refactor: standardize max_tokens handling and introduce default constant

- Updated LLMProfile and related components to enforce max_tokens validation, ensuring values are positive and adhere to a global default.
- Removed hardcoded max_tokens limits in various configurations, replacing them with a reference to DEFAULT_MAX_OUTPUT_TOKENS for consistency.
- Enhanced unit tests to validate the new max_tokens behavior and ensure proper lifting of values to the global floor.
- Introduced a new constants file to manage default values, improving maintainability and clarity across the codebase.

Author: shenminglinyi

application/ai/llm_control_service.py

@@ -58,7 +58,14 @@ def _validate_temperature(cls, value: float) -> float:
             raise ValueError('temperature must be between 0 and 2')
         return value
 
-    @field_validator('max_tokens', 'timeout_seconds')
+    @field_validator('max_tokens')
+    @classmethod
+    def _validate_max_tokens(cls, value: int) -> int:
+        if value <= 0:
+            raise ValueError('value must be positive')
+        return max(value, DEFAULT_MAX_OUTPUT_TOKENS)
+
+    @field_validator('timeout_seconds')
     @classmethod
     def _validate_positive_int(cls, value: int) -> int:
         if value <= 0:

application/engine/dag/models.py

@@ -199,7 +199,7 @@ class NodeConfig(BaseModel):
     max_retries: int = Field(default=1, ge=0, le=5)
     timeout_seconds: int = Field(default=60, ge=10, le=600)
     temperature: float = Field(default=0.7, ge=0.0, le=2.0)
-    max_tokens: Optional[int] = Field(default=None, ge=100, le=120000)
+    max_tokens: Optional[int] = Field(default=None, ge=100)
 
 
 # ─── DAG 定义模型 ───

engine/runtime/daemon_host.py

@@ -2012,7 +2012,7 @@ async def _stream_one_beat(
             except Exception:
                 pass  # 降级：无锚点则不加
 
-        # 输出 token 策略与整章生成一致：固定 16000，不随目标字数动态变化。
+        # 输出 token 策略与整章生成一致：统一走全局输出下限，不随目标字数动态变化。
         from engine.runtime.generation_token_policy import CHAPTER_PROSE_MAX_TOKENS
 
         max_tokens = CHAPTER_PROSE_MAX_TOKENS

frontend/src/components/autopilot/NodeEditorDrawer.vue

@@ -49,7 +49,6 @@
             v-model:value="localConfig.maxTokens"
             size="small"
             :min="100"
-            :max="16000"
             :step="100"
             placeholder="默认"
             clearable

frontend/src/components/settings/sections/EngineMatrixInferenceCollapse.vue

@@ -20,7 +20,6 @@
           class="em-infer-input"
           :value="maxTokens"
           :min="1"
-          :max="200000"
           :step="256"
           size="small"
           @update:value="onMaxTokens"
@@ -43,6 +42,8 @@
 </template>
 
 <script setup lang="ts">
+import { DEFAULT_MAX_OUTPUT_TOKENS } from '@/constants/llm'
+
 const props = defineProps<{
   temperature: number
   maxTokens: number
@@ -60,7 +61,7 @@ function onTemperature(v: number | null) {
 }
 
 function onMaxTokens(v: number | null) {
-  emit('update:maxTokens', Math.max(1, Math.floor(v ?? 4096)))
+  emit('update:maxTokens', Math.max(1, Math.floor(v ?? DEFAULT_MAX_OUTPUT_TOKENS)))
 }
 
 function onTimeout(v: number | null) {

frontend/src/components/settings/sections/EngineMatrixSection.vue

@@ -113,6 +113,7 @@
 import { computed, onMounted, reactive, ref } from 'vue'
 import { useMessage } from 'naive-ui'
 import { llmControlApi, type LLMControlPanelData, type LLMProfile, type LLMProtocol } from '@/api/llmControl'
+import { DEFAULT_MAX_OUTPUT_TOKENS } from '@/constants/llm'
 import EndpointGrid from './EngineMatrixEndpointGrid.vue'
 import InferenceCollapse from './EngineMatrixInferenceCollapse.vue'
 
@@ -161,21 +162,21 @@ const formData = reactive<ModelRoleConfig>({
   default_model_base_url: '',
   default_model: '',
   default_temperature: 0.7,
-  default_max_tokens: 16000,
+  default_max_tokens: DEFAULT_MAX_OUTPUT_TOKENS,
   default_timeout_seconds: 300,
   cheap_model_provider: 'openai',
   cheap_model_api_key: '',
   cheap_model_base_url: '',
   cheap_model: '',
   cheap_temperature: 0.5,
-  cheap_max_tokens: 16000,
+  cheap_max_tokens: DEFAULT_MAX_OUTPUT_TOKENS,
   cheap_timeout_seconds: 300,
   knowledge_model_provider: 'openai',
   knowledge_model_api_key: '',
   knowledge_model_base_url: '',
   knowledge_model: '',
   knowledge_temperature: 0.3,
-  knowledge_max_tokens: 16000,
+  knowledge_max_tokens: DEFAULT_MAX_OUTPUT_TOKENS,
   knowledge_timeout_seconds: 300,
 })
 

frontend/src/components/workbench/LLMControlPanel.vue

@@ -257,6 +257,7 @@ import {
   type LLMRuntimeSummary,
   type ModelItem,
 } from '../../api/llmControl'
+import { DEFAULT_MAX_OUTPUT_TOKENS } from '@/constants/llm'
 import { readStorageJson, writeStorageJson } from '@/utils/storage'
 import { formatApiError } from '../../utils/apiError'
 
@@ -409,7 +410,7 @@ function buildProfileFromPreset(preset?: LLMPreset): LLMProfile {
     api_key: '',
     model: preset?.default_model || '',
     temperature: 0.7,
-    max_tokens: 16000,
+    max_tokens: DEFAULT_MAX_OUTPUT_TOKENS,
     timeout_seconds: 300,
     extra_headers: {},
     extra_query: {},

frontend/src/constants/llm.ts

@@ -0,0 +1 @@
+export const DEFAULT_MAX_OUTPUT_TOKENS = 120000

infrastructure/persistence/database/schema.sql

@@ -868,7 +868,7 @@ CREATE TABLE IF NOT EXISTS llm_profiles (
     api_key TEXT NOT NULL DEFAULT '',
     model TEXT NOT NULL DEFAULT '',
     temperature REAL NOT NULL DEFAULT 0.7,
-    max_tokens INTEGER NOT NULL DEFAULT 4096,
+    max_tokens INTEGER NOT NULL DEFAULT 120000,
     timeout_seconds INTEGER NOT NULL DEFAULT 300,
     extra_headers TEXT NOT NULL DEFAULT '{}',
     extra_query TEXT NOT NULL DEFAULT '{}',

interfaces/api/v1/engine/dag/dag_routes.py

@@ -421,7 +421,7 @@ class UpdateNodeConfigRequest(BaseModel):
     max_retries: Optional[int] = Field(default=None, ge=0, le=5)
     timeout_seconds: Optional[int] = Field(default=None, ge=10, le=600)
     temperature: Optional[float] = Field(default=None, ge=0.0, le=2.0)
-    max_tokens: Optional[int] = Field(default=None, ge=100, le=120000)
+    max_tokens: Optional[int] = Field(default=None, ge=100)
 
 
 @router.put("/{novel_id}/nodes/{node_id}")