ -1,7 +1,13 @@
 <script lang="ts">
 	import { ModelsSelectorDropdown, ModelsSelectorSheet } from '$lib/components/app';
 	import { useBackendAvailability } from '$lib/hooks/use-backend-availability.svelte';
-	import { conversationsStore, deviceStore, modelsStore, serverStore } from '$lib/stores';
+	import {
+		backendsStore,
+		conversationsStore,
+		deviceStore,
+		modelsStore,
+		serverStore
+	} from '$lib/stores';
 	import { getConversationModel } from '$lib/utils';
 
 	interface Props {
@@ -54,10 +60,16 @@
 
 	$effect(() => {
 		if (conversationModel && conversationModel !== lastSyncedConversationModel) {
-			if (modelsStore.models.some((m) => m.model === conversationModel)) {
+			const option = modelsStore.models.find((m) => m.model === conversationModel);
+
+			// only sync models served by the active backend; a model from another
+			// backend must not yank the active tab (and trigger a full backend
+			// switch) just because the conversation used it. sends resolve their
+			// backend explicitly via ensureModelBackend
+			if (option && option.backendId === backendsStore.active.id) {
 				modelsStore.selectedModelName = conversationModel;
 				modelsStore.selectModelByName(conversationModel);
-			} else {
+			} else if (!option) {
 				modelsStore.selectedModelName = null;
 				modelsStore.clearSelection();
 			}

tools/ui/src/lib/stores/models/props.svelte.ts

@@ -164,6 +164,10 @@ export class ModelPropsManager {
 		// /props only exists on llama.cpp servers
 		if (!serverStore.capabilities.props) return null;
 
+		// props not loaded yet (mid backend switch): the server role and model
+		// load state are not trustworthy, so do not guess
+		if (!serverStore.props) return null;
+
 		const cached = this.cache.get(modelId);
 
 		if (cached) return cached;