fix: make model validation token budget configurable

lib/admin/settings-registry.ts

@@ -88,6 +88,17 @@ export const SETTINGS_REGISTRY: SettingDef[] = [
         label: "Max Output Tokens",
         min: 1,
     },
+    {
+        key: "MODEL_VALIDATION_MAX_TOKENS",
+        group: "generation",
+        type: "number",
+        label: "Model Validation Max Tokens",
+        description:
+            "Token budget for model test requests. Increase for reasoning models that may emit thinking tokens first.",
+        min: 1,
+        max: 64000,
+        default: "1000",
+    },
 
     // ── Access Control ───────────────────────────────────────────────
     {

lib/i18n/dictionaries/en.json

@@ -494,6 +494,10 @@
             "MAX_OUTPUT_TOKENS": {
                 "label": "Max Output Tokens"
             },
+            "MODEL_VALIDATION_MAX_TOKENS": {
+                "label": "Model Validation Max Tokens",
+                "description": "Token budget for model test requests. Increase for reasoning models that may emit thinking tokens first."
+            },
             "ACCESS_CODE_LIST": {
                 "label": "Access Codes",
                 "description": "Comma-separated list. Users must enter one to chat. Empty = open access."

lib/i18n/dictionaries/ja.json

@@ -494,6 +494,10 @@
             "MAX_OUTPUT_TOKENS": {
                 "label": "最大出力トークン数"
             },
+            "MODEL_VALIDATION_MAX_TOKENS": {
+                "label": "モデル検証の最大トークン数",
+                "description": "モデルテストリクエストのトークン予算。推論モデルが先に思考トークンを出す場合は増やしてください。"
+            },
             "ACCESS_CODE_LIST": {
                 "label": "アクセスコード",
                 "description": "カンマ区切りのリスト。チャットにはいずれかの入力が必要です。空 = オープンアクセス。"

lib/i18n/dictionaries/zh-Hant.json

@@ -494,6 +494,10 @@
             "MAX_OUTPUT_TOKENS": {
                 "label": "最大輸出 token 數"
             },
+            "MODEL_VALIDATION_MAX_TOKENS": {
+                "label": "模型驗證最大 token 數",
+                "description": "模型測試請求的 token 預算。推理模型可能先輸出思考 token，可適當調大。"
+            },
             "ACCESS_CODE_LIST": {
                 "label": "存取碼",
                 "description": "以逗號分隔的清單。使用者需輸入其中之一才能聊天。留空 = 開放存取。"

lib/i18n/dictionaries/zh.json

@@ -494,6 +494,10 @@
             "MAX_OUTPUT_TOKENS": {
                 "label": "最大输出 token 数"
             },
+            "MODEL_VALIDATION_MAX_TOKENS": {
+                "label": "模型验证最大 token 数",
+                "description": "模型测试请求的 token 预算。推理模型可能先输出思考 token，可适当调大。"
+            },
             "ACCESS_CODE_LIST": {
                 "label": "访问码",
                 "description": "以逗号分隔的列表。用户需输入其中之一才能聊天。留空 = 开放访问。"

lib/model-validation.ts

@@ -0,0 +1,28 @@
+export const DEFAULT_MODEL_VALIDATION_MAX_TOKENS = 1000
+export const MAX_MODEL_VALIDATION_MAX_TOKENS = 64000
+
+export function parseModelValidationMaxTokens(value: unknown): number | null {
+    if (value === undefined || value === null || value === "") return null
+
+    const numeric =
+        typeof value === "number" ? value : Number(String(value).trim())
+
+    if (
+        !Number.isFinite(numeric) ||
+        !Number.isInteger(numeric) ||
+        numeric < 1
+    ) {
+        return null
+    }
+
+    return Math.min(numeric, MAX_MODEL_VALIDATION_MAX_TOKENS)
+}
+
+export function resolveModelValidationMaxTokens(...sources: unknown[]): number {
+    for (const source of sources) {
+        const parsed = parseModelValidationMaxTokens(source)
+        if (parsed !== null) return parsed
+    }
+
+    return DEFAULT_MODEL_VALIDATION_MAX_TOKENS
+}