deploy: 2026-07-15 14:48 部署更新

demo/web/server.py

@@ -18,7 +18,7 @@ import psycopg2
 import psycopg2.pool
 from fastapi import FastAPI, HTTPException, Query
 from fastapi.staticfiles import StaticFiles
-from fastapi.responses import RedirectResponse
+from fastapi.responses import RedirectResponse, StreamingResponse
 from pydantic import BaseModel
 
 # ---------------------------------------------------------------------------
@@ -932,6 +932,8 @@ def generate_paradigm(
         ],
         "response_format": {"type": "json_object"},
         "temperature": 0.4,
+        "stream": True,
+        "stream_options": {"include_usage": False},
     }).encode("utf-8")
 
     req = _urllib.Request(
@@ -944,14 +946,40 @@ def generate_paradigm(
         method="POST",
     )
 
-    try:
-        with _urllib.urlopen(req, timeout=60) as resp:
-            result = json.loads(resp.read().decode("utf-8"))
-        content = result["choices"][0]["message"]["content"]
-        parsed = json.loads(content)
-        return parsed
-    except Exception as exc:
-        raise HTTPException(status_code=502, detail=f"LLM 调用失败: {exc}")
+    def stream_generator():
+        """SSE 流式生成器，逐块推送 LLM content delta。"""
+        try:
+            resp = _urllib.urlopen(req, timeout=120)
+            buffer = b""
+            while True:
+                chunk = resp.read(4096)
+                if not chunk:
+                    break
+                buffer += chunk
+                while b"\n" in buffer:
+                    line, buffer = buffer.split(b"\n", 1)
+                    line = line.strip()
+                    if not line:
+                        continue
+                    if line.startswith(b"data:"):
+                        data_str = line[5:].strip()
+                        if data_str == b"[DONE]":
+                            yield "data: [DONE]\n\n"
+                            return
+                        try:
+                            chunk_obj = json.loads(data_str)
+                            delta = chunk_obj.get("choices", [{}])[0].get("delta", {})
+                            content = delta.get("content", "")
+                            if content:
+                                yield f"data: {json.dumps({'content': content}, ensure_ascii=False)}\n\n"
+                        except (json.JSONDecodeError, IndexError, KeyError):
+                            continue
+            resp.close()
+        except Exception as exc:
+            yield f"data: {json.dumps({'error': str(exc)}, ensure_ascii=False)}\n\n"
+            yield "data: [DONE]\n\n"
+
+    return StreamingResponse(stream_generator(), media_type="text/event-stream")
 
 
 @app.get("/api/sync/status")

demo/web/static/index.html

@@ -1132,13 +1132,63 @@ async function generateParadigm() {
   const result = document.getElementById('paradigm-result');
   btn.disabled = true;
   btn.textContent = '生成中...';
-  result.innerHTML = '<p class="text-sm text-gray-400">正在调用千问 LLM 生成标准范式对话，请稍候...</p>';
+  result.innerHTML = `
+    <div class="bg-white rounded-lg shadow p-6">
+      <div class="flex items-center gap-2 mb-4">
+        <svg class="animate-spin w-5 h-5 text-green-600" fill="none" viewBox="0 0 24 24"><circle class="opacity-25" cx="12" cy="12" r="10" stroke="currentColor" stroke-width="4"/><path class="opacity-75" fill="currentColor" d="M4 12a8 8 0 018-8V0C5.373 0 0 5.373 0 12h4z"/></svg>
+        <span class="text-sm text-gray-500">正在调用千问 LLM 生成标准范式对话...</span>
+      </div>
+      <pre id="paradigm-stream-text" class="text-xs text-gray-600 bg-gray-50 rounded p-3 max-h-96 overflow-auto whitespace-pre-wrap break-all font-mono"></pre>
+    </div>
+  `;
+  const streamText = document.getElementById('paradigm-stream-text');
+  let fullContent = '';
   try {
     const productId = document.getElementById('paradigm-product-select').value;
-    let paradigmPath = `/conversations/${customerId}/paradigm`;
-    if (productId) paradigmPath += `?product_id=${productId}`;
-    const data = await api(paradigmPath, { method: 'POST' });
-    renderParadigm(data);
+    let paradigmUrl = `${API}/conversations/${customerId}/paradigm`;
+    if (productId) paradigmUrl += `?product_id=${productId}`;
+    const resp = await fetch(paradigmUrl, { method: 'POST' });
+    if (!resp.ok) throw new Error(`HTTP ${resp.status}`);
+    const reader = resp.body.getReader();
+    const decoder = new TextDecoder();
+    let buffer = '';
+    while (true) {
+      const { done, value } = await reader.read();
+      if (done) break;
+      buffer += decoder.decode(value, { stream: true });
+      const lines = buffer.split('\n');
+      buffer = lines.pop();
+      for (const line of lines) {
+        if (!line.startsWith('data: ')) continue;
+        const dataStr = line.slice(6).trim();
+        if (dataStr === '[DONE]') continue;
+        try {
+          const obj = JSON.parse(dataStr);
+          if (obj.error) throw new Error(obj.error);
+          if (obj.content) {
+            fullContent += obj.content;
+            streamText.textContent = fullContent;
+            streamText.scrollTop = streamText.scrollHeight;
+          }
+        } catch (e) {
+          // 忽略解析错误的行
+        }
+      }
+    }
+    // 尝试解析完整 JSON 并渲染结构化结果
+    try {
+      const data = JSON.parse(fullContent);
+      renderParadigm(data);
+    } catch (e) {
+      // JSON 解析失败，保留原始流式文本
+      result.innerHTML = `
+        <div class="bg-white rounded-lg shadow p-6">
+          <h3 class="text-lg font-bold mb-2">标准范式对话模板</h3>
+          <p class="text-sm text-orange-500 mb-3">LLM 返回内容无法解析为 JSON，以下为原始输出：</p>
+          <pre class="text-xs text-gray-600 bg-gray-50 rounded p-3 max-h-96 overflow-auto whitespace-pre-wrap break-all">${escapeHtml(fullContent)}</pre>
+        </div>
+      `;
+    }
   } catch (e) {
     result.innerHTML = `<p class="text-sm text-red-500">生成失败: ${e.message}</p>`;
   } finally {