fix(web): fall back to backend defaults for unset search LLM parameters (#19079)

Fix search app LLM settings defaulting to 0 and overriding backend defaults on save. top_p=0 collapses sampling. Needs proper default handling.

web/src/pages/next-search/llm-setting-defaults.test.ts

@@ -0,0 +1,55 @@
+import { llmSettingDefaults, resolveInitialLlmSetting } from './llm-setting-defaults';
+
+// The four generation parameters must fall back to the backend defaults
+// (LLM_SETTING_DEFAULTS in api/db/services/llm_service.py), not to 0: an
+// unset top_p persisted as 0 would collapse nucleus sampling on the next
+// save because the enabled flags default to true.
+
+describe('resolveInitialLlmSetting', () => {
+  test('unset llm_setting falls back to the backend defaults', () => {
+    expect(resolveInitialLlmSetting(undefined)).toEqual(llmSettingDefaults);
+    expect(resolveInitialLlmSetting(null)).toEqual(llmSettingDefaults);
+    expect(resolveInitialLlmSetting({})).toEqual(llmSettingDefaults);
+  });
+
+  test('stored values are kept', () => {
+    expect(
+      resolveInitialLlmSetting({
+        temperature: 0.8,
+        top_p: 0.9,
+        frequency_penalty: 0.2,
+        presence_penalty: 0.5,
+      }),
+    ).toEqual({
+      temperature: 0.8,
+      top_p: 0.9,
+      frequency_penalty: 0.2,
+      presence_penalty: 0.5,
+    });
+  });
+
+  test('a stored 0 is a deliberate choice and must not be replaced', () => {
+    expect(
+      resolveInitialLlmSetting({
+        temperature: 0,
+        top_p: 0,
+        frequency_penalty: 0,
+        presence_penalty: 0,
+      }),
+    ).toEqual({
+      temperature: 0,
+      top_p: 0,
+      frequency_penalty: 0,
+      presence_penalty: 0,
+    });
+  });
+
+  test('partially stored values only fill the missing ones', () => {
+    expect(resolveInitialLlmSetting({ temperature: 0.5 })).toEqual({
+      temperature: 0.5,
+      top_p: llmSettingDefaults.top_p,
+      frequency_penalty: llmSettingDefaults.frequency_penalty,
+      presence_penalty: llmSettingDefaults.presence_penalty,
+    });
+  });
+});

web/src/pages/next-search/llm-setting-defaults.ts

@@ -0,0 +1,50 @@
+/*
+ *  Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+// Mirrors LLM_SETTING_DEFAULTS in api/db/services/llm_service.py: when a
+// generation parameter was never saved, the backend resolves it to these
+// defaults. Falling back to 0 in the form would persist 0 as an explicit
+// value and override the backend defaults on the next save (e.g. top_p 0
+// collapses nucleus sampling).
+export const llmSettingDefaults = {
+  temperature: 0.1,
+  top_p: 0.3,
+  frequency_penalty: 0.7,
+  presence_penalty: 0.4,
+} as const;
+
+type LlmSettingNumericFields = {
+  temperature?: number | null;
+  top_p?: number | null;
+  frequency_penalty?: number | null;
+  presence_penalty?: number | null;
+};
+
+/**
+ * Fill the four generation parameters for the form's initial values. A stored
+ * 0 is a deliberate choice and must be kept; only unset values fall back to
+ * the backend defaults.
+ */
+export const resolveInitialLlmSetting = (
+  llm_setting?: LlmSettingNumericFields | null,
+) => ({
+  temperature: llm_setting?.temperature ?? llmSettingDefaults.temperature,
+  top_p: llm_setting?.top_p ?? llmSettingDefaults.top_p,
+  frequency_penalty:
+    llm_setting?.frequency_penalty ?? llmSettingDefaults.frequency_penalty,
+  presence_penalty:
+    llm_setting?.presence_penalty ?? llmSettingDefaults.presence_penalty,
+});

web/src/pages/next-search/search-setting.tsx

@@ -53,6 +53,7 @@ import {
   useUpdateSearch,
 } from '../next-searches/hooks';
 import { RerankFormFields } from '@/components/rerank';
+import { resolveInitialLlmSetting } from './llm-setting-defaults';
 import {
   SearchSettingFormData,
   useRevalidatePersistedModels,
@@ -116,10 +117,7 @@ function SearchSetting({
         llm_setting: {
           llm_id: search_config?.chat_id || '',
           parameter: llm_setting?.parameter || '',
-          temperature: llm_setting?.temperature || 0,
-          top_p: llm_setting?.top_p || 0,
-          frequency_penalty: llm_setting?.frequency_penalty || 0,
-          presence_penalty: llm_setting?.presence_penalty || 0,
+          ...resolveInitialLlmSetting(llm_setting),
         },
         chat_settingcross_languages: [],
         highlight: false,