重构初始化逻辑以支持根据环境选择不同的LLM模型

app/service/fashion_agent/init_llm.py

@@ -7,22 +7,24 @@ from langchain_qwq import ChatQwen
 load_dotenv()
 
 QWEN_API_KEY_INTL = os.environ.get("QWEN_API_KEY_INTL", "")
+SERVE_ENV = os.environ.get("SERVE_ENV", "dev")  # prod 使用 Qwen API，dev 使用本地 Ollama
 
 
 def build_llm(enable_thinking: bool = False):
-    # llm = ChatQwen(
+    if SERVE_ENV == "prod":
-    #     model="qwen3.6-plus",
+        llm = ChatQwen(
-    #     timeout=None,
+            model="qwen3.6-plus",
-    #     max_retries=2,
+            timeout=None,
-    #     enable_thinking=enable_thinking,
+            max_retries=2,
-    #     streaming=True,
+            enable_thinking=enable_thinking,
-    #     api_key=QWEN_API_KEY_INTL,
+            streaming=True,
-    # )
+            api_key=QWEN_API_KEY_INTL,
-    # return llm
+        )
+    else:
         llm = ChatOllama(
             model="qwen3.6:27b",
             base_url="http://127.0.0.1:11434",
-        reasoning=enable_thinking,  # 核心：开启思考，自动拆分reasoning_content
+            reasoning=enable_thinking,
             temperature=0.7,
             num_ctx=32768,
         )
@@ -42,6 +44,17 @@ def build_llm(enable_thinking: bool = False):
 # print(response)
 
 
+if SERVE_ENV == "prod":
+    qwen_plus_llm = ChatQwen(
+        model="qwen-plus",
+        timeout=None,
+        max_retries=2,
+        streaming=False,
+        temperature=0.25,
+        top_p=0.8,
+        api_key=QWEN_API_KEY_INTL,
+    )
+else:
     qwen_plus_llm = ChatOllama(
         model="qwen3.6:27b", base_url="http://127.0.0.1:11434", reasoning=False, temperature=0.7, num_ctx=32768, extra_kwargs={"stream": False}
     )