add vLLM Server

Author: zhiwei

configs/config_qwen.toml

@@ -30,7 +30,6 @@ max_follow_ups = 3
 summarizer_model_id = "Qwen"
 
 [browser_tool]
-model_id = "Qwen"
 headless = false
 disable_security = true
 extra_chromium_args = []

src/models/models.py

@@ -28,7 +28,7 @@ def init_models(self, use_local_proxy: bool = False):
         self._register_google_models(use_local_proxy=use_local_proxy)
         self._register_qwen_models(use_local_proxy=use_local_proxy)
         self._register_langchain_models(use_local_proxy=use_local_proxy)
-    
+        self._register_vllm_models(use_local_proxy=use_local_proxy)
     def _check_local_api_key(self, local_api_key_name: str, remote_api_key_name: str) -> str:
         api_key = os.getenv(local_api_key_name, PLACEHOLDER)
         if api_key == PLACEHOLDER:
@@ -383,4 +383,56 @@ def _register_langchain_models(self, use_local_proxy: bool = False):
                     api_key=api_key,
                     base_url=api_base,
                 )
-                self.registed_models[model_name] = model
+                self.registed_models[model_name] = model
+    def _register_vllm_models(self, use_local_proxy: bool = False):
+        # qwen
+        api_key = self._check_local_api_key(local_api_key_name="QWEN_API_KEY", 
+                                                remote_api_key_name="QWEN_API_KEY")
+        api_base = self._check_local_api_base(local_api_base_name="QWEN_API_BASE", 
+                                                    remote_api_base_name="QWEN_API_BASE")
+        models = [
+            {
+                "model_name": "Qwen",
+                "model_id": "Qwen",
+            }
+        ]
+        for model in models:
+            model_name = model["model_name"]
+            model_id = model["model_id"]
+            
+            client = OpenAI(
+                api_key=api_key,
+                base_url=api_base,
+            )
+            model = OpenAIServerModel(
+                model_id=model_id,
+                http_client=client,
+                custom_role_conversions=custom_role_conversions,
+            )
+            self.registed_models[model_name] = model
+
+        # Qwen-VL
+        api_key_VL = self._check_local_api_key(local_api_key_name="QWEN_VL_API_KEY", 
+                                                remote_api_key_name="QWEN_VL_API_KEY")
+        api_base_VL = self._check_local_api_base(local_api_base_name="QWEN_VL_API_BASE", 
+                                                    remote_api_base_name="QWEN_VL_API_BASE")
+        models = [
+            {
+                "model_name": "Qwen-VL",
+                "model_id": "Qwen-VL",
+            }
+        ]
+        for model in models:
+            model_name = model["model_name"]
+            model_id = model["model_id"]
+
+            client = OpenAI(
+                api_key=api_key_VL,
+                base_url=api_base_VL,
+            )
+            model = OpenAIServerModel(
+                model_id=model_id,
+                http_client=client,
+                custom_role_conversions=custom_role_conversions,
+            )
+            self.registed_models[model_name] = model

src/tools/deep_researcher.py

@@ -566,7 +566,7 @@ async def _analyze_content(
         return insights
 
     async def _summary(self, query: str, reference_materials: str) -> str:
-        model = model_manager.registed_models["gpt-4o-search-preview"]
+        model = model_manager.registed_models["Qwen"]
 
         messages = [
             {"role": "user", "content": query}