Optimize tablet load: defer health check, lighten service worker, drop Google Fonts.

Co-authored-by: Cursor <cursoragent@cursor.com>

app.py

@@ -197,9 +197,14 @@ PWA_HEAD = """
   var deferredPrompt = null;
 
   if ("serviceWorker" in navigator) {
-    window.addEventListener("load", function () {
+    function registerSW() {
       navigator.serviceWorker.register("/sw.js", { scope: "/" }).catch(function () {});
-    });
+    }
+    if ("requestIdleCallback" in window) {
+      requestIdleCallback(registerSW, { timeout: 5000 });
+    } else {
+      setTimeout(registerSW, 3000);
+    }
   }
 
   function isStandalone() {
@@ -711,24 +716,37 @@ def _status_html(title: str, message: str, level: str = "warn") -> str:
     )
 
 
-def ui_check_ollama_html() -> str:
+def ui_check_ollama_html(force: bool = False) -> str:
-    ok, msg = check_ollama_health()
+    ok, msg = check_ollama_health(force=force)
     return _status_html("Ollama 节点", msg, "ok" if ok else "err")
 
 
+def ui_initial_load() -> tuple[str, str]:
+    """首屏立即返回，不发起网络请求，避免平板白屏等待。"""
+    return (
+        _status_html("Ollama 节点", "后台检测中，请稍候…", "warn"),
+        ui_speaker_status_html(),
+    )
+
+
+def ui_refresh_status_html(force: bool = False) -> tuple[str, str]:
+    """刷新 Ollama + 音色状态（供 Timer / 按钮调用）。"""
+    return ui_check_ollama_html(force=force), ui_speaker_status_html()
+
+
 def ui_speaker_status_html() -> str:
     ok, msg = speaker_is_ready()
     return _status_html("音色状态", msg, "ok" if ok else "warn")
 
 
 def build_theme() -> gr.themes.Base:
-    """高对比度暗色主题（Gradio 6.0 需在 launch() 传入）。"""
+    """高对比度暗色主题；使用系统字体，避免平板拉取 Google Fonts 卡顿。"""
     return gr.themes.Base(
         primary_hue="blue",
         secondary_hue="blue",
         neutral_hue="slate",
-        font=[gr.themes.GoogleFont("Inter"), "system-ui", "sans-serif"],
+        font=["system-ui", "-apple-system", "Segoe UI", "Roboto", "sans-serif"],
-        font_mono=[gr.themes.GoogleFont("JetBrains Mono"), "Consolas", "monospace"],
+        font_mono=["Consolas", "Monaco", "Courier New", "monospace"],
     ).set(
         body_background_fill="#0f1419",
         body_background_fill_dark="#0f1419",
@@ -780,7 +798,25 @@ def build_app() -> gr.Blocks:
             refresh_btn = gr.Button("🔄 刷新状态", variant="secondary", scale=0, min_width=120)
 
         refresh_btn.click(
-            fn=lambda: (ui_check_ollama_html(), ui_speaker_status_html()),
+            fn=lambda: ui_refresh_status_html(force=True),
+            outputs=[ollama_status, speaker_status],
+        )
+
+        # 首屏秒开：仅本地检测音色，Ollama 延后到 Timer
+        demo.load(
+            fn=ui_initial_load,
+            outputs=[ollama_status, speaker_status],
+        )
+
+        # 1 秒后后台检测 Ollama；之后每 30s 刷新（30s 内走缓存）
+        status_timer = gr.Timer(value=1, active=True)
+        status_timer.tick(
+            fn=lambda: ui_refresh_status_html(force=False),
+            outputs=[ollama_status, speaker_status],
+        )
+        status_timer_slow = gr.Timer(value=30, active=True)
+        status_timer_slow.tick(
+            fn=lambda: ui_refresh_status_html(force=True),
             outputs=[ollama_status, speaker_status],
         )
 
@@ -886,11 +922,6 @@ def build_app() -> gr.Blocks:
                     [pipe_raw, pipe_polished, pipe_output, pipeline_log],
                 )
 
-        demo.load(
-            fn=lambda: (ui_check_ollama_html(), ui_speaker_status_html()),
-            outputs=[ollama_status, speaker_status],
-        )
-
     return demo
 
 

config.py

@@ -23,6 +23,11 @@ PORT = 5683
 # HTTP 请求超时（秒）
 OLLAMA_TIMEOUT = 60
 
+# 健康检查快速超时（秒）— 避免平板首屏被长时间阻塞
+HEALTH_CHECK_CONNECT_TIMEOUT = 2
+HEALTH_CHECK_READ_TIMEOUT = 3
+HEALTH_CHECK_CACHE_SECONDS = 30
+
 # ---------------------------------------------------------------------------
 # LLM 系统 Prompt
 # ---------------------------------------------------------------------------

llm_service.py

@@ -6,14 +6,26 @@
 from __future__ import annotations
 
 import logging
+import time
 from typing import Tuple
 
 import requests
 
-from config import MODEL_NAME, OLLAMA_TIMEOUT, OLLAMA_URL, SYSTEM_PROMPT
+from config import (
+    HEALTH_CHECK_CACHE_SECONDS,
+    HEALTH_CHECK_CONNECT_TIMEOUT,
+    HEALTH_CHECK_READ_TIMEOUT,
+    MODEL_NAME,
+    OLLAMA_TIMEOUT,
+    OLLAMA_URL,
+    SYSTEM_PROMPT,
+)
 
 logger = logging.getLogger(__name__)
 
+# 健康检查短时缓存，避免平板/手机反复打开页面时重复等待
+_health_cache: dict = {"ts": 0.0, "ok": False, "msg": ""}
+
 
 def _build_payload(raw_text: str) -> dict:
     """构造 Ollama /api/chat 非流式请求体。"""
@@ -139,24 +151,48 @@ def polish_text(raw_text: str) -> Tuple[bool, str]:
         return False, err
 
 
-def check_ollama_health() -> Tuple[bool, str]:
+def check_ollama_health(force: bool = False) -> Tuple[bool, str]:
     """
     快速检测 Ollama 节点是否在线（不触发完整推理）。
+    默认 2+3 秒超时，结果缓存 30 秒，避免平板首屏长时间白屏。
 
     Returns:
         (online, message)
     """
+    global _health_cache
+
+    now = time.time()
+    if (
+        not force
+        and _health_cache["msg"]
+        and (now - _health_cache["ts"]) < HEALTH_CHECK_CACHE_SECONDS
+    ):
+        return _health_cache["ok"], _health_cache["msg"]
+
     base_url = OLLAMA_URL.rsplit("/api/", 1)[0]
+    timeout = (HEALTH_CHECK_CONNECT_TIMEOUT, HEALTH_CHECK_READ_TIMEOUT)
+
     try:
-        resp = requests.get(f"{base_url}/api/tags", timeout=10)
+        resp = requests.get(f"{base_url}/api/tags", timeout=timeout)
         resp.raise_for_status()
         tags = resp.json().get("models", [])
         model_names = [m.get("name", "") for m in tags]
         if any(MODEL_NAME.split(":")[0] in name for name in model_names):
-            return True, f"Ollama 在线，已检测到模型: {MODEL_NAME}"
+            msg = f"Ollama 在线，已检测到模型: {MODEL_NAME}"
-        return True, (
+            ok = True
-            f"Ollama 在线，但未找到模型 {MODEL_NAME}，"
+        else:
-            f"请执行: ollama pull {MODEL_NAME}"
+            ok = True
+            msg = (
+                f"Ollama 在线，但未找到模型 {MODEL_NAME}，"
+                f"请执行: ollama pull {MODEL_NAME}"
+            )
+    except requests.exceptions.Timeout:
+        ok, msg = False, (
+            f"Ollama 检测超时（>{HEALTH_CHECK_READ_TIMEOUT}s）。"
+            "页面已加载，可稍后点击「刷新状态」重试。"
         )
     except Exception as exc:
-        return False, f"Ollama 不可达: {exc}"
+        ok, msg = False, f"Ollama 不可达: {exc}"
+
+    _health_cache.update({"ts": now, "ok": ok, "msg": msg})
+    return ok, msg

pwa/sw.js

@@ -1,14 +1,10 @@
 /**
- * Trading Studio PWA Service Worker
+ * Trading Studio PWA Service Worker（轻量版）
- * 缓存应用壳，支持离线打开已访问页面；API 请求始终走网络。
+ * 仅处理导航请求，不拦截 Gradio 静态资源 — 避免平板端加载变慢。
  */
-const CACHE_NAME = "trading-studio-v1";
+const CACHE_NAME = "trading-studio-v2";
-const SHELL = ["/", "/manifest.webmanifest"];
 
 self.addEventListener("install", (event) => {
-  event.waitUntil(
-    caches.open(CACHE_NAME).then((cache) => cache.addAll(SHELL)).catch(() => {})
-  );
   self.skipWaiting();
 });
 
@@ -23,28 +19,13 @@ self.addEventListener("activate", (event) => {
 
 self.addEventListener("fetch", (event) => {
   const { request } = event;
-  const url = new URL(request.url);
 
-  // API / 文件上传 / Gradio 动态接口不走缓存
+  // 仅缓存页面导航，Gradio JS/CSS/API 全部直连网络
-  if (
+  if (request.method !== "GET" || request.mode !== "navigate") {
-    request.method !== "GET" ||
-    url.pathname.startsWith("/gradio_api") ||
-    url.pathname.startsWith("/file=") ||
-    url.pathname.startsWith("/upload") ||
-    url.pathname.includes("call")
-  ) {
     return;
   }
 
   event.respondWith(
-    fetch(request)
+    fetch(request).catch(() => caches.match("/"))
-      .then((response) => {
-        if (response.ok && url.origin === self.location.origin) {
-          const clone = response.clone();
-          caches.open(CACHE_NAME).then((cache) => cache.put(request, clone));
-        }
-        return response;
-      })
-      .catch(() => caches.match(request).then((r) => r || caches.match("/")))
   );
 });